1
1
openmpi/orte/orted/pmix/pmix_server.c

730 строки
26 KiB
C
Исходник Обычный вид История

Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/*
* Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2011 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2006-2013 Los Alamos National Security, LLC.
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
* All rights reserved.
* Copyright (c) 2009-2012 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014 Mellanox Technologies, Inc.
* All rights reserved.
* Copyright (c) 2014-2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#include "orte_config.h"
#include "orte/types.h"
#include "opal/types.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#include <fcntl.h>
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif
#ifdef HAVE_ARPA_INET_H
#include <arpa/inet.h>
#endif
#ifdef HAVE_NETDB_H
#include <netdb.h>
#endif
#include <ctype.h>
#include "opal_stdint.h"
#include "opal/class/opal_hotel.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
#include "opal/class/opal_list.h"
#include "opal/mca/base/mca_base_var.h"
#include "opal/mca/pmix/pmix.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
#include "opal/util/opal_environ.h"
#include "opal/util/show_help.h"
#include "opal/util/error.h"
#include "opal/util/output.h"
#include "opal/util/os_path.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
#include "opal/util/argv.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/grpcomm/grpcomm.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/base/rml_contact.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
#include "orte/util/name_fns.h"
#include "orte/util/proc_info.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
#include "orte/util/session_dir.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_globals.h"
#include "pmix_server.h"
#include "pmix_server_internal.h"
/*
* Local utility functions
*/
static void pmix_server_dmdx_recv(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tg, void *cbdata);
static void pmix_server_dmdx_resp(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tg, void *cbdata);
pmix_server_globals_t orte_pmix_server_globals = {0};
static opal_pmix_server_module_t pmix_server = {
.client_connected = pmix_server_client_connected_fn,
.client_finalized = pmix_server_client_finalized_fn,
.abort = pmix_server_abort_fn,
.fence_nb = pmix_server_fencenb_fn,
.direct_modex = pmix_server_dmodex_req_fn,
.publish = pmix_server_publish_fn,
.lookup = pmix_server_lookup_fn,
.unpublish = pmix_server_unpublish_fn,
.spawn = pmix_server_spawn_fn,
.connect = pmix_server_connect_fn,
.disconnect = pmix_server_disconnect_fn,
.register_events = pmix_server_register_events_fn,
.deregister_events = pmix_server_deregister_events_fn,
.notify_event = pmix_server_notify_event,
.query = pmix_server_query_fn,
.tool_connected = pmix_tool_connected_fn,
.log = pmix_server_log_fn
};
void pmix_server_register_params(void)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
/* register a verbosity */
orte_pmix_server_globals.verbosity = -1;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
(void) mca_base_var_register ("orte", "pmix", NULL, "server_verbose",
"Debug verbosity for PMIx server",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orte_pmix_server_globals.verbosity);
if (0 <= orte_pmix_server_globals.verbosity) {
orte_pmix_server_globals.output = opal_output_open(NULL);
opal_output_set_verbosity(orte_pmix_server_globals.output,
orte_pmix_server_globals.verbosity);
}
/* specify the size of the hotel */
orte_pmix_server_globals.num_rooms = 256;
(void) mca_base_var_register ("orte", "pmix", NULL, "server_max_reqs",
"Maximum number of backlogged PMIx server direct modex requests",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orte_pmix_server_globals.num_rooms);
/* specify the timeout for the hotel */
orte_pmix_server_globals.timeout = 2;
(void) mca_base_var_register ("orte", "pmix", NULL, "server_max_wait",
"Maximum time (in seconds) the PMIx server should wait to service direct modex requests",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orte_pmix_server_globals.timeout);
/* register the URI of the UNIVERSAL data server */
orte_pmix_server_globals.server_uri = NULL;
(void) mca_base_var_register ("orte", "pmix", NULL, "server_uri",
"URI of a session-level keyval server for publish/lookup operations",
MCA_BASE_VAR_TYPE_STRING, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orte_pmix_server_globals.server_uri);
/* whether or not to wait for the universal server */
orte_pmix_server_globals.wait_for_server = false;
(void) mca_base_var_register ("orte", "pmix", NULL, "wait_for_server",
"Whether or not to wait for the session-level server to start",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_ALL,
&orte_pmix_server_globals.wait_for_server);
}
static void eviction_cbfunc(struct opal_hotel_t *hotel,
int room_num, void *occupant)
{
pmix_server_req_t *req = (pmix_server_req_t*)occupant;
bool timeout = false;
int rc;
/* decrement the request timeout */
req->timeout -= orte_pmix_server_globals.timeout;
if (req->timeout > 0) {
req->timeout -= orte_pmix_server_globals.timeout;
if (0 >= req->timeout) {
timeout = true;
}
}
if (!timeout) {
/* not done yet - check us back in */
if (OPAL_SUCCESS == (rc = opal_hotel_checkin(&orte_pmix_server_globals.reqs, req, &req->room_num))) {
return;
}
ORTE_ERROR_LOG(rc);
/* fall thru and return an error so the caller doesn't hang */
}
/* don't let the caller hang */
if (NULL != req->opcbfunc) {
req->opcbfunc(OPAL_ERR_TIMEOUT, req->cbdata);
} else if (NULL != req->mdxcbfunc) {
req->mdxcbfunc(OPAL_ERR_TIMEOUT, NULL, 0, req->cbdata, NULL, NULL);
} else if (NULL != req->spcbfunc) {
req->spcbfunc(OPAL_ERR_TIMEOUT, ORTE_JOBID_INVALID, req->cbdata);
} else if (NULL != req->lkcbfunc) {
req->lkcbfunc(OPAL_ERR_TIMEOUT, NULL, req->cbdata);
}
OBJ_RELEASE(req);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/*
* Initialize global variables used w/in the server.
*/
int pmix_server_init(void)
{
int rc;
opal_list_t info;
opal_value_t *kv;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
if (orte_pmix_server_globals.initialized) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return ORTE_SUCCESS;
}
orte_pmix_server_globals.initialized = true;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* setup the server's state variables */
OBJ_CONSTRUCT(&orte_pmix_server_globals.reqs, opal_hotel_t);
if (OPAL_SUCCESS != (rc = opal_hotel_init(&orte_pmix_server_globals.reqs,
orte_pmix_server_globals.num_rooms,
orte_event_base, orte_pmix_server_globals.timeout*1000000,
ORTE_ERROR_PRI, eviction_cbfunc))) {
ORTE_ERROR_LOG(rc);
return rc;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
OBJ_CONSTRUCT(&orte_pmix_server_globals.notifications, opal_list_t);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* setup recv for direct modex requests */
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DIRECT_MODEX,
ORTE_RML_PERSISTENT, pmix_server_dmdx_recv, NULL);
/* setup recv for replies to direct modex requests */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DIRECT_MODEX_RESP,
ORTE_RML_PERSISTENT, pmix_server_dmdx_resp, NULL);
/* setup recv for replies to proxy launch requests */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_LAUNCH_RESP,
ORTE_RML_PERSISTENT, pmix_server_launch_resp, NULL);
/* setup recv for replies from data server */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DATA_CLIENT,
ORTE_RML_PERSISTENT, pmix_server_keyval_client, NULL);
/* setup recv for notifications */
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_NOTIFICATION,
ORTE_RML_PERSISTENT, pmix_server_notify, NULL);
/* ensure the PMIx server uses the proper rendezvous directory */
opal_setenv("PMIX_SERVER_TMPDIR", orte_process_info.proc_session_dir, true, &environ);
/* pass the server the local topology - we do this so the procs won't read the
* topology themselves as this could overwhelm the local
* system on large-scale SMPs */
OBJ_CONSTRUCT(&info, opal_list_t);
if (NULL != opal_hwloc_topology) {
char *xmlbuffer=NULL;
int len;
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_LOCAL_TOPO);
if (0 != hwloc_topology_export_xmlbuffer(opal_hwloc_topology, &xmlbuffer, &len)) {
OBJ_RELEASE(kv);
OBJ_DESTRUCT(&info);
return ORTE_ERROR;
}
kv->data.string = xmlbuffer;
kv->type = OPAL_STRING;
opal_list_append(&info, &kv->super);
}
/* tell the server to allow tool connections */
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_SERVER_TOOL_SUPPORT);
kv->type = OPAL_BOOL;
kv->data.flag = true;
opal_list_append(&info, &kv->super);
/* tell the server our temp directory */
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_SERVER_TMPDIR);
kv->type = OPAL_STRING;
kv->data.string = opal_os_path(false, orte_process_info.jobfam_session_dir, NULL);
opal_list_append(&info, &kv->super);
/* use the same for the system temp directory - this is
* where the system-level tool connections will go */
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_SYSTEM_TMPDIR);
kv->type = OPAL_STRING;
kv->data.string = strdup(orte_process_info.tmpdir_base);
opal_list_append(&info, &kv->super);
/* setup the local server */
if (ORTE_SUCCESS != (rc = opal_pmix.server_init(&pmix_server, &info))) {
/* pmix will provide a nice show_help output here */
return rc;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
OPAL_LIST_DESTRUCT(&info);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* if the universal server wasn't specified, then we use
* our own HNP for that purpose */
if (NULL == orte_pmix_server_globals.server_uri) {
orte_pmix_server_globals.server = *ORTE_PROC_MY_HNP;
} else {
char *server;
opal_buffer_t buf;
if (0 == strncmp(orte_pmix_server_globals.server_uri, "file", strlen("file")) ||
0 == strncmp(orte_pmix_server_globals.server_uri, "FILE", strlen("FILE"))) {
char input[1024], *filename;
FILE *fp;
/* it is a file - get the filename */
filename = strchr(orte_pmix_server_globals.server_uri, ':');
if (NULL == filename) {
/* filename is not correctly formatted */
orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-bad", true,
orte_basename, orte_pmix_server_globals.server_uri);
return ORTE_ERR_BAD_PARAM;
}
++filename; /* space past the : */
if (0 >= strlen(filename)) {
/* they forgot to give us the name! */
orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-missing", true,
orte_basename, orte_pmix_server_globals.server_uri);
return ORTE_ERR_BAD_PARAM;
}
/* open the file and extract the uri */
fp = fopen(filename, "r");
if (NULL == fp) { /* can't find or read file! */
orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-access", true,
orte_basename, orte_pmix_server_globals.server_uri);
return ORTE_ERR_BAD_PARAM;
}
if (NULL == fgets(input, 1024, fp)) {
/* something malformed about file */
fclose(fp);
orte_show_help("help-orterun.txt", "orterun:ompi-server-file-bad", true,
orte_basename, orte_pmix_server_globals.server_uri,
orte_basename);
return ORTE_ERR_BAD_PARAM;
}
fclose(fp);
input[strlen(input)-1] = '\0'; /* remove newline */
server = strdup(input);
} else {
server = strdup(orte_pmix_server_globals.server_uri);
}
/* setup our route to the server */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
opal_dss.pack(&buf, &server, 1, OPAL_STRING);
if (ORTE_SUCCESS != (rc = orte_rml_base_update_contact_info(&buf))) {
ORTE_ERROR_LOG(rc);
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return rc;
}
OBJ_DESTRUCT(&buf);
/* parse the URI to get the server's name */
if (ORTE_SUCCESS != (rc = orte_rml_base_parse_uris(server, &orte_pmix_server_globals.server, NULL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* check if we are to wait for the server to start - resolves
* a race condition that can occur when the server is run
* as a background job - e.g., in scripts
*/
if (orte_pmix_server_globals.wait_for_server) {
/* ping the server */
struct timeval timeout;
timeout.tv_sec = orte_pmix_server_globals.timeout;
timeout.tv_usec = 0;
if (ORTE_SUCCESS != (rc = orte_rml.ping(orte_mgmt_conduit, server, &timeout))) {
/* try it one more time */
if (ORTE_SUCCESS != (rc = orte_rml.ping(orte_mgmt_conduit, server, &timeout))) {
/* okay give up */
orte_show_help("help-orterun.txt", "orterun:server-not-found", true,
orte_basename, server,
(long)orte_pmix_server_globals.timeout,
ORTE_ERROR_NAME(rc));
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return rc;
}
}
}
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return rc;
}
void pmix_server_finalize(void)
{
if (!orte_pmix_server_globals.initialized) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return;
}
opal_output_verbose(2, orte_pmix_server_globals.output,
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
"%s Finalizing PMIX server",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* stop receives */
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DIRECT_MODEX);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DIRECT_MODEX_RESP);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_LAUNCH_RESP);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DATA_CLIENT);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_NOTIFICATION);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* shutdown the local server */
opal_pmix.server_finalize();
/* cleanup collectives */
OBJ_DESTRUCT(&orte_pmix_server_globals.reqs);
OPAL_LIST_DESTRUCT(&orte_pmix_server_globals.notifications);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
static void send_error(int status, opal_process_name_t *idreq,
orte_process_name_t *remote)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
opal_buffer_t *reply;
Fat SMPs (i.e., systems with nodes containing large numbers of cpus) were failing to start due to connection failures of the opal/pmix support. Root cause was that (a) we were setting the client socket to non-blocking before calling connect, and (b) the server was using the event library to harvest the accepts, and also did the handshake while in that event. So the server would backup beyond the connection backlog limit, and we would fail. Changing the client to leave its socket as blocking during the connect doesn't solve the problem by itself - you also have to introduce a sleep delay once the backlog is hit to avoid simply machine-gunning your way thru retries. This gets somewhat difficult to adjust as you don't want to unnecessarily prolong startup time. We've solved this before by adding a listening thread that simply reaps accepts and shoves them into the event library for subsequent processing. This would resolve the problem, but meant yet another daemon-level thread. So I centralized the listening thread support and let multiple elements register listeners on it. Thus, each daemon now has a single listening thread that reaps accepts from multiple sources - for now, the orte/pmix server and the oob/usock support are using it. I'll add in the oob/tcp component later. This still didn't fully resolve the SMP problem, especially on coprocessor cards (e.g., KNC). Removing the shared memory dstore support helped further improve the behavior - it looks like there is some kind of memory paging issue there that needs further understanding. Given that the shared memory support was about to be lost when I bring over the PMIx integration (until it is restored in that library), it seemed like a reasonable thing to just remove it at this point.
2015-05-30 00:28:26 +03:00
int rc;
reply = OBJ_NEW(opal_buffer_t);
/* pack the status */
if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, &status, 1, OPAL_INT))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(reply);
2014-12-14 07:30:21 +03:00
return;
}
/* pack the id of the requested proc */
if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, idreq, 1, OPAL_NAME))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(reply);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return;
}
/* send the response */
orte_rml.send_buffer_nb(orte_mgmt_conduit,
remote, reply,
ORTE_RML_TAG_DIRECT_MODEX_RESP,
orte_rml_send_callback, NULL);
return;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
static void _mdxresp(int sd, short args, void *cbdata)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
pmix_server_req_t *req = (pmix_server_req_t*)cbdata;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
int rc;
opal_buffer_t *reply;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* check us out of the hotel */
opal_hotel_checkout(&orte_pmix_server_globals.reqs, req->room_num);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
reply = OBJ_NEW(opal_buffer_t);
/* return the status */
if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, &req->status, 1, OPAL_INT))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(reply);
goto done;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/* pack the id of the requested proc */
if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, &req->target, 1, OPAL_NAME))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(reply);
goto done;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/* pack the remote daemon's request room number */
if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, &req->remote_room_num, 1, OPAL_INT))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(reply);
goto done;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/* return any provided data */
opal_dss.copy_payload(reply, &req->msg);
/* send the response */
orte_rml.send_buffer_nb(orte_mgmt_conduit,
&req->proxy, reply,
ORTE_RML_TAG_DIRECT_MODEX_RESP,
orte_rml_send_callback, NULL);
done:
/* if they asked for a release, give it to them */
if (NULL != req->rlcbfunc) {
req->rlcbfunc(req->cbdata);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
OBJ_RELEASE(req);
return;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/* the modex_resp function takes place in the local PMIx server's
* progress thread - we must therefore thread-shift it so we can
* access our global data */
static void modex_resp(int status,
const char *data, size_t sz, void *cbdata,
opal_pmix_release_cbfunc_t relcbfunc, void *relcbdata)
{
pmix_server_req_t *req = (pmix_server_req_t*)cbdata;
opal_buffer_t xfer;
req->status = status;
/* we need to preserve the data as the caller
* will free it upon our return */
OBJ_CONSTRUCT(&xfer, opal_buffer_t);
opal_dss.load(&xfer, (void*)data, sz);
opal_dss.copy_payload(&req->msg, &xfer);
xfer.base_ptr = NULL; // protect the incoming data
OBJ_DESTRUCT(&xfer);
/* point to the callback */
req->rlcbfunc = relcbfunc;
req->cbdata = relcbdata;
opal_event_set(orte_event_base, &(req->ev),
-1, OPAL_EV_WRITE, _mdxresp, req);
opal_event_set_priority(&(req->ev), ORTE_MSG_PRI);
opal_event_active(&(req->ev), OPAL_EV_WRITE, 1);
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
static void pmix_server_dmdx_recv(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tg, void *cbdata)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
int rc, room_num;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
int32_t cnt;
opal_process_name_t idreq;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
orte_process_name_t name;
orte_job_t *jdata;
orte_proc_t *proc;
pmix_server_req_t *req;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* unpack the id of the proc whose data is being requested */
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &idreq, &cnt, OPAL_NAME))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
return;
}
opal_output_verbose(2, orte_pmix_server_globals.output,
"%s dmdx:recv request from proc %s for proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(sender),
ORTE_NAME_PRINT(&idreq));
/* and the remote daemon's tracking room number */
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &room_num, &cnt, OPAL_INT))) {
ORTE_ERROR_LOG(rc);
return;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
/* is this proc one of mine? */
memcpy((char*)&name, (char*)&idreq, sizeof(orte_process_name_t));
if (NULL == (jdata = orte_get_job_data_object(name.jobid))) {
/* not having the jdata means that we haven't unpacked the
* the launch message for this job yet - this is a race
* condition, so just log the request and we will fill
* it later */
req = OBJ_NEW(pmix_server_req_t);
req->proxy = *sender;
req->target = idreq;
req->remote_room_num = room_num;
if (OPAL_SUCCESS != (rc = opal_hotel_checkin(&orte_pmix_server_globals.reqs, req, &req->room_num))) {
OBJ_RELEASE(req);
send_error(rc, &idreq, sender);
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return;
}
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, name.vpid))) {
/* this is truly an error, so notify the sender */
send_error(ORTE_ERR_NOT_FOUND, &idreq, sender);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
return;
}
if (!ORTE_FLAG_TEST(proc, ORTE_PROC_FLAG_LOCAL)) {
/* send back an error - they obviously have made a mistake */
send_error(ORTE_ERR_NOT_FOUND, &idreq, sender);
return;
}
/* track the request since the call down to the PMIx server
* is asynchronous */
req = OBJ_NEW(pmix_server_req_t);
req->proxy = *sender;
req->target = idreq;
req->remote_room_num = room_num;
if (OPAL_SUCCESS != (rc = opal_hotel_checkin(&orte_pmix_server_globals.reqs, req, &req->room_num))) {
OBJ_RELEASE(req);
send_error(rc, &idreq, sender);
return;
}
/* ask our local pmix server for the data */
if (OPAL_SUCCESS != (rc = opal_pmix.server_dmodex_request(&idreq, modex_resp, req))) {
ORTE_ERROR_LOG(rc);
opal_hotel_checkout(&orte_pmix_server_globals.reqs, req->room_num);
OBJ_RELEASE(req);
send_error(rc, &idreq, sender);
return;
}
return;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
typedef struct {
opal_object_t super;
char *data;
int32_t ndata;
} datacaddy_t;
static void dccon(datacaddy_t *p)
{
p->data = NULL;
}
static void dcdes(datacaddy_t *p)
{
if (NULL != p->data) {
free(p->data);
}
}
static OBJ_CLASS_INSTANCE(datacaddy_t,
opal_object_t,
dccon, dcdes);
static void relcbfunc(void *relcbdata)
{
datacaddy_t *d = (datacaddy_t*)relcbdata;
OBJ_RELEASE(d);
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
static void pmix_server_dmdx_resp(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tg, void *cbdata)
{
int rc, ret, room_num, rnum;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
int32_t cnt;
opal_process_name_t target;
pmix_server_req_t *req;
datacaddy_t *d;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
opal_output_verbose(2, orte_pmix_server_globals.output,
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
"%s dmdx:recv response from proc %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(sender));
/* unpack the status */
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &ret, &cnt, OPAL_INT))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
return;
}
/* unpack the id of the target whose info we just received */
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &target, &cnt, OPAL_NAME))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
ORTE_ERROR_LOG(rc);
return;
}
/* unpack our tracking room number */
cnt = 1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &room_num, &cnt, OPAL_INT))) {
ORTE_ERROR_LOG(rc);
return;
}
/* unload the remainder of the buffer */
d = OBJ_NEW(datacaddy_t);
if (OPAL_SUCCESS != (rc = opal_dss.unload(buffer, (void**)&d->data, &d->ndata))) {
ORTE_ERROR_LOG(rc);
return;
}
/* check the request out of the tracking hotel */
opal_hotel_checkout_and_return_occupant(&orte_pmix_server_globals.reqs, room_num, (void**)&req);
/* return the returned data to the requestor */
2015-08-30 07:19:27 +03:00
if (NULL != req) {
if (NULL != req->mdxcbfunc) {
OBJ_RETAIN(d);
req->mdxcbfunc(ret, d->data, d->ndata, req->cbdata, relcbfunc, d);
2015-08-30 07:19:27 +03:00
}
OBJ_RELEASE(req);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
/* now see if anyone else was waiting for data from this target */
for (rnum=0; rnum < orte_pmix_server_globals.reqs.num_rooms; rnum++) {
opal_hotel_knock(&orte_pmix_server_globals.reqs, rnum, (void**)&req);
if (NULL == req) {
continue;
}
if (req->target.jobid == target.jobid &&
req->target.vpid == target.vpid) {
if (NULL != req->mdxcbfunc) {
OBJ_RETAIN(d);
req->mdxcbfunc(ret, d->data, d->ndata, req->cbdata, relcbfunc, d);
}
opal_hotel_checkout(&orte_pmix_server_globals.reqs, rnum);
OBJ_RELEASE(req);
}
}
OBJ_RELEASE(d); // maintain accounting
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
static void opcon(orte_pmix_server_op_caddy_t *p)
{
p->procs = NULL;
p->eprocs = NULL;
p->info = NULL;
p->cbfunc = NULL;
p->infocbfunc = NULL;
p->toolcbfunc = NULL;
p->cbdata = NULL;
}
OBJ_CLASS_INSTANCE(orte_pmix_server_op_caddy_t,
opal_object_t,
opcon, NULL);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
static void rqcon(pmix_server_req_t *p)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
p->target = *ORTE_NAME_INVALID;
p->proxy = *ORTE_NAME_INVALID;
p->timeout = orte_pmix_server_globals.timeout;
p->jdata = NULL;
OBJ_CONSTRUCT(&p->msg, opal_buffer_t);
p->opcbfunc = NULL;
p->mdxcbfunc = NULL;
p->spcbfunc = NULL;
p->lkcbfunc = NULL;
p->rlcbfunc = NULL;
p->cbdata = NULL;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
static void rqdes(pmix_server_req_t *p)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
if (NULL != p->jdata) {
OBJ_RELEASE(p->jdata);
}
OBJ_DESTRUCT(&p->msg);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
OBJ_CLASS_INSTANCE(pmix_server_req_t,
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
opal_object_t,
rqcon, rqdes);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
static void mdcon(orte_pmix_mdx_caddy_t *p)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
p->sig = NULL;
p->cbfunc = NULL;
p->cbdata = NULL;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
static void mddes(orte_pmix_mdx_caddy_t *p)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
{
if (NULL != p->sig) {
OBJ_RELEASE(p->sig);
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 22:56:47 +04:00
}
OBJ_CLASS_INSTANCE(orte_pmix_mdx_caddy_t,
opal_object_t,
mdcon, mddes);