1
1
openmpi/orte/mca/plm/rsh/plm_rsh_module.c

1616 строки
59 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2007 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2006 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) 2006-2007 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2007-2012 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2008-2009 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2011 IBM Corporation. All rights reserved.
* Copyright (c) 2014-2015 Intel Corporation. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
* These symbols are in a file by themselves to provide nice linker
* semantics. Since linkers generally pull in symbols by object
* files, keeping these symbols as the only symbols in this file
* prevents utility programs such as "ompi_info" from having to import
* entire components just to query their version and parameters.
*/
#include "orte_config.h"
#include "orte/constants.h"
#include <stdlib.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <errno.h>
#include <string.h>
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#ifdef HAVE_TIME_H
#include <time.h>
#endif
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif
#include <fcntl.h>
#include <signal.h>
#ifdef HAVE_PWD_H
#include <pwd.h>
#endif
#include "opal/mca/installdirs/installdirs.h"
#include "opal/util/output.h"
#include "opal/mca/base/base.h"
Update libevent to the 2.0 series, currently at 2.0.7rc. We will update to their final release when it becomes available. Currently known errors exist in unused portions of the libevent code. This revision passes the IBM test suite on a Linux machine and on a standalone Mac. This is a fairly intrusive change, but outside of the moving of opal/event to opal/mca/event, the only changes involved (a) changing all calls to opal_event functions to reflect the new framework instead, and (b) ensuring that all opal_event_t objects are properly constructed since they are now true opal_objects. Note: Shiqing has just returned from vacation and has not yet had a chance to complete the Windows integration. Thus, this commit almost certainly breaks Windows support on the trunk. However, I want this to have a chance to soak for as long as possible before I become less available a week from today (going to be at a class for 5 days, and thus will only be sparingly available) so we can find and fix any problems. Biggest change is moving the libevent code from opal/event to a new opal/mca/event framework. This was done to make it much easier to update libevent in the future. New versions can be inserted as a new component and tested in parallel with the current version until validated, then we can remove the earlier version if we so choose. This is a statically built framework ala installdirs, so only one component will build at a time. There is no selection logic - the sole compiled component simply loads its function pointers into the opal_event struct. I have gone thru the code base and converted all the libevent calls I could find. However, I cannot compile nor test every environment. It is therefore quite likely that errors remain in the system. Please keep an eye open for two things: 1. compile-time errors: these will be obvious as calls to the old functions (e.g., opal_evtimer_new) must be replaced by the new framework APIs (e.g., opal_event.evtimer_new) 2. run-time errors: these will likely show up as segfaults due to missing constructors on opal_event_t objects. It appears that it became a typical practice for people to "init" an opal_event_t by simply using memset to zero it out. This will no longer work - you must either OBJ_NEW or OBJ_CONSTRUCT an opal_event_t. I tried to catch these cases, but may have missed some. Believe me, you'll know when you hit it. There is also the issue of the new libevent "no recursion" behavior. As I described on a recent email, we will have to discuss this and figure out what, if anything, we need to do. This commit was SVN r23925.
2010-10-24 18:35:54 +00:00
#include "opal/mca/event/event.h"
#include "opal/util/argv.h"
#include "opal/util/opal_environ.h"
#include "opal/util/basename.h"
#include "opal/util/path.h"
#include "opal/class/opal_pointer_array.h"
#include "orte/util/show_help.h"
#include "orte/runtime/orte_wait.h"
#include "orte/runtime/orte_globals.h"
#include "orte/util/name_fns.h"
#include "orte/util/nidmap.h"
#include "orte/util/proc_info.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/ess/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rmaps/rmaps.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/rml/base/rml_contact.h"
#include "orte/mca/state/state.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/base.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/plm/rsh/plm_rsh.h"
static int rsh_init(void);
static int rsh_launch(orte_job_t *jdata);
static int remote_spawn(opal_buffer_t *launch);
static int rsh_terminate_orteds(void);
static int rsh_finalize(void);
orte_plm_base_module_t orte_plm_rsh_module = {
rsh_init,
orte_plm_base_set_hnp_name,
rsh_launch,
remote_spawn,
orte_plm_base_orted_terminate_job,
rsh_terminate_orteds,
orte_plm_base_orted_kill_local_procs,
orte_plm_base_orted_signal_local_procs,
rsh_finalize
};
typedef struct {
opal_list_item_t super;
int argc;
char **argv;
orte_proc_t *daemon;
} orte_plm_rsh_caddy_t;
static void caddy_const(orte_plm_rsh_caddy_t *ptr)
{
ptr->argv = NULL;
ptr->daemon = NULL;
}
static void caddy_dest(orte_plm_rsh_caddy_t *ptr)
{
if (NULL != ptr->argv) {
opal_argv_free(ptr->argv);
}
if (NULL != ptr->daemon) {
OBJ_RELEASE(ptr->daemon);
}
}
OBJ_CLASS_INSTANCE(orte_plm_rsh_caddy_t,
opal_list_item_t,
caddy_const, caddy_dest);
typedef enum {
ORTE_PLM_RSH_SHELL_BASH = 0,
ORTE_PLM_RSH_SHELL_ZSH,
ORTE_PLM_RSH_SHELL_TCSH,
ORTE_PLM_RSH_SHELL_CSH,
ORTE_PLM_RSH_SHELL_KSH,
ORTE_PLM_RSH_SHELL_SH,
ORTE_PLM_RSH_SHELL_UNKNOWN
} orte_plm_rsh_shell_t;
/* These strings *must* follow the same order as the enum ORTE_PLM_RSH_SHELL_* */
static const char *orte_plm_rsh_shell_name[7] = {
"bash",
"zsh",
"tcsh", /* tcsh has to be first otherwise strstr finds csh */
"csh",
"ksh",
"sh",
"unknown"
};
/*
* Local functions
*/
static void set_handler_default(int sig);
static orte_plm_rsh_shell_t find_shell(char *shell);
static int launch_agent_setup(const char *agent, char *path);
static void ssh_child(int argc, char **argv) __opal_attribute_noreturn__;
static int rsh_probe(char *nodename,
orte_plm_rsh_shell_t *shell);
static int setup_shell(orte_plm_rsh_shell_t *rshell,
orte_plm_rsh_shell_t *lshell,
char *nodename, int *argc, char ***argv);
static void launch_daemons(int fd, short args, void *cbdata);
static void process_launch_list(int fd, short args, void *cbdata);
/* local global storage */
static char *rsh_agent_path=NULL;
static char **rsh_agent_argv=NULL;
static int num_in_progress=0;
static opal_list_t launch_list;
static opal_event_t launch_event;
/**
* Init the module
*/
static int rsh_init(void)
{
char *tmp;
int rc;
/* we were selected, so setup the launch agent */
if (mca_plm_rsh_component.using_qrsh) {
/* perform base setup for qrsh */
(void)asprintf(&tmp, "%s/bin/%s", getenv("SGE_ROOT"), getenv("ARC"));
if (ORTE_SUCCESS != (rc = launch_agent_setup("qrsh", tmp))) {
ORTE_ERROR_LOG(rc);
free(tmp);
return rc;
}
free(tmp);
/* automatically add -inherit and grid engine PE related flags */
opal_argv_append_nosize(&rsh_agent_argv, "-inherit");
/* Don't use the "-noshell" flag as qrsh would have a problem
* swallowing a long command */
opal_argv_append_nosize(&rsh_agent_argv, "-nostdin");
opal_argv_append_nosize(&rsh_agent_argv, "-V");
if (0 < opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
opal_argv_append_nosize(&rsh_agent_argv, "-verbose");
tmp = opal_argv_join(rsh_agent_argv, ' ');
opal_output_verbose(1, orte_plm_base_framework.framework_output,
"%s plm:rsh: using \"%s\" for launching\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
free(tmp);
}
} else if(mca_plm_rsh_component.using_llspawn) {
/* perform base setup for llspawn */
if (ORTE_SUCCESS != (rc = launch_agent_setup("llspawn", NULL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
opal_output_verbose(1, orte_plm_base_framework.framework_output,
"%s plm:rsh: using \"%s\" for launching\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
rsh_agent_path);
} else {
/* not using qrsh or llspawn - use MCA-specified agent */
if (ORTE_SUCCESS != (rc = launch_agent_setup(mca_plm_rsh_component.agent, NULL))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
/* point to our launch command */
if (ORTE_SUCCESS != (rc = orte_state.add_job_state(ORTE_JOB_STATE_LAUNCH_DAEMONS,
launch_daemons, ORTE_SYS_PRI))) {
ORTE_ERROR_LOG(rc);
return rc;
}
/* setup the event for metering the launch */
OBJ_CONSTRUCT(&launch_list, opal_list_t);
opal_event_set(orte_event_base, &launch_event, -1, 0, process_launch_list, NULL);
opal_event_set_priority(&launch_event, ORTE_SYS_PRI);
/* start the recvs */
if (ORTE_SUCCESS != (rc = orte_plm_base_comm_start())) {
ORTE_ERROR_LOG(rc);
}
/* we assign daemon nodes at launch */
orte_plm_globals.daemon_nodes_assigned_at_launch = true;
return rc;
}
/**
* Callback on daemon exit.
*/
static void rsh_wait_daemon(orte_proc_t *daemon, void* cbdata)
{
orte_job_t *jdata;
orte_plm_rsh_caddy_t *caddy=(orte_plm_rsh_caddy_t*)cbdata;
if (orte_orteds_term_ordered || orte_abnormal_term_ordered) {
/* ignore any such report - it will occur if we left the
* session attached, e.g., while debugging
*/
OBJ_RELEASE(caddy);
return;
}
if (! WIFEXITED(daemon->exit_code) ||
! WEXITSTATUS(daemon->exit_code) == 0) { /* if abnormal exit */
/* if we are not the HNP, send a message to the HNP alerting it
* to the failure
*/
if (!ORTE_PROC_IS_HNP) {
opal_buffer_t *buf;
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s daemon %d failed with status %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)daemon->name.vpid, WEXITSTATUS(daemon->exit_code)));
buf = OBJ_NEW(opal_buffer_t);
opal_dss.pack(buf, &(daemon->name.vpid), 1, ORTE_VPID);
opal_dss.pack(buf, &daemon->exit_code, 1, OPAL_INT);
orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, buf,
As per the RFC, bring in the ORTE async progress code and the rewrite of OOB: *** THIS RFC INCLUDES A MINOR CHANGE TO THE MPI-RTE INTERFACE *** Note: during the course of this work, it was necessary to completely separate the MPI and RTE progress engines. There were multiple places in the MPI layer where ORTE_WAIT_FOR_COMPLETION was being used. A new OMPI_WAIT_FOR_COMPLETION macro was created (defined in ompi/mca/rte/rte.h) that simply cycles across opal_progress until the provided flag becomes false. Places where the MPI layer blocked waiting for RTE to complete an event have been modified to use this macro. *************************************************************************************** I am reissuing this RFC because of the time that has passed since its original release. Since its initial release and review, I have debugged it further to ensure it fully supports tests like loop_spawn. It therefore seems ready for merge back to the trunk. Given its prior review, I have set the timeout for one week. The code is in https://bitbucket.org/rhc/ompi-oob2 WHAT: Rewrite of ORTE OOB WHY: Support asynchronous progress and a host of other features WHEN: Wed, August 21 SYNOPSIS: The current OOB has served us well, but a number of limitations have been identified over the years. Specifically: * it is only progressed when called via opal_progress, which can lead to hangs or recursive calls into libevent (which is not supported by that code) * we've had issues when multiple NICs are available as the code doesn't "shift" messages between transports - thus, all nodes had to be available via the same TCP interface. * the OOB "unloads" incoming opal_buffer_t objects during the transmission, thus preventing use of OBJ_RETAIN in the code when repeatedly sending the same message to multiple recipients * there is no failover mechanism across NICs - if the selected NIC (or its attached switch) fails, we are forced to abort * only one transport (i.e., component) can be "active" The revised OOB resolves these problems: * async progress is used for all application processes, with the progress thread blocking in the event library * each available TCP NIC is supported by its own TCP module. The ability to asynchronously progress each module independently is provided, but not enabled by default (a runtime MCA parameter turns it "on") * multi-address TCP NICs (e.g., a NIC with both an IPv4 and IPv6 address, or with virtual interfaces) are supported - reachability is determined by comparing the contact info for a peer against all addresses within the range covered by the address/mask pairs for the NIC. * a message that arrives on one TCP NIC is automatically shifted to whatever NIC that is connected to the next "hop" if that peer cannot be reached by the incoming NIC. If no TCP module will reach the peer, then the OOB attempts to send the message via all other available components - if none can reach the peer, then an "error" is reported back to the RML, which then calls the errmgr for instructions. * opal_buffer_t now conforms to standard object rules re OBJ_RETAIN as we no longer "unload" the incoming object * NIC failure is reported to the TCP component, which then tries to resend the message across any other available TCP NIC. If that doesn't work, then the message is given back to the OOB base to try using other components. If all that fails, then the error is reported to the RML, which reports to the errmgr for instructions * obviously from the above, multiple OOB components (e.g., TCP and UD) can be active in parallel * the matching code has been moved to the RML (and out of the OOB/TCP component) so it is independent of transport * routing is done by the individual OOB modules (as opposed to the RML). Thus, both routed and non-routed transports can simultaneously be active * all blocking send/recv APIs have been removed. Everything operates asynchronously. KNOWN LIMITATIONS: * although provision is made for component failover as described above, the code for doing so has not been fully implemented yet. At the moment, if all connections for a given peer fail, the errmgr is notified of a "lost connection", which by default results in termination of the job if it was a lifeline * the IPv6 code is present and compiles, but is not complete. Since the current IPv6 support in the OOB doesn't work anyway, I don't consider this a blocker * routing is performed at the individual module level, yet the active routed component is selected on a global basis. We probably should update that to reflect that different transports may need/choose to route in different ways * obviously, not every error path has been tested nor necessarily covered * determining abnormal termination is more challenging than in the old code as we now potentially have multiple ways of connecting to a process. Ideally, we would declare "connection failed" when *all* transports can no longer reach the process, but that requires some additional (possibly complex) code. For now, the code replicates the old behavior only somewhat modified - i.e., if a module sees its connection fail, it checks to see if it is a lifeline. If so, it notifies the errmgr that the lifeline is lost - otherwise, it notifies the errmgr that a non-lifeline connection was lost. * reachability is determined solely on the basis of a shared subnet address/mask - more sophisticated algorithms (e.g., the one used in the tcp btl) are required to handle routing via gateways * the RML needs to assign sequence numbers to each message on a per-peer basis. The receiving RML will then deliver messages in order, thus preventing out-of-order messaging in the case where messages travel across different transports or a message needs to be redirected/resent due to failure of a NIC This commit was SVN r29058.
2013-08-22 16:37:40 +00:00
ORTE_RML_TAG_REPORT_REMOTE_LAUNCH,
orte_rml_send_callback, NULL);
/* note that this daemon failed */
daemon->state = ORTE_PROC_STATE_FAILED_TO_START;
} else {
jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s daemon %d failed with status %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(int)daemon->name.vpid, WEXITSTATUS(daemon->exit_code)));
/* set the exit status */
ORTE_UPDATE_EXIT_STATUS(WEXITSTATUS(daemon->exit_code));
/* note that this daemon failed */
daemon->state = ORTE_PROC_STATE_FAILED_TO_START;
/* increment the #daemons terminated so we will exit properly */
jdata->num_terminated++;
/* remove it from the routing table to ensure num_routes
* returns the correct value
*/
orte_routed.route_lost(&daemon->name);
/* report that the daemon has failed so we can exit */
ORTE_ACTIVATE_PROC_STATE(&daemon->name, ORTE_PROC_STATE_FAILED_TO_START);
}
}
/* release any delay */
--num_in_progress;
if (num_in_progress < mca_plm_rsh_component.num_concurrent) {
/* trigger continuation of the launch */
opal_event_active(&launch_event, EV_WRITE, 1);
}
/* cleanup */
OBJ_RELEASE(caddy);
}
static int setup_launch(int *argcptr, char ***argvptr,
char *nodename,
int *node_name_index1,
int *proc_vpid_index, char *prefix_dir)
{
int argc;
char **argv;
char *param, *value;
orte_plm_rsh_shell_t remote_shell, local_shell;
int orted_argc;
char **orted_argv;
char *orted_cmd, *orted_prefix, *final_cmd;
int orted_index;
int rc;
int i, j;
bool found;
char *lib_base=NULL, *bin_base=NULL;
char *opal_prefix = getenv("OPAL_PREFIX");
char* full_orted_cmd = NULL;
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* Figure out the basenames for the libdir and bindir. This
requires some explanation:
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
- Use opal_install_dirs.libdir and opal_install_dirs.bindir.
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
- After a discussion on the devel-core mailing list, the
developers decided that we should use the local directory
basenames as the basis for the prefix on the remote note.
This does not handle a few notable cases (e.g., if the
libdir/bindir is not simply a subdir under the prefix, if the
libdir/bindir basename is not the same on the remote node as
it is here on the local node, etc.), but we decided that
--prefix was meant to handle "the common case". If you need
something more complex than this, a) edit your shell startup
files to set PATH/LD_LIBRARY_PATH properly on the remove
node, or b) use some new/to-be-defined options that
explicitly allow setting the bindir/libdir on the remote
node. We decided to implement these options (e.g.,
--remote-bindir and --remote-libdir) to orterun when it
actually becomes a problem for someone (vs. a hypothetical
situation).
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
Hence, for now, we simply take the basename of this install's
libdir and bindir and use it to append this install's prefix
and use that on the remote node.
*/
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/*
* Build argv array
*/
argv = opal_argv_copy(rsh_agent_argv);
argc = opal_argv_count(rsh_agent_argv);
/* if any ssh args were provided, now is the time to add them */
if (NULL != mca_plm_rsh_component.ssh_args) {
char **ssh_argv;
ssh_argv = opal_argv_split(mca_plm_rsh_component.ssh_args, ' ');
for (i=0; NULL != ssh_argv[i]; i++) {
opal_argv_append(&argc, &argv, ssh_argv[i]);
}
opal_argv_free(ssh_argv);
}
*node_name_index1 = argc;
opal_argv_append(&argc, &argv, "<template>");
/* setup the correct shell info */
if (ORTE_SUCCESS != (rc = setup_shell(&remote_shell, &local_shell,
nodename, &argc, &argv))) {
ORTE_ERROR_LOG(rc);
return rc;
}
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* now get the orted cmd - as specified by user - into our tmp array.
* The function returns the location where the actual orted command is
* located - usually in the final spot, but someone could
* have added options. For example, it should be legal for them to use
* "orted --debug-devel" so they get debug output from the orteds, but
* not from mpirun. Also, they may have a customized version of orted
* that takes arguments in addition to the std ones we already support
*/
orted_argc = 0;
orted_argv = NULL;
orted_index = orte_plm_base_setup_orted_cmd(&orted_argc, &orted_argv);
/* look at the returned orted cmd argv to check several cases:
*
* - only "orted" was given. This is the default and thus most common
* case. In this situation, there is nothing we need to do
*
* - something was given that doesn't include "orted" - i.e., someone
* has substituted their own daemon. There isn't anything we can
* do here, so we want to avoid adding prefixes to the cmd
*
* - something was given that precedes "orted". For example, someone
* may have specified "valgrind [options] orted". In this case, we
* need to separate out that "orted_prefix" section so it can be
* treated separately below
*
* - something was given that follows "orted". An example was given above.
* In this case, we need to construct the effective "orted_cmd" so it
* can be treated properly below
*
* Obviously, the latter two cases can be combined - just to make it
* even more interesting! Gotta love rsh/ssh...
*/
if (0 == orted_index) {
/* single word cmd - this is the default scenario, but there could
* be options specified so we need to account for that possibility.
* However, we don't need/want a prefix as nothing precedes the orted
* cmd itself
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
*/
orted_cmd = opal_argv_join(orted_argv, ' ');
orted_prefix = NULL;
} else {
/* okay, so the "orted" cmd is somewhere in this array, with
* something preceding it and perhaps things following it.
*/
orted_prefix = opal_argv_join_range(orted_argv, 0, orted_index, ' ');
orted_cmd = opal_argv_join_range(orted_argv, orted_index, opal_argv_count(orted_argv), ' ');
}
opal_argv_free(orted_argv); /* done with this */
/* if the user specified a library path to pass, set it up now */
param = opal_basename(opal_install_dirs.libdir);
if (NULL != mca_plm_rsh_component.pass_libpath) {
if (NULL != prefix_dir) {
asprintf(&lib_base, "%s:%s/%s", mca_plm_rsh_component.pass_libpath, prefix_dir, param);
} else {
asprintf(&lib_base, "%s:%s", mca_plm_rsh_component.pass_libpath, param);
}
} else if (NULL != prefix_dir) {
asprintf(&lib_base, "%s/%s", prefix_dir, param);
}
2015-04-18 07:48:22 -07:00
free(param);
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* we now need to assemble the actual cmd that will be executed - this depends
* upon whether or not a prefix directory is being used
*/
if (NULL != prefix_dir) {
/* if we have a prefix directory, we need to set the PATH and
* LD_LIBRARY_PATH on the remote node, and prepend just the orted_cmd
* with the prefix directory
*/
value = opal_basename(opal_install_dirs.bindir);
asprintf(&bin_base, "%s/%s", prefix_dir, value);
2015-04-18 07:48:22 -07:00
free(value);
if (NULL != orted_cmd) {
if (0 == strcmp(orted_cmd, "orted")) {
/* if the cmd is our standard one, then add the prefix */
(void)asprintf(&full_orted_cmd, "%s/%s", bin_base, orted_cmd);
} else {
/* someone specified something different, so don't prefix it */
full_orted_cmd = strdup(orted_cmd);
}
free(orted_cmd);
}
2015-04-18 07:48:22 -07:00
} else {
full_orted_cmd = orted_cmd;
}
2015-04-18 07:48:22 -07:00
if (NULL != lib_base || NULL != bin_base) {
if (ORTE_PLM_RSH_SHELL_SH == remote_shell ||
ORTE_PLM_RSH_SHELL_KSH == remote_shell ||
ORTE_PLM_RSH_SHELL_ZSH == remote_shell ||
ORTE_PLM_RSH_SHELL_BASH == remote_shell) {
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* if there is nothing preceding orted, then we can just
* assemble the cmd with the orted_cmd at the end. Otherwise,
* we have to insert the orted_prefix in the right place
*/
(void)asprintf (&final_cmd,
"%s%s%s PATH=%s%s$PATH ; export PATH ; "
"LD_LIBRARY_PATH=%s%s$LD_LIBRARY_PATH ; export LD_LIBRARY_PATH ; "
"DYLD_LIBRARY_PATH=%s%s$DYLD_LIBRARY_PATH ; export DYLD_LIBRARY_PATH ; "
"%s %s",
(opal_prefix != NULL ? "OPAL_PREFIX=" : " "),
(opal_prefix != NULL ? opal_prefix : " "),
(opal_prefix != NULL ? " ; export OPAL_PREFIX;" : " "),
(NULL != bin_base ? bin_base : " "),
(NULL != bin_base ? ":" : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? ":" : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? ":" : " "),
(orted_prefix != NULL ? orted_prefix : " "),
(full_orted_cmd != NULL ? full_orted_cmd : " "));
} else if (ORTE_PLM_RSH_SHELL_TCSH == remote_shell ||
ORTE_PLM_RSH_SHELL_CSH == remote_shell) {
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* [t]csh is a bit more challenging -- we
have to check whether LD_LIBRARY_PATH
is already set before we try to set it.
Must be very careful about obeying
[t]csh's order of evaluation and not
using a variable before it is defined.
See this thread for more details:
http://www.open-mpi.org/community/lists/users/2006/01/0517.php. */
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
/* if there is nothing preceding orted, then we can just
* assemble the cmd with the orted_cmd at the end. Otherwise,
* we have to insert the orted_prefix in the right place
*/
(void)asprintf (&final_cmd,
"%s%s%s set path = ( %s $path ) ; "
"if ( $?LD_LIBRARY_PATH == 1 ) "
"set OMPI_have_llp ; "
"if ( $?LD_LIBRARY_PATH == 0 ) "
"setenv LD_LIBRARY_PATH %s ; "
"if ( $?OMPI_have_llp == 1 ) "
"setenv LD_LIBRARY_PATH %s%s$LD_LIBRARY_PATH ; "
"if ( $?DYLD_LIBRARY_PATH == 1 ) "
"set OMPI_have_dllp ; "
"if ( $?DYLD_LIBRARY_PATH == 0 ) "
"setenv DYLD_LIBRARY_PATH %s ; "
"if ( $?OMPI_have_dllp == 1 ) "
"setenv DYLD_LIBRARY_PATH %s%s$DYLD_LIBRARY_PATH ; "
"%s %s",
(opal_prefix != NULL ? "setenv OPAL_PREFIX " : " "),
(opal_prefix != NULL ? opal_prefix : " "),
(opal_prefix != NULL ? " ;" : " "),
(NULL != bin_base ? bin_base : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? ":" : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? lib_base : " "),
(NULL != lib_base ? ":" : " "),
(orted_prefix != NULL ? orted_prefix : " "),
(full_orted_cmd != NULL ? full_orted_cmd : " "));
} else {
orte_show_help("help-plm-rsh.txt", "cannot-resolve-shell-with-prefix", true,
(NULL == opal_prefix) ? "NULL" : opal_prefix,
prefix_dir);
if (NULL != bin_base) {
free(bin_base);
}
if (NULL != lib_base) {
free(lib_base);
}
if (NULL != orted_prefix) free(orted_prefix);
if (NULL != full_orted_cmd) free(full_orted_cmd);
return ORTE_ERR_SILENT;
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
}
if (NULL != bin_base) {
free(bin_base);
}
if (NULL != lib_base) {
free(lib_base);
}
if( NULL != full_orted_cmd ) {
free(full_orted_cmd);
}
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
} else {
/* no prefix directory, so just aggregate the result */
(void)asprintf(&final_cmd, "%s %s",
(orted_prefix != NULL ? orted_prefix : ""),
2015-04-18 07:48:22 -07:00
(full_orted_cmd != NULL ? full_orted_cmd : ""));
if (NULL != full_orted_cmd) {
free(full_orted_cmd);
}
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
}
/* now add the final cmd to the argv array */
opal_argv_append(&argc, &argv, final_cmd);
free(final_cmd); /* done with this */
if (NULL != orted_prefix) free(orted_prefix);
/* if we are not tree launching or debugging, tell the daemon
* to daemonize so we can launch the next group
*/
if (mca_plm_rsh_component.no_tree_spawn &&
!orte_debug_flag &&
!orte_debug_daemons_flag &&
!orte_debug_daemons_file_flag &&
!orte_leave_session_attached &&
/* Daemonize when not using qrsh. Or, if using qrsh, only
* daemonize if told to by user with daemonize_qrsh flag. */
((!mca_plm_rsh_component.using_qrsh) ||
(mca_plm_rsh_component.using_qrsh && mca_plm_rsh_component.daemonize_qrsh)) &&
((!mca_plm_rsh_component.using_llspawn) ||
(mca_plm_rsh_component.using_llspawn && mca_plm_rsh_component.daemonize_llspawn))) {
opal_argv_append(&argc, &argv, "--daemonize");
}
/*
* Add the basic arguments to the orted command line, including
* all debug options
*/
orte_plm_base_orted_append_basic_args(&argc, &argv,
"env",
proc_vpid_index,
NULL);
/* ensure that only the ssh plm is selected on the remote daemon */
opal_argv_append_nosize(&argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append_nosize(&argv, "plm");
opal_argv_append_nosize(&argv, "rsh");
/* unless told otherwise... */
if (mca_plm_rsh_component.pass_environ_mca_params) {
/* now check our local environment for MCA params - add them
* only if they aren't already present
*/
for (i = 0; NULL != environ[i]; ++i) {
if (0 == strncmp(OPAL_MCA_PREFIX"mca_base_env_list", environ[i],
strlen(OPAL_MCA_PREFIX"mca_base_env_list"))) {
/* ignore this one */
continue;
}
if (0 == strncmp(OPAL_MCA_PREFIX, environ[i], 9)) {
/* check for duplicate in app->env - this
* would have been placed there by the
* cmd line processor. By convention, we
* always let the cmd line override the
* environment
*/
param = strdup(&environ[i][9]);
value = strchr(param, '=');
*value = '\0';
value++;
found = false;
/* see if this param exists on the cmd line */
for (j=0; NULL != argv[j]; j++) {
if (0 == strcmp(param, argv[j])) {
found = true;
break;
}
}
if (!found) {
/* add it */
opal_argv_append(&argc, &argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append(&argc, &argv, param);
opal_argv_append(&argc, &argv, value);
}
free(param);
}
}
}
/* protect the params */
mca_base_cmd_line_wrap_args(argv);
/* tell the daemon we are in a tree spawn */
opal_argv_append(&argc, &argv, "--tree-spawn");
value = opal_argv_join(argv, ' ');
if (sysconf(_SC_ARG_MAX) < (int)strlen(value)) {
orte_show_help("help-plm-rsh.txt", "cmd-line-too-long",
true, strlen(value), sysconf(_SC_ARG_MAX));
free(value);
return ORTE_ERR_SILENT;
}
free(value);
if (ORTE_PLM_RSH_SHELL_SH == remote_shell ||
ORTE_PLM_RSH_SHELL_KSH == remote_shell) {
opal_argv_append(&argc, &argv, ")");
}
if (0 < opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
param = opal_argv_join(argv, ' ');
opal_output(orte_plm_base_framework.framework_output,
"%s plm:rsh: final template argv:\n\t%s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == param) ? "NULL" : param);
if (NULL != param) free(param);
}
/* all done */
*argcptr = argc;
*argvptr = argv;
return ORTE_SUCCESS;
}
/* actually ssh the child */
static void ssh_child(int argc, char **argv)
{
char** env;
char* var;
long fd, fdmax = sysconf(_SC_OPEN_MAX);
char *exec_path;
char **exec_argv;
int fdin;
sigset_t sigs;
/* setup environment */
env = opal_argv_copy(orte_launch_environ);
/* We don't need to sense an oversubscribed condition and set the sched_yield
* for the node as we are only launching the daemons at this time. The daemons
* are now smart enough to set the oversubscribed condition themselves when
* they launch the local procs.
*/
/* We cannot launch locally as this would cause multiple daemons to
* exist on a node (HNP counts as a daemon). This is taken care of
* by the earlier check for daemon_preexists, so we only have to worry
* about remote launches here
*/
exec_argv = argv;
exec_path = strdup(rsh_agent_path);
/* Don't let ssh slurp all of our stdin! */
fdin = open("/dev/null", O_RDWR);
dup2(fdin, 0);
close(fdin);
/* close all file descriptors w/ exception of stdin/stdout/stderr */
for(fd=3; fd<fdmax; fd++)
close(fd);
/* Set signal handlers back to the default. Do this close
to the execve() because the event library may (and likely
will) reset them. If we don't do this, the event
library may have left some set that, at least on some
OS's, don't get reset via fork() or exec(). Hence, the
orted could be unkillable (for example). */
set_handler_default(SIGTERM);
set_handler_default(SIGINT);
set_handler_default(SIGHUP);
set_handler_default(SIGPIPE);
set_handler_default(SIGCHLD);
/* Unblock all signals, for many of the same reasons that
we set the default handlers, above. This is noticable
on Linux where the event library blocks SIGTERM, but we
don't want that blocked by the orted (or, more
specifically, we don't want it to be blocked by the
orted and then inherited by the ORTE processes that it
forks, making them unkillable by SIGTERM). */
sigprocmask(0, 0, &sigs);
sigprocmask(SIG_UNBLOCK, &sigs, 0);
/* exec the daemon */
var = opal_argv_join(argv, ' ');
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: executing: (%s) [%s]",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
exec_path, (NULL == var) ? "NULL" : var));
if (NULL != var) free(var);
execve(exec_path, exec_argv, env);
opal_output(0, "plm:rsh: execv of %s failed with errno=%s(%d)\n",
exec_path, strerror(errno), errno);
exit(-1);
}
/*
* launch a set of daemons from a remote daemon
*/
static int remote_spawn(opal_buffer_t *launch)
{
int node_name_index1;
int proc_vpid_index;
char **argv = NULL;
char *prefix, *hostname, *var;
int argc;
int rc=ORTE_SUCCESS;
bool failed_launch = true;
orte_std_cntr_t n;
opal_byte_object_t *bo;
orte_process_name_t target;
orte_plm_rsh_caddy_t *caddy;
orte_job_t *daemons;
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 18:56:47 +00:00
opal_list_t coll;
orte_namelist_t *child;
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: remote spawn called",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
/* if we hit any errors, tell the HNP it was us */
target.vpid = ORTE_PROC_MY_NAME->vpid;
/* extract the prefix from the launch buffer */
n = 1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(launch, &prefix, &n, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* extract the byte object holding the nidmap */
n=1;
if (ORTE_SUCCESS != (rc = opal_dss.unpack(launch, &bo, &n, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* update our nidmap - this will free data in the byte object */
if (ORTE_SUCCESS != (rc = orte_util_decode_daemon_nodemap(bo))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* ensure the routing plan is updated */
orte_routed.update_routing_plan();
/* get the updated routing list */
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 18:56:47 +00:00
OBJ_CONSTRUCT(&coll, opal_list_t);
orte_routed.get_routing_list(&coll);
/* if I have no children, just 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 18:56:47 +00:00
if (0 == opal_list_get_size(&coll)) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: remote spawn - have no children!",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
failed_launch = false;
rc = ORTE_SUCCESS;
OBJ_DESTRUCT(&coll);
goto cleanup;
}
/* setup the launch */
if (ORTE_SUCCESS != (rc = setup_launch(&argc, &argv, orte_process_info.nodename, &node_name_index1,
&proc_vpid_index, prefix))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(&coll);
goto cleanup;
}
/* get the daemon job object */
if (NULL == (daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
OBJ_DESTRUCT(&coll);
goto cleanup;
}
target.jobid = ORTE_PROC_MY_NAME->jobid;
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 18:56:47 +00:00
OPAL_LIST_FOREACH(child, &coll, orte_namelist_t) {
target.vpid = child->name.vpid;
/* get the host where this daemon resides */
if (NULL == (hostname = orte_get_proc_hostname(&target))) {
opal_output(0, "%s unable to get hostname for daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_VPID_PRINT(child->name.vpid));
rc = ORTE_ERR_NOT_FOUND;
OBJ_DESTRUCT(&coll);
goto cleanup;
}
free(argv[node_name_index1]);
argv[node_name_index1] = strdup(hostname);
/* pass the vpid */
rc = orte_util_convert_vpid_to_string(&var, target.vpid);
if (ORTE_SUCCESS != rc) {
opal_output(0, "orte_plm_rsh: unable to get daemon vpid as string");
exit(-1);
}
free(argv[proc_vpid_index]);
argv[proc_vpid_index] = strdup(var);
free(var);
/* we are in an event, so no need to protect the list */
caddy = OBJ_NEW(orte_plm_rsh_caddy_t);
caddy->argc = argc;
caddy->argv = opal_argv_copy(argv);
/* fake a proc structure for the new daemon - will be released
* upon startup
*/
caddy->daemon = OBJ_NEW(orte_proc_t);
caddy->daemon->name.jobid = ORTE_PROC_MY_NAME->jobid;
caddy->daemon->name.vpid = target.vpid;
opal_list_append(&launch_list, &caddy->super);
}
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 18:56:47 +00:00
OPAL_LIST_DESTRUCT(&coll);
/* trigger the event to start processing the launch list */
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: activating launch event",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
opal_event_active(&launch_event, EV_WRITE, 1);
/* declare the launch a success */
failed_launch = false;
Per the July technical meeting: Standardize the handling of the orte launch agent option across PLMs. This has been a consistent complaint I have received - each PLM would register its own MCA param to get input on the launch agent for remote nodes (in fact, one or two didn't, but most did). This would then get handled in various and contradictory ways. Some PLMs would accept only a one-word input. Others accepted multi-word args such as "valgrind orted", but then some would error by putting any prefix specified on the cmd line in front of the incorrect argument. For example, while using the rsh launcher, if you specified "valgrind orted" as your launch agent and had "--prefix foo" on you cmd line, you would attempt to execute "ssh foo/valgrind orted" - which obviously wouldn't work. This was all -very- confusing to users, who had to know which PLM was being used so they could even set the right mca param in the first place! And since we don't warn about non-recognized or non-used mca params, half of the time they would wind up not doing what they thought they were telling us to do. To solve this problem, we did the following: 1. removed all mca params from the individual plms for the launch agent 2. added a new mca param "orte_launch_agent" for this purpose. To further simplify for users, this comes with a new cmd line option "--launch-agent" that can take a multi-word string argument. The value of the param defaults to "orted". 3. added a PLM base function that processes the orte_launch_agent value and adds the contents to a provided argv array. This can subsequently be harvested at-will to handle multi-word values 4. modified the PLMs to use this new function. All the PLMs except for the rsh PLM required very minor change - just called the function and moved on. The rsh PLM required much larger changes as - because of the rsh/ssh cmd line limitations - we had to correctly prepend any provided prefix to the correct argv entry. 5. added a new opal_argv_join_range function that allows the caller to "join" argv entries between two specified indices Please let me know of any problems. I tried to make this as clean as possible, but cannot compile all PLMs to ensure all is correct. This commit was SVN r19097.
2008-07-30 18:26:24 +00:00
cleanup:
if (NULL != argv) {
opal_argv_free(argv);
}
/* check for failed launch */
if (failed_launch) {
/* report cannot launch this daemon to HNP */
opal_buffer_t *buf;
buf = OBJ_NEW(opal_buffer_t);
opal_dss.pack(buf, &target.vpid, 1, ORTE_VPID);
opal_dss.pack(buf, &rc, 1, OPAL_INT);
orte_rml.send_buffer_nb(ORTE_PROC_MY_HNP, buf,
As per the RFC, bring in the ORTE async progress code and the rewrite of OOB: *** THIS RFC INCLUDES A MINOR CHANGE TO THE MPI-RTE INTERFACE *** Note: during the course of this work, it was necessary to completely separate the MPI and RTE progress engines. There were multiple places in the MPI layer where ORTE_WAIT_FOR_COMPLETION was being used. A new OMPI_WAIT_FOR_COMPLETION macro was created (defined in ompi/mca/rte/rte.h) that simply cycles across opal_progress until the provided flag becomes false. Places where the MPI layer blocked waiting for RTE to complete an event have been modified to use this macro. *************************************************************************************** I am reissuing this RFC because of the time that has passed since its original release. Since its initial release and review, I have debugged it further to ensure it fully supports tests like loop_spawn. It therefore seems ready for merge back to the trunk. Given its prior review, I have set the timeout for one week. The code is in https://bitbucket.org/rhc/ompi-oob2 WHAT: Rewrite of ORTE OOB WHY: Support asynchronous progress and a host of other features WHEN: Wed, August 21 SYNOPSIS: The current OOB has served us well, but a number of limitations have been identified over the years. Specifically: * it is only progressed when called via opal_progress, which can lead to hangs or recursive calls into libevent (which is not supported by that code) * we've had issues when multiple NICs are available as the code doesn't "shift" messages between transports - thus, all nodes had to be available via the same TCP interface. * the OOB "unloads" incoming opal_buffer_t objects during the transmission, thus preventing use of OBJ_RETAIN in the code when repeatedly sending the same message to multiple recipients * there is no failover mechanism across NICs - if the selected NIC (or its attached switch) fails, we are forced to abort * only one transport (i.e., component) can be "active" The revised OOB resolves these problems: * async progress is used for all application processes, with the progress thread blocking in the event library * each available TCP NIC is supported by its own TCP module. The ability to asynchronously progress each module independently is provided, but not enabled by default (a runtime MCA parameter turns it "on") * multi-address TCP NICs (e.g., a NIC with both an IPv4 and IPv6 address, or with virtual interfaces) are supported - reachability is determined by comparing the contact info for a peer against all addresses within the range covered by the address/mask pairs for the NIC. * a message that arrives on one TCP NIC is automatically shifted to whatever NIC that is connected to the next "hop" if that peer cannot be reached by the incoming NIC. If no TCP module will reach the peer, then the OOB attempts to send the message via all other available components - if none can reach the peer, then an "error" is reported back to the RML, which then calls the errmgr for instructions. * opal_buffer_t now conforms to standard object rules re OBJ_RETAIN as we no longer "unload" the incoming object * NIC failure is reported to the TCP component, which then tries to resend the message across any other available TCP NIC. If that doesn't work, then the message is given back to the OOB base to try using other components. If all that fails, then the error is reported to the RML, which reports to the errmgr for instructions * obviously from the above, multiple OOB components (e.g., TCP and UD) can be active in parallel * the matching code has been moved to the RML (and out of the OOB/TCP component) so it is independent of transport * routing is done by the individual OOB modules (as opposed to the RML). Thus, both routed and non-routed transports can simultaneously be active * all blocking send/recv APIs have been removed. Everything operates asynchronously. KNOWN LIMITATIONS: * although provision is made for component failover as described above, the code for doing so has not been fully implemented yet. At the moment, if all connections for a given peer fail, the errmgr is notified of a "lost connection", which by default results in termination of the job if it was a lifeline * the IPv6 code is present and compiles, but is not complete. Since the current IPv6 support in the OOB doesn't work anyway, I don't consider this a blocker * routing is performed at the individual module level, yet the active routed component is selected on a global basis. We probably should update that to reflect that different transports may need/choose to route in different ways * obviously, not every error path has been tested nor necessarily covered * determining abnormal termination is more challenging than in the old code as we now potentially have multiple ways of connecting to a process. Ideally, we would declare "connection failed" when *all* transports can no longer reach the process, but that requires some additional (possibly complex) code. For now, the code replicates the old behavior only somewhat modified - i.e., if a module sees its connection fail, it checks to see if it is a lifeline. If so, it notifies the errmgr that the lifeline is lost - otherwise, it notifies the errmgr that a non-lifeline connection was lost. * reachability is determined solely on the basis of a shared subnet address/mask - more sophisticated algorithms (e.g., the one used in the tcp btl) are required to handle routing via gateways * the RML needs to assign sequence numbers to each message on a per-peer basis. The receiving RML will then deliver messages in order, thus preventing out-of-order messaging in the case where messages travel across different transports or a message needs to be redirected/resent due to failure of a NIC This commit was SVN r29058.
2013-08-22 16:37:40 +00:00
ORTE_RML_TAG_REPORT_REMOTE_LAUNCH,
orte_rml_send_callback, NULL);
}
return rc;
}
/*
* Launch a daemon (bootproxy) on each node. The daemon will be responsible
* for launching the application.
*/
static int rsh_launch(orte_job_t *jdata)
{
if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_RESTART)) {
/* this is a restart situation - skip to the mapping stage */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);
} else {
/* new job - set it up */
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_INIT);
}
return ORTE_SUCCESS;
}
static void process_launch_list(int fd, short args, void *cbdata)
{
opal_list_item_t *item;
pid_t pid;
orte_plm_rsh_caddy_t *caddy;
while (num_in_progress < mca_plm_rsh_component.num_concurrent) {
item = opal_list_remove_first(&launch_list);
if (NULL == item) {
/* we are done */
break;
}
caddy = (orte_plm_rsh_caddy_t*)item;
/* register the sigchild callback */
ORTE_FLAG_SET(caddy->daemon, ORTE_PROC_FLAG_ALIVE);
orte_wait_cb(caddy->daemon, rsh_wait_daemon, (void*)caddy);
/* fork a child to exec the rsh/ssh session */
pid = fork();
if (pid < 0) {
ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_CHILDREN);
orte_wait_cb_cancel(caddy->daemon);
continue;
}
/* child */
if (pid == 0) {
/* do the ssh launch - this will exit if it fails */
ssh_child(caddy->argc, caddy->argv);
} else { /* father */
/* indicate this daemon has been launched */
caddy->daemon->state = ORTE_PROC_STATE_RUNNING;
/* record the pid */
caddy->daemon->pid = pid;
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: recording launch of daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(caddy->daemon->name))));
num_in_progress++;
}
}
}
static void launch_daemons(int fd, short args, void *cbdata)
{
orte_job_map_t *map = NULL;
int node_name_index1;
int proc_vpid_index;
char **argv = NULL;
char *prefix_dir=NULL, *var;
int argc;
int rc;
orte_app_context_t *app;
orte_node_t *node, *nd;
orte_std_cntr_t nnode;
orte_job_t *daemons;
orte_state_caddy_t *state = (orte_state_caddy_t*)cbdata;
orte_plm_rsh_caddy_t *caddy;
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 18:56:47 +00:00
opal_list_t coll;
char *username;
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 18:56:47 +00:00
orte_namelist_t *child;
/* if we are launching debugger daemons, then just go
* do it - no new daemons will be launched
*/
if (ORTE_FLAG_TEST(state->jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
state->jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
ORTE_ACTIVATE_JOB_STATE(state->jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
OBJ_RELEASE(state);
return;
}
/* setup the virtual machine */
daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
if (ORTE_SUCCESS != (rc = orte_plm_base_setup_virtual_machine(state->jdata))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/* if we don't want to launch, then don't attempt to
* launch the daemons - the user really wants to just
* look at the proposed process map
*/
if (orte_do_not_launch) {
/* set the state to indicate the daemons reported - this
* will trigger the daemons_reported event and cause the
* job to move to the following step
*/
state->jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
ORTE_ACTIVATE_JOB_STATE(state->jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
OBJ_RELEASE(state);
return;
}
/* Get the map for this job */
if (NULL == (map = daemons->map)) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
goto cleanup;
}
if (0 == map->num_new_daemons) {
/* set the state to indicate the daemons reported - this
* will trigger the daemons_reported event and cause the
* job to move to the following step
*/
state->jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
ORTE_ACTIVATE_JOB_STATE(state->jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
OBJ_RELEASE(state);
return;
}
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: launching vm",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
if ((0 < opal_output_get_verbosity(orte_plm_base_framework.framework_output) ||
orte_leave_session_attached) &&
mca_plm_rsh_component.num_concurrent < map->num_new_daemons) {
/**
* If we are in '--debug-daemons' we keep the ssh connection
* alive for the span of the run. If we use this option
* AND we launch on more than "num_concurrent" machines
* then we will deadlock. No connections are terminated
* until the job is complete, no job is started
* since all the orteds are waiting for all the others
* to come online, and the others ore not launched because
* we are waiting on those that have started to terminate
* their ssh tunnels. :(
* As we cannot run in this situation, pretty print the error
* and return an error code.
*/
This commit represents a bunch of work on a Mercurial side branch. As such, the commit message back to the master SVN repository is fairly long. = ORTE Job-Level Output Messages = Add two new interfaces that should be used for all new code throughout the ORTE and OMPI layers (we already make the search-and-replace on the existing ORTE / OMPI layers): * orte_output(): (and corresponding friends ORTE_OUTPUT, orte_output_verbose, etc.) This function sends the output directly to the HNP for processing as part of a job-specific output channel. It supports all the same outputs as opal_output() (syslog, file, stdout, stderr), but for stdout/stderr, the output is sent to the HNP for processing and output. More on this below. * orte_show_help(): This function is a drop-in-replacement for opal_show_help(), with two differences in functionality: 1. the rendered text help message output is sent to the HNP for display (rather than outputting directly into the process' stderr stream) 1. the HNP detects duplicate help messages and does not display them (so that you don't see the same error message N times, once from each of your N MPI processes); instead, it counts "new" instances of the help message and displays a message every ~5 seconds when there are new ones ("I got X new copies of the help message...") opal_show_help and opal_output still exist, but they only output in the current process. The intent for the new orte_* functions is that they can apply job-level intelligence to the output. As such, we recommend that all new ORTE and OMPI code use the new orte_* functions, not thei opal_* functions. === New code === For ORTE and OMPI programmers, here's what you need to do differently in new code: * Do not include opal/util/show_help.h or opal/util/output.h. Instead, include orte/util/output.h (this one header file has declarations for both the orte_output() series of functions and orte_show_help()). * Effectively s/opal_output/orte_output/gi throughout your code. Note that orte_output_open() takes a slightly different argument list (as a way to pass data to the filtering stream -- see below), so you if explicitly call opal_output_open(), you'll need to slightly adapt to the new signature of orte_output_open(). * Literally s/opal_show_help/orte_show_help/. The function signature is identical. === Notes === * orte_output'ing to stream 0 will do similar to what opal_output'ing did, so leaving a hard-coded "0" as the first argument is safe. * For systems that do not use ORTE's RML or the HNP, the effect of orte_output_* and orte_show_help will be identical to their opal counterparts (the additional information passed to orte_output_open() will be lost!). Indeed, the orte_* functions simply become trivial wrappers to their opal_* counterparts. Note that we have not tested this; the code is simple but it is quite possible that we mucked something up. = Filter Framework = Messages sent view the new orte_* functions described above and messages output via the IOF on the HNP will now optionally be passed through a new "filter" framework before being output to stdout/stderr. The "filter" OPAL MCA framework is intended to allow preprocessing to messages before they are sent to their final destinations. The first component that was written in the filter framework was to create an XML stream, segregating all the messages into different XML tags, etc. This will allow 3rd party tools to read the stdout/stderr from the HNP and be able to know exactly what each text message is (e.g., a help message, another OMPI infrastructure message, stdout from the user process, stderr from the user process, etc.). Filtering is not active by default. Filter components must be specifically requested, such as: {{{ $ mpirun --mca filter xml ... }}} There can only be one filter component active. = New MCA Parameters = The new functionality described above introduces two new MCA parameters: * '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that help messages will be aggregated, as described above. If set to 0, all help messages will be displayed, even if they are duplicates (i.e., the original behavior). * '''orte_base_show_output_recursions''': An MCA parameter to help debug one of the known issues, described below. It is likely that this MCA parameter will disappear before v1.3 final. = Known Issues = * The XML filter component is not complete. The current output from this component is preliminary and not real XML. A bit more work needs to be done to configure.m4 search for an appropriate XML library/link it in/use it at run time. * There are possible recursion loops in the orte_output() and orte_show_help() functions -- e.g., if RML send calls orte_output() or orte_show_help(). We have some ideas how to fix these, but figured that it was ok to commit before feature freeze with known issues. The code currently contains sub-optimal workarounds so that this will not be a problem, but it would be good to actually solve the problem rather than have hackish workarounds before v1.3 final. This commit was SVN r18434.
2008-05-13 20:00:55 +00:00
orte_show_help("help-plm-rsh.txt", "deadlock-params",
true, mca_plm_rsh_component.num_concurrent, map->num_new_daemons);
ORTE_ERROR_LOG(ORTE_ERR_FATAL);
OBJ_RELEASE(state);
return;
}
/*
* After a discussion between Ralph & Jeff, we concluded that we
* really are handling the prefix dir option incorrectly. It currently
* is associated with an app_context, yet it really refers to the
* location where OpenRTE/Open MPI is installed on a NODE. Fixing
* this right now would involve significant change to orterun as well
* as elsewhere, so we will intentionally leave this incorrect at this
* point. The error, however, is identical to that seen in all prior
* releases of OpenRTE/Open MPI, so our behavior is no worse than before.
*
* A note to fix this, along with ideas on how to do so, has been filed
* on the project's Trac system under "feature enhancement".
*
* For now, default to the prefix_dir provided in the first app_context.
* Since there always MUST be at least one app_context, we are safe in
* doing this.
*/
app = (orte_app_context_t*)opal_pointer_array_get_item(state->jdata->apps, 0);
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 18:56:47 +00:00
if (!orte_get_attribute(&app->attributes, ORTE_APP_PREFIX_DIR, (void**)&prefix_dir, OPAL_STRING)) {
/* check to see if enable-orterun-prefix-by-default was given - if
* this is being done by a singleton, then orterun will not be there
* to put the prefix in the app. So make sure we check to find it */
if ((bool)ORTE_WANT_ORTERUN_PREFIX_BY_DEFAULT) {
prefix_dir = strdup(opal_install_dirs.prefix);
}
}
/* we also need at least one node name so we can check what shell is
* being used, if we have to
*/
node = NULL;
for (nnode = 0; nnode < map->nodes->size; nnode++) {
if (NULL != (nd = (orte_node_t*)opal_pointer_array_get_item(map->nodes, nnode))) {
node = nd;
/* if the node is me, then we continue - we would
* prefer to find some other node so we can tell what the remote
* shell is, if necessary
*/
if (0 != strcmp(node->name, orte_process_info.nodename)) {
break;
}
}
}
/* if we are tree launching, find our children and create the launch cmd */
if (!mca_plm_rsh_component.no_tree_spawn) {
orte_daemon_cmd_flag_t command = ORTE_DAEMON_TREE_SPAWN;
opal_byte_object_t bo, *boptr;
orte_job_t *jdatorted;
/* get the tree spawn buffer */
orte_tree_launch_cmd = OBJ_NEW(opal_buffer_t);
/* insert the tree_spawn cmd */
if (ORTE_SUCCESS != (rc = opal_dss.pack(orte_tree_launch_cmd, &command, 1, ORTE_DAEMON_CMD))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
goto cleanup;
}
/* pack the prefix since this will be needed by the next wave */
if (ORTE_SUCCESS != (rc = opal_dss.pack(orte_tree_launch_cmd, &prefix_dir, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
goto cleanup;
}
/* construct a nodemap of all daemons we know about */
if (ORTE_SUCCESS != (rc = orte_util_encode_nodemap(&bo, false))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
goto cleanup;
}
/* store it */
boptr = &bo;
if (ORTE_SUCCESS != (rc = opal_dss.pack(orte_tree_launch_cmd, &boptr, 1, OPAL_BYTE_OBJECT))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(orte_tree_launch_cmd);
free(bo.bytes);
goto cleanup;
}
/* release the data since it has now been copied into our buffer */
free(bo.bytes);
/* get the orted job data object */
if (NULL == (jdatorted = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
rc = ORTE_ERR_NOT_FOUND;
goto cleanup;
}
/* get the updated routing list */
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 18:56:47 +00:00
OBJ_CONSTRUCT(&coll, opal_list_t);
orte_routed.get_routing_list(&coll);
}
/* setup the launch */
if (ORTE_SUCCESS != (rc = setup_launch(&argc, &argv, node->name, &node_name_index1,
&proc_vpid_index, prefix_dir))) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/*
* Iterate through each of the nodes
*/
for (nnode=0; nnode < map->nodes->size; nnode++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, nnode))) {
continue;
}
/* if we are tree launching, only launch our own children */
if (!mca_plm_rsh_component.no_tree_spawn) {
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 18:56:47 +00:00
OPAL_LIST_FOREACH(child, &coll, orte_namelist_t) {
if (child->name.vpid == node->daemon->name.vpid) {
goto launch;
}
}
/* didn't find it - ignore this node */
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh:launch daemon %s not a child of mine",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_VPID_PRINT(node->daemon->name.vpid)));
continue;
}
launch:
/* if this daemon already exists, don't launch it! */
if (ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_DAEMON_LAUNCHED)) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh:launch daemon already exists on node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
node->name));
continue;
}
/* if the node's daemon has not been defined, then we
* have an error!
*/
if (NULL == node->daemon) {
ORTE_ERROR_LOG(ORTE_ERR_FATAL);
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh:launch daemon failed to be defined on node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
node->name));
continue;
}
/* setup node name */
free(argv[node_name_index1]);
username = NULL;
if (orte_get_attribute(&node->attributes, ORTE_NODE_USERNAME, (void**)&username, OPAL_STRING)) {
(void)asprintf (&argv[node_name_index1], "%s@%s",
username, node->name);
free(username);
} else {
argv[node_name_index1] = strdup(node->name);
}
/* pass the vpid */
rc = orte_util_convert_vpid_to_string(&var, node->daemon->name.vpid);
if (ORTE_SUCCESS != rc) {
opal_output(0, "orte_plm_rsh: unable to get daemon vpid as string");
exit(-1);
}
free(argv[proc_vpid_index]);
argv[proc_vpid_index] = strdup(var);
free(var);
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: adding node %s to launch list",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
node->name));
/* we are in an event, so no need to protect the list */
caddy = OBJ_NEW(orte_plm_rsh_caddy_t);
caddy->argc = argc;
caddy->argv = opal_argv_copy(argv);
caddy->daemon = node->daemon;
OBJ_RETAIN(caddy->daemon);
opal_list_append(&launch_list, &caddy->super);
}
/* set the job state to indicate the daemons are launched */
state->jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
/* trigger the event to start processing the launch list */
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: activating launch event",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
opal_event_active(&launch_event, EV_WRITE, 1);
/* now that we've launched the daemons, let the daemon callback
* function determine they are all alive and trigger the next stage
*/
OBJ_RELEASE(state);
opal_argv_free(argv);
return;
cleanup:
OBJ_RELEASE(state);
As per the RFC, bring in the ORTE async progress code and the rewrite of OOB: *** THIS RFC INCLUDES A MINOR CHANGE TO THE MPI-RTE INTERFACE *** Note: during the course of this work, it was necessary to completely separate the MPI and RTE progress engines. There were multiple places in the MPI layer where ORTE_WAIT_FOR_COMPLETION was being used. A new OMPI_WAIT_FOR_COMPLETION macro was created (defined in ompi/mca/rte/rte.h) that simply cycles across opal_progress until the provided flag becomes false. Places where the MPI layer blocked waiting for RTE to complete an event have been modified to use this macro. *************************************************************************************** I am reissuing this RFC because of the time that has passed since its original release. Since its initial release and review, I have debugged it further to ensure it fully supports tests like loop_spawn. It therefore seems ready for merge back to the trunk. Given its prior review, I have set the timeout for one week. The code is in https://bitbucket.org/rhc/ompi-oob2 WHAT: Rewrite of ORTE OOB WHY: Support asynchronous progress and a host of other features WHEN: Wed, August 21 SYNOPSIS: The current OOB has served us well, but a number of limitations have been identified over the years. Specifically: * it is only progressed when called via opal_progress, which can lead to hangs or recursive calls into libevent (which is not supported by that code) * we've had issues when multiple NICs are available as the code doesn't "shift" messages between transports - thus, all nodes had to be available via the same TCP interface. * the OOB "unloads" incoming opal_buffer_t objects during the transmission, thus preventing use of OBJ_RETAIN in the code when repeatedly sending the same message to multiple recipients * there is no failover mechanism across NICs - if the selected NIC (or its attached switch) fails, we are forced to abort * only one transport (i.e., component) can be "active" The revised OOB resolves these problems: * async progress is used for all application processes, with the progress thread blocking in the event library * each available TCP NIC is supported by its own TCP module. The ability to asynchronously progress each module independently is provided, but not enabled by default (a runtime MCA parameter turns it "on") * multi-address TCP NICs (e.g., a NIC with both an IPv4 and IPv6 address, or with virtual interfaces) are supported - reachability is determined by comparing the contact info for a peer against all addresses within the range covered by the address/mask pairs for the NIC. * a message that arrives on one TCP NIC is automatically shifted to whatever NIC that is connected to the next "hop" if that peer cannot be reached by the incoming NIC. If no TCP module will reach the peer, then the OOB attempts to send the message via all other available components - if none can reach the peer, then an "error" is reported back to the RML, which then calls the errmgr for instructions. * opal_buffer_t now conforms to standard object rules re OBJ_RETAIN as we no longer "unload" the incoming object * NIC failure is reported to the TCP component, which then tries to resend the message across any other available TCP NIC. If that doesn't work, then the message is given back to the OOB base to try using other components. If all that fails, then the error is reported to the RML, which reports to the errmgr for instructions * obviously from the above, multiple OOB components (e.g., TCP and UD) can be active in parallel * the matching code has been moved to the RML (and out of the OOB/TCP component) so it is independent of transport * routing is done by the individual OOB modules (as opposed to the RML). Thus, both routed and non-routed transports can simultaneously be active * all blocking send/recv APIs have been removed. Everything operates asynchronously. KNOWN LIMITATIONS: * although provision is made for component failover as described above, the code for doing so has not been fully implemented yet. At the moment, if all connections for a given peer fail, the errmgr is notified of a "lost connection", which by default results in termination of the job if it was a lifeline * the IPv6 code is present and compiles, but is not complete. Since the current IPv6 support in the OOB doesn't work anyway, I don't consider this a blocker * routing is performed at the individual module level, yet the active routed component is selected on a global basis. We probably should update that to reflect that different transports may need/choose to route in different ways * obviously, not every error path has been tested nor necessarily covered * determining abnormal termination is more challenging than in the old code as we now potentially have multiple ways of connecting to a process. Ideally, we would declare "connection failed" when *all* transports can no longer reach the process, but that requires some additional (possibly complex) code. For now, the code replicates the old behavior only somewhat modified - i.e., if a module sees its connection fail, it checks to see if it is a lifeline. If so, it notifies the errmgr that the lifeline is lost - otherwise, it notifies the errmgr that a non-lifeline connection was lost. * reachability is determined solely on the basis of a shared subnet address/mask - more sophisticated algorithms (e.g., the one used in the tcp btl) are required to handle routing via gateways * the RML needs to assign sequence numbers to each message on a per-peer basis. The receiving RML will then deliver messages in order, thus preventing out-of-order messaging in the case where messages travel across different transports or a message needs to be redirected/resent due to failure of a NIC This commit was SVN r29058.
2013-08-22 16:37:40 +00:00
ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
}
/**
* Terminate the orteds for a given job
*/
static int rsh_terminate_orteds(void)
{
int rc;
if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
static int rsh_finalize(void)
{
int rc;
/* remove launch event */
opal_event_del(&launch_event);
OBJ_DESTRUCT(&launch_list);
/* cleanup any pending recvs */
if (ORTE_SUCCESS != (rc = orte_plm_base_comm_stop())) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
static void set_handler_default(int sig)
{
struct sigaction act;
act.sa_handler = SIG_DFL;
act.sa_flags = 0;
sigemptyset(&act.sa_mask);
sigaction(sig, &act, (struct sigaction *)0);
}
static orte_plm_rsh_shell_t find_shell(char *shell)
{
int i = 0;
char *sh_name = NULL;
if( (NULL == shell) || (strlen(shell) == 1) ) {
/* Malformed shell */
return ORTE_PLM_RSH_SHELL_UNKNOWN;
}
sh_name = rindex(shell, '/');
if( NULL == sh_name ) {
/* Malformed shell */
return ORTE_PLM_RSH_SHELL_UNKNOWN;
}
/* skip the '/' */
++sh_name;
for (i = 0; i < (int)(sizeof (orte_plm_rsh_shell_name) /
sizeof(orte_plm_rsh_shell_name[0])); ++i) {
if (NULL != strstr(sh_name, orte_plm_rsh_shell_name[i])) {
return (orte_plm_rsh_shell_t)i;
}
}
/* We didn't find it */
return ORTE_PLM_RSH_SHELL_UNKNOWN;
}
static int launch_agent_setup(const char *agent, char *path)
{
char *bname;
int i;
/* if no agent was provided, then report not found */
if (NULL == mca_plm_rsh_component.agent && NULL == agent) {
return ORTE_ERR_NOT_FOUND;
}
/* search for the argv */
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:rsh_setup on agent %s path %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == agent) ? mca_plm_rsh_component.agent : agent,
(NULL == path) ? "NULL" : path));
rsh_agent_argv = orte_plm_rsh_search(agent, path);
if (0 == opal_argv_count(rsh_agent_argv)) {
/* nothing was found */
return ORTE_ERR_NOT_FOUND;
}
/* see if we can find the agent in the path */
rsh_agent_path = opal_path_findv(rsh_agent_argv[0], X_OK, environ, path);
if (NULL == rsh_agent_path) {
/* not an error - just report not found */
opal_argv_free(rsh_agent_argv);
return ORTE_ERR_NOT_FOUND;
}
bname = opal_basename(rsh_agent_argv[0]);
if (NULL == bname) {
return ORTE_SUCCESS;
}
if (0 == strcmp(bname, "ssh")) {
/* if xterm option was given, add '-X', ensuring we don't do it twice */
if (NULL != orte_xterm) {
opal_argv_append_unique_nosize(&rsh_agent_argv, "-X", false);
} else if (0 >= opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
/* if debug was not specified, and the user didn't explicitly
* specify X11 forwarding/non-forwarding, add "-x" if it
* isn't already there (check either case)
*/
for (i = 1; NULL != rsh_agent_argv[i]; ++i) {
if (0 == strcasecmp("-x", rsh_agent_argv[i])) {
break;
}
}
if (NULL == rsh_agent_argv[i]) {
opal_argv_append_nosize(&rsh_agent_argv, "-x");
}
}
}
free(bname);
/* the caller can append any additional argv's they desire */
return ORTE_SUCCESS;
}
/**
* Check the Shell variable and system type on the specified node
*/
static int rsh_probe(char *nodename,
orte_plm_rsh_shell_t *shell)
{
char ** argv;
int argc, rc = ORTE_SUCCESS, i;
int fd[2];
pid_t pid;
char outbuf[4096];
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: going to check SHELL variable on node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
nodename));
*shell = ORTE_PLM_RSH_SHELL_UNKNOWN;
if (pipe(fd)) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: pipe failed with errno=%d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
errno));
return ORTE_ERR_IN_ERRNO;
}
if ((pid = fork()) < 0) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: fork failed with errno=%d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
errno));
return ORTE_ERR_IN_ERRNO;
}
else if (pid == 0) { /* child */
if (dup2(fd[1], 1) < 0) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: dup2 failed with errno=%d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
errno));
exit(01);
}
/* Build argv array */
argv = opal_argv_copy(rsh_agent_argv);
argc = opal_argv_count(rsh_agent_argv);
opal_argv_append(&argc, &argv, nodename);
opal_argv_append(&argc, &argv, "echo $SHELL");
execvp(argv[0], argv);
exit(errno);
}
if (close(fd[1])) {
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: close failed with errno=%d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
errno));
return ORTE_ERR_IN_ERRNO;
}
{
ssize_t ret = 1;
char* ptr = outbuf;
size_t outbufsize = sizeof(outbuf);
do {
ret = read (fd[0], ptr, outbufsize-1);
if (ret < 0) {
if (errno == EINTR)
continue;
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: Unable to detect the remote shell (error %s)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
strerror(errno)));
rc = ORTE_ERR_IN_ERRNO;
break;
}
if( outbufsize > 1 ) {
outbufsize -= ret;
ptr += ret;
}
} while( 0 != ret );
*ptr = '\0';
}
close(fd[0]);
if( outbuf[0] != '\0' ) {
char *sh_name = rindex(outbuf, '/');
if( NULL != sh_name ) {
sh_name++; /* skip '/' */
/* Search for the substring of known shell-names */
for (i = 0; i < (int)(sizeof (orte_plm_rsh_shell_name)/
sizeof(orte_plm_rsh_shell_name[0])); i++) {
if ( NULL != strstr(sh_name, orte_plm_rsh_shell_name[i]) ) {
*shell = (orte_plm_rsh_shell_t)i;
break;
}
}
}
}
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: node %s has SHELL: %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
nodename,
(ORTE_PLM_RSH_SHELL_UNKNOWN == *shell) ? "UNHANDLED" : (char*)orte_plm_rsh_shell_name[*shell]));
return rc;
}
static int setup_shell(orte_plm_rsh_shell_t *rshell,
orte_plm_rsh_shell_t *lshell,
char *nodename, int *argc, char ***argv)
{
orte_plm_rsh_shell_t remote_shell, local_shell;
char *param;
int rc;
/* What is our local shell? */
local_shell = ORTE_PLM_RSH_SHELL_UNKNOWN;
#if OPAL_ENABLE_GETPWUID
{
struct passwd *p;
p = getpwuid(getuid());
if( NULL == p ) {
/* This user is unknown to the system. Therefore, there is no reason we
* spawn whatsoever in his name. Give up with a HUGE error message.
*/
orte_show_help( "help-plm-rsh.txt", "unknown-user", true, (int)getuid() );
return ORTE_ERR_FATAL;
}
param = p->pw_shell;
local_shell = find_shell(p->pw_shell);
}
#endif
/* If we didn't find it in getpwuid(), try looking at the $SHELL
environment variable (see https://svn.open-mpi.org/trac/ompi/ticket/1060)
*/
if (ORTE_PLM_RSH_SHELL_UNKNOWN == local_shell &&
NULL != (param = getenv("SHELL"))) {
local_shell = find_shell(param);
}
if (ORTE_PLM_RSH_SHELL_UNKNOWN == local_shell) {
opal_output(0, "WARNING: local probe returned unhandled shell:%s assuming bash\n",
(NULL != param) ? param : "unknown");
local_shell = ORTE_PLM_RSH_SHELL_BASH;
}
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: local shell: %d (%s)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
local_shell, orte_plm_rsh_shell_name[local_shell]));
/* What is our remote shell? */
if (mca_plm_rsh_component.assume_same_shell) {
remote_shell = local_shell;
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: assuming same remote shell as local shell",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
} else {
rc = rsh_probe(nodename, &remote_shell);
if (ORTE_SUCCESS != rc) {
ORTE_ERROR_LOG(rc);
return rc;
}
if (ORTE_PLM_RSH_SHELL_UNKNOWN == remote_shell) {
opal_output(0, "WARNING: rsh probe returned unhandled shell; assuming bash\n");
remote_shell = ORTE_PLM_RSH_SHELL_BASH;
}
}
OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
"%s plm:rsh: remote shell: %d (%s)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
remote_shell, orte_plm_rsh_shell_name[remote_shell]));
/* Do we need to source .profile on the remote side?
- sh: yes (see bash(1))
- ksh: yes (see ksh(1))
- bash: no (see bash(1))
- [t]csh: no (see csh(1) and tcsh(1))
- zsh: no (see http://zsh.sourceforge.net/FAQ/zshfaq03.html#l19)
*/
if (ORTE_PLM_RSH_SHELL_SH == remote_shell ||
ORTE_PLM_RSH_SHELL_KSH == remote_shell) {
int i;
char **tmp;
tmp = opal_argv_split("( test ! -r ./.profile || . ./.profile;", ' ');
if (NULL == tmp) {
return ORTE_ERR_OUT_OF_RESOURCE;
}
for (i = 0; NULL != tmp[i]; ++i) {
opal_argv_append(argc, argv, tmp[i]);
}
opal_argv_free(tmp);
}
/* pass results back */
*rshell = remote_shell;
*lshell = local_shell;
return ORTE_SUCCESS;
}