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) 2004-2007 The Trustees of Indiana University and Indiana
|
|
|
|
|
* University Research and Technology
|
|
|
|
|
* Corporation. All rights reserved.
|
|
|
|
|
* Copyright (c) 2004-2006 The University of Tennessee and The University
|
|
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
|
* reserved.
|
2015-06-24 06:59:57 +03:00
|
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
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
|
|
|
* University of Stuttgart. All rights reserved.
|
|
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
|
* All rights reserved.
|
2015-06-24 06:59:57 +03:00
|
|
|
* 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) 2010-2011 Cisco Systems, Inc. All rights reserved.
|
2016-10-08 19:24:28 +03:00
|
|
|
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
|
2014-11-06 15:12:25 +03:00
|
|
|
* Copyright (c) 2014 Mellanox Technologies, Inc.
|
|
|
|
|
* All rights reserved.
|
2014-11-12 04:00:42 +03:00
|
|
|
* Copyright (c) 2014 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
|
|
|
* $COPYRIGHT$
|
2015-06-24 06:59:57 +03:00
|
|
|
*
|
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
|
|
|
* Additional copyrights may follow
|
2015-06-24 06:59:57 +03:00
|
|
|
*
|
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
|
|
|
* $HEADER$
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#ifndef _PMIX_SERVER_INTERNAL_H_
|
|
|
|
|
#define _PMIX_SERVER_INTERNAL_H_
|
|
|
|
|
|
|
|
|
|
#include "orte_config.h"
|
|
|
|
|
#include "orte/types.h"
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_SYS_SOCKET_H
|
|
|
|
|
#include <sys/socket.h>
|
|
|
|
|
#endif
|
|
|
|
|
#ifdef HAVE_SYS_UN_H
|
|
|
|
|
#include <sys/un.h>
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
#include "opal/types.h"
|
2015-06-18 19:53:20 +03:00
|
|
|
#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/mca/base/base.h"
|
|
|
|
|
#include "opal/mca/event/event.h"
|
|
|
|
|
#include "opal/mca/pmix/pmix.h"
|
|
|
|
|
#include "opal/util/proc.h"
|
|
|
|
|
|
2015-06-18 19:53:20 +03:00
|
|
|
#include "orte/mca/grpcomm/base/base.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
|
|
|
|
2015-06-18 19:53:20 +03:00
|
|
|
BEGIN_C_DECLS
|
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
|
|
|
|
2015-06-18 19:53:20 +03:00
|
|
|
/* object for tracking requests so we can
|
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
|
|
|
* correctly route the eventual reply */
|
2015-06-18 19:53:20 +03:00
|
|
|
typedef struct {
|
|
|
|
|
opal_object_t super;
|
|
|
|
|
opal_event_t ev;
|
2015-11-22 06:10:12 +03:00
|
|
|
int status;
|
2015-09-04 18:29:09 +03:00
|
|
|
int timeout;
|
2015-06-18 19:53:20 +03:00
|
|
|
int room_num;
|
|
|
|
|
int remote_room_num;
|
|
|
|
|
orte_process_name_t proxy;
|
2014-11-12 04:00:42 +03:00
|
|
|
opal_process_name_t target;
|
2015-06-18 19:53:20 +03:00
|
|
|
orte_job_t *jdata;
|
|
|
|
|
opal_buffer_t msg;
|
|
|
|
|
opal_pmix_op_cbfunc_t opcbfunc;
|
|
|
|
|
opal_pmix_modex_cbfunc_t mdxcbfunc;
|
|
|
|
|
opal_pmix_spawn_cbfunc_t spcbfunc;
|
|
|
|
|
opal_pmix_lookup_cbfunc_t lkcbfunc;
|
2015-11-22 06:10:12 +03:00
|
|
|
opal_pmix_release_cbfunc_t rlcbfunc;
|
2015-06-18 19:53:20 +03:00
|
|
|
void *cbdata;
|
|
|
|
|
} pmix_server_req_t;
|
|
|
|
|
OBJ_CLASS_DECLARATION(pmix_server_req_t);
|
|
|
|
|
|
|
|
|
|
/* object for thread-shifting server operations */
|
|
|
|
|
typedef struct {
|
|
|
|
|
opal_object_t super;
|
|
|
|
|
opal_event_t ev;
|
2016-06-06 07:39:44 +03:00
|
|
|
int status;
|
|
|
|
|
opal_process_name_t *proc;
|
|
|
|
|
const char *msg;
|
|
|
|
|
void *server_object;
|
2015-09-07 19:03:59 +03:00
|
|
|
opal_list_t *procs;
|
2016-01-11 19:46:31 +03:00
|
|
|
opal_list_t *eprocs;
|
2015-09-07 19:03:59 +03:00
|
|
|
opal_list_t *info;
|
2015-06-18 19:53:20 +03:00
|
|
|
opal_pmix_op_cbfunc_t cbfunc;
|
2016-06-18 01:15:13 +03:00
|
|
|
opal_pmix_info_cbfunc_t infocbfunc;
|
|
|
|
|
opal_pmix_tool_connection_cbfunc_t toolcbfunc;
|
2015-06-18 19:53:20 +03:00
|
|
|
void *cbdata;
|
|
|
|
|
} orte_pmix_server_op_caddy_t;
|
|
|
|
|
OBJ_CLASS_DECLARATION(orte_pmix_server_op_caddy_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
|
|
|
|
2015-06-18 19:53:20 +03:00
|
|
|
typedef struct {
|
|
|
|
|
opal_object_t super;
|
|
|
|
|
orte_grpcomm_signature_t *sig;
|
|
|
|
|
opal_pmix_modex_cbfunc_t cbfunc;
|
|
|
|
|
void *cbdata;
|
|
|
|
|
} orte_pmix_mdx_caddy_t;
|
|
|
|
|
OBJ_CLASS_DECLARATION(orte_pmix_mdx_caddy_t);
|
|
|
|
|
|
2016-06-06 07:39:44 +03:00
|
|
|
#define ORTE_DMX_REQ(p, cf, ocf, ocd) \
|
|
|
|
|
do { \
|
|
|
|
|
pmix_server_req_t *_req; \
|
|
|
|
|
_req = OBJ_NEW(pmix_server_req_t); \
|
|
|
|
|
_req->target = (p); \
|
|
|
|
|
_req->mdxcbfunc = (ocf); \
|
|
|
|
|
_req->cbdata = (ocd); \
|
|
|
|
|
opal_event_set(orte_event_base, &(_req->ev), \
|
|
|
|
|
-1, OPAL_EV_WRITE, (cf), _req); \
|
|
|
|
|
opal_event_set_priority(&(_req->ev), ORTE_MSG_PRI); \
|
|
|
|
|
opal_event_active(&(_req->ev), OPAL_EV_WRITE, 1); \
|
|
|
|
|
} while(0);
|
2015-06-18 19:53:20 +03:00
|
|
|
|
2016-06-06 07:39:44 +03:00
|
|
|
#define ORTE_SPN_REQ(j, cf, ocf, ocd) \
|
|
|
|
|
do { \
|
|
|
|
|
pmix_server_req_t *_req; \
|
|
|
|
|
_req = OBJ_NEW(pmix_server_req_t); \
|
|
|
|
|
_req->jdata = (j); \
|
|
|
|
|
_req->spcbfunc = (ocf); \
|
|
|
|
|
_req->cbdata = (ocd); \
|
|
|
|
|
opal_event_set(orte_event_base, &(_req->ev), \
|
|
|
|
|
-1, OPAL_EV_WRITE, (cf), _req); \
|
|
|
|
|
opal_event_set_priority(&(_req->ev), ORTE_MSG_PRI); \
|
|
|
|
|
opal_event_active(&(_req->ev), OPAL_EV_WRITE, 1); \
|
|
|
|
|
} while(0);
|
2015-06-18 19:53:20 +03:00
|
|
|
|
2016-06-06 07:39:44 +03:00
|
|
|
#define ORTE_PMIX_OPERATION(p, i, fn, cf, cb) \
|
|
|
|
|
do { \
|
|
|
|
|
orte_pmix_server_op_caddy_t *_cd; \
|
|
|
|
|
_cd = OBJ_NEW(orte_pmix_server_op_caddy_t); \
|
|
|
|
|
_cd->procs = (p); \
|
|
|
|
|
_cd->info = (i); \
|
|
|
|
|
_cd->cbfunc = (cf); \
|
|
|
|
|
_cd->cbdata = (cb); \
|
|
|
|
|
opal_event_set(orte_event_base, &(_cd->ev), -1, \
|
|
|
|
|
OPAL_EV_WRITE, (fn), _cd); \
|
|
|
|
|
opal_event_set_priority(&(_cd->ev), ORTE_MSG_PRI); \
|
|
|
|
|
opal_event_active(&(_cd->ev), OPAL_EV_WRITE, 1); \
|
|
|
|
|
} while(0);
|
2015-06-18 19:53:20 +03:00
|
|
|
|
2016-06-06 07:39:44 +03:00
|
|
|
#define ORTE_PMIX_THREADSHIFT(p, s, st, m, pl, fn, cf, cb) \
|
|
|
|
|
do { \
|
|
|
|
|
orte_pmix_server_op_caddy_t *_cd; \
|
|
|
|
|
_cd = OBJ_NEW(orte_pmix_server_op_caddy_t); \
|
|
|
|
|
_cd->proc = (p); \
|
|
|
|
|
_cd->server_object = (s); \
|
|
|
|
|
_cd->status = (st); \
|
|
|
|
|
_cd->msg = (m); \
|
|
|
|
|
_cd->procs = (pl); \
|
|
|
|
|
_cd->cbfunc = (cf); \
|
|
|
|
|
_cd->cbdata = (cb); \
|
|
|
|
|
opal_event_set(orte_event_base, &(_cd->ev), -1, \
|
|
|
|
|
OPAL_EV_WRITE, (fn), _cd); \
|
|
|
|
|
opal_event_set_priority(&(_cd->ev), ORTE_MSG_PRI); \
|
|
|
|
|
opal_event_active(&(_cd->ev), OPAL_EV_WRITE, 1); \
|
|
|
|
|
} while(0);
|
2015-06-18 19:53:20 +03:00
|
|
|
|
|
|
|
|
/* define the server module functions */
|
2016-06-02 02:34:03 +03:00
|
|
|
extern int pmix_server_client_connected_fn(opal_process_name_t *proc, void* server_object,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
2015-06-18 19:53:20 +03:00
|
|
|
extern int pmix_server_client_finalized_fn(opal_process_name_t *proc, void* server_object,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_abort_fn(opal_process_name_t *proc, void *server_object,
|
|
|
|
|
int status, const char msg[],
|
|
|
|
|
opal_list_t *procs_to_abort,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_fencenb_fn(opal_list_t *procs, opal_list_t *info,
|
|
|
|
|
char *data, size_t ndata,
|
|
|
|
|
opal_pmix_modex_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_dmodex_req_fn(opal_process_name_t *proc, opal_list_t *info,
|
|
|
|
|
opal_pmix_modex_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_publish_fn(opal_process_name_t *proc,
|
|
|
|
|
opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
2015-09-04 18:29:09 +03:00
|
|
|
extern int pmix_server_lookup_fn(opal_process_name_t *proc, char **keys,
|
|
|
|
|
opal_list_t *info,
|
2015-06-18 19:53:20 +03:00
|
|
|
opal_pmix_lookup_cbfunc_t cbfunc, void *cbdata);
|
2015-09-04 18:29:09 +03:00
|
|
|
extern int pmix_server_unpublish_fn(opal_process_name_t *proc, char **keys,
|
|
|
|
|
opal_list_t *info,
|
2015-06-18 19:53:20 +03:00
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_spawn_fn(opal_process_name_t *requestor,
|
|
|
|
|
opal_list_t *job_info, opal_list_t *apps,
|
|
|
|
|
opal_pmix_spawn_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_connect_fn(opal_list_t *procs, opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_disconnect_fn(opal_list_t *procs, opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern int pmix_server_register_events_fn(opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc,
|
|
|
|
|
void *cbdata);
|
2015-12-17 06:06:47 +03:00
|
|
|
extern int pmix_server_deregister_events_fn(opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc,
|
|
|
|
|
void *cbdata);
|
2016-10-08 19:24:28 +03:00
|
|
|
extern int pmix_server_notify_event(int code, opal_process_name_t *source,
|
|
|
|
|
opal_list_t *info,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
|
2016-06-18 01:15:13 +03:00
|
|
|
extern int pmix_server_query_fn(opal_process_name_t *requestor,
|
2016-08-08 06:41:54 +03:00
|
|
|
opal_list_t *queries,
|
2016-06-18 01:15:13 +03:00
|
|
|
opal_pmix_info_cbfunc_t cbfunc, void *cbdata);
|
|
|
|
|
extern void pmix_tool_connected_fn(opal_list_t *info,
|
|
|
|
|
opal_pmix_tool_connection_cbfunc_t cbfunc,
|
|
|
|
|
void *cbdata);
|
2015-06-18 19:53:20 +03:00
|
|
|
|
2016-08-08 06:41:54 +03:00
|
|
|
extern void pmix_server_log_fn(opal_process_name_t *requestor,
|
|
|
|
|
opal_list_t *info,
|
|
|
|
|
opal_list_t *directives,
|
|
|
|
|
opal_pmix_op_cbfunc_t cbfunc,
|
|
|
|
|
void *cbdata);
|
|
|
|
|
|
2015-08-31 06:54:45 +03:00
|
|
|
/* declare the RML recv functions for responses */
|
2015-06-18 19:53:20 +03:00
|
|
|
extern void pmix_server_launch_resp(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
|
|
|
|
2015-08-31 06:54:45 +03:00
|
|
|
extern void pmix_server_keyval_client(int status, orte_process_name_t* sender,
|
|
|
|
|
opal_buffer_t *buffer,
|
|
|
|
|
orte_rml_tag_t tg, void *cbdata);
|
|
|
|
|
|
2016-01-11 19:46:31 +03:00
|
|
|
extern void pmix_server_notify(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
|
|
|
/* exposed shared variables */
|
2015-06-18 19:53:20 +03:00
|
|
|
typedef struct {
|
|
|
|
|
bool initialized;
|
|
|
|
|
int verbosity;
|
|
|
|
|
int output;
|
|
|
|
|
opal_hotel_t reqs;
|
|
|
|
|
int num_rooms;
|
|
|
|
|
int timeout;
|
2015-09-04 18:29:09 +03:00
|
|
|
char *server_uri;
|
|
|
|
|
bool wait_for_server;
|
2015-06-18 19:53:20 +03:00
|
|
|
orte_process_name_t server;
|
2016-01-11 19:46:31 +03:00
|
|
|
opal_list_t notifications;
|
2015-06-18 19:53:20 +03:00
|
|
|
} pmix_server_globals_t;
|
|
|
|
|
|
|
|
|
|
extern pmix_server_globals_t orte_pmix_server_globals;
|
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
|
|
|
|
|
|
|
|
END_C_DECLS
|
|
|
|
|
|
|
|
|
|
#endif /* PMIX_SERVER_INTERNAL_H_ */
|
|
|
|
|
|