Provide backward compatible keys so that the non-PMIx components in the opal/pmix framework don't have to adjust as we continue to work on finalizing the PMIx reference scheme. Activate and utilize the new PMIx show_help capability to provide more meaningful error output when the server cannot start.

Ralph Castain · Ralph Castain · commit be8424b691f8 · 2016-08-13T12:13:04.000-07:00
Add a contrib script to cleanup permissions incorrectly modified due to things like smb mounts

dd
diff --git a/contrib/cleanperms b/contrib/cleanperms
@@ -0,0 +1,11 @@
+#!/usr/bin/bash
+
+find . -type f -name "*.c" -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name Makefile.am -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name "*.h" -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name Makefile.include -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name Makefile -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name "*.m4" -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name "*.ac" -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name "*.txt" -perm /u+x -print -exec chmod -x {} \;
+find . -type f -name "*.l" -perm /u+x -print -exec chmod -x {} \;
diff --git a/opal/mca/pmix/pmix2x/pmix/src/server/Makefile.include b/opal/mca/pmix/pmix2x/pmix/src/server/Makefile.include
@@ -11,6 +11,8 @@
 # $HEADER$
 #
 
+dist_pmixdata_DATA += server/help-pmix-server.txt
+
 headers += \
         server/pmix_server_ops.h
 
diff --git a/opal/mca/pmix/pmix2x/pmix/src/server/help-pmix-server.txt b/opal/mca/pmix/pmix2x/pmix/src/server/help-pmix-server.txt
@@ -0,0 +1,35 @@
+# -*- text -*-
+#
+# Copyright (c) 2016      Intel, Inc. All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+#
+[rnd-path-too-long]
+The PMIx server was unable to setup a rendezvous file due to your
+system's restriction for Unix's socket's path-length.
+
+   Temporary directory: %s
+   Rendezvous filename: %s
+
+Please try to set TMPDIR to something short (like /tmp) or change
+your computer's name to something shorter (see uname -n).
+[listener-failed-start]
+The PMIx server was unable to start its listening thread. This is
+usually due to a conflicting stale named pipe from a prior failed
+job, thus preventing the server from binding to its assigned socket.
+
+  Rendezvous filename: %s
+
+Please remove the stale file and try again.
+[data-store-failed]
+The PMIx server was unable to store the specified key-value:
+
+  Key: %s
+
+The precise reason for the failure was provided in the above
+"error-log" message. This is probably something that should
+be referred to the PMIx developers.
diff --git a/opal/mca/pmix/pmix2x/pmix/src/server/pmix_server.c b/opal/mca/pmix/pmix2x/pmix/src/server/pmix_server.c
@@ -52,6 +52,10 @@
 #include "src/util/error.h"
 #include "src/util/output.h"
 #include "src/util/pmix_environ.h"
+#include "src/util/show_help.h"
+#include "src/mca/base/base.h"
+#include "src/mca/base/pmix_mca_base_var.h"
+#include "src/mca/pinstalldirs/base/base.h"
 #include "src/runtime/pmix_progress_threads.h"
 #include "src/usock/usock.h"
 #include "src/sec/pmix_sec.h"
@@ -121,11 +125,25 @@ static pmix_status_t initialize_server_base(pmix_server_module_t *module)
     char *tdir, *evar;
     char * pmix_pid;
     pmix_listener_t *listener;
+    pmix_status_t ret;
 
     /* initialize the output system */
     if (!pmix_output_init()) {
+        fprintf(stderr, "PMIx server was unable to initialize its output system\n");
         return PMIX_ERR_INIT;
     }
+    /* initialize install dirs code */
+    if (PMIX_SUCCESS != (ret = pmix_mca_base_framework_open(&pmix_pinstalldirs_base_framework, 0))) {
+        fprintf(stderr, "pmix_pinstalldirs_base_open() failed -- process will likely abort (%s:%d, returned %d instead of PMIX_SUCCESS)\n",
+                __FILE__, __LINE__, ret);
+        return ret;
+    }
+
+    if (PMIX_SUCCESS != pmix_show_help_init()) {
+        fprintf(stderr, "PMIx server was unable to initialize its show_help system\n");
+        return PMIX_ERR_INIT;
+    }
+
     /* setup the globals */
     pmix_globals_init();
     memset(&pmix_server_globals, 0, sizeof(pmix_server_globals));
@@ -198,7 +216,9 @@ static pmix_status_t initialize_server_base(pmix_server_module_t *module)
     if (0 > asprintf(&pmix_pid, "%s/pmix-%d", tdir, mypid)) {
         return PMIX_ERR_NOMEM;
     }
+
     if ((strlen(pmix_pid) + 1) > sizeof(listener->address.sun_path)-1) {
+        pmix_show_help("help-pmix-server.txt", "rnd-path-too-long", true, tdir, pmix_pid);
         free(pmix_pid);
         return PMIX_ERR_INVALID_LENGTH;
     }
@@ -352,6 +372,7 @@ PMIX_EXPORT pmix_status_t PMIx_server_init(pmix_server_module_t *module,
                 return PMIX_ERR_NOMEM;
             }
             if ((strlen(pmix_pid) + 1) > sizeof(tl->address.sun_path)-1) {
+                pmix_show_help("help-pmix-server.txt", "rnd-path-too-long", true, tdir, pmix_pid);
                 free(pmix_pid);
                 return PMIX_ERR_INVALID_LENGTH;
             }
@@ -380,6 +401,7 @@ PMIX_EXPORT pmix_status_t PMIx_server_init(pmix_server_module_t *module,
                 return PMIX_ERR_NOMEM;
             }
             if ((strlen(pmix_pid) + 1) > sizeof(tl->address.sun_path)-1) {
+                pmix_show_help("help-pmix-server.txt", "rnd-path-too-long", true, tdir, pmix_pid);
                 free(pmix_pid);
                 return PMIX_ERR_INVALID_LENGTH;
             }
@@ -413,6 +435,7 @@ PMIX_EXPORT pmix_status_t PMIx_server_init(pmix_server_module_t *module,
     }
     if (need_listener) {
         if (PMIX_SUCCESS != pmix_start_listening()) {
+            pmix_show_help("help-pmix-server.txt", "listener-failed-start", true, tl->address.sun_path);
             PMIx_server_finalize();
             return PMIX_ERR_INIT;
         }
@@ -441,6 +464,7 @@ PMIX_EXPORT pmix_status_t PMIx_server_init(pmix_server_module_t *module,
             kv.value = &info[n].value;
             if (PMIX_SUCCESS != (rc = pmix_bfrop.pack(&pmix_server_globals.gdata, &kv, 1, PMIX_KVAL))) {
                 PMIX_ERROR_LOG(rc);
+                pmix_show_help("help-pmix-server.txt", "data-store-failed", true, kv.key);
                 /* protect the incoming data */
                 kv.key = NULL;
                 kv.value = NULL;
diff --git a/orte/mca/ess/pmi/ess_pmi_module.c b/orte/mca/ess/pmi/ess_pmi_module.c
@@ -314,12 +314,13 @@ static int rte_init(void)
         }
         /* retrieve the local peers */
         OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_LOCAL_PEERS,
-                              ORTE_PROC_MY_NAME, &val, OPAL_STRING);
+                              &wildcard_rank, &val, OPAL_STRING);
         if (OPAL_SUCCESS == ret && NULL != val) {
             peers = opal_argv_split(val, ',');
             free(val);
             /* and their cpusets, if available */
-            OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCAL_CPUSETS, ORTE_PROC_MY_NAME, &val, OPAL_STRING);
+            OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCAL_CPUSETS,
+                                           &wildcard_rank, &val, OPAL_STRING);
             if (OPAL_SUCCESS == ret && NULL != val) {
                 cpusets = opal_argv_split(val, ':');
                 free(val);
diff --git a/orte/orted/pmix/pmix_server.c b/orte/orted/pmix/pmix_server.c
@@ -272,10 +272,7 @@ int pmix_server_init(void)
 
     /* setup the local server */
     if (ORTE_SUCCESS != (rc = opal_pmix.server_init(&pmix_server, &info))) {
-        ORTE_ERROR_LOG(rc);
-        /* memory cleanup will occur when finalize is called */
-        orte_show_help("help-orterun.txt", "orterun:pmix-failed", true,
-                       orte_process_info.proc_session_dir);
+        /* pmix will provide a nice show_help output here */
         return rc;
     }
     OPAL_LIST_DESTRUCT(&info);
diff --git a/orte/orted/pmix/pmix_server_register_fns.c b/orte/orted/pmix/pmix_server_register_fns.c
@@ -54,10 +54,10 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
 {
     int rc;
     orte_proc_t *pptr;
-    int i, k, n, nlocalprocs;
+    int i, k, n;
     opal_list_t *info, *pmap;
     opal_value_t *kv;
-    orte_node_t *node, *n2;
+    orte_node_t *node, *mynode;
     opal_vpid_t vpid;
     char **list, **procs, **micro, *tmp, *regex, *cpulist, *peerlist;
     orte_job_t *dmns;
@@ -164,8 +164,8 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
         OPAL_LIST_RELEASE(info);
         return ORTE_ERR_NOT_FOUND;
     }
-    node = pptr->node;
-    if (NULL == node) {
+    mynode = pptr->node;
+    if (NULL == mynode) {
         /* cannot happen */
         ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
         OPAL_LIST_RELEASE(info);
@@ -175,14 +175,14 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
     kv = OBJ_NEW(opal_value_t);
     kv->key = strdup(OPAL_PMIX_NODEID);
     kv->type = OPAL_UINT32;
-    kv->data.uint32 = node->index;
+    kv->data.uint32 = mynode->index;
     opal_list_append(info, &kv->super);
 
     /* pass our node size */
     kv = OBJ_NEW(opal_value_t);
     kv->key = strdup(OPAL_PMIX_NODE_SIZE);
     kv->type = OPAL_UINT32;
-    kv->data.uint32 = node->num_procs;
+    kv->data.uint32 = mynode->num_procs;
     opal_list_append(info, &kv->super);
 
     /* univ size */
@@ -220,43 +220,29 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
     kv->data.uint32 = jdata->total_slots_alloc;
     opal_list_append(info, &kv->super);
 
-    /* identify our local node object within the map,
-     * if we were included */
-    node = NULL;
-    map = (orte_job_map_t*)jdata->map;
-    for (i=0; i < map->nodes->size; i++) {
-        if (NULL == (n2 = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
+    /* register any local clients */
+    vpid = ORTE_VPID_MAX;
+    for (i=0; i < mynode->procs->size; i++) {
+        if (NULL == (pptr = (orte_proc_t*)opal_pointer_array_get_item(mynode->procs, i))) {
             continue;
         }
-        if (n2 == pptr->node) {
-            node = n2;
-            break;
-        }
-    }
-    if (NULL != node) {
-        vpid = ORTE_VPID_MAX;
-        for (i=0; i < node->procs->size; i++) {
-            if (NULL == (pptr = (orte_proc_t*)opal_pointer_array_get_item(node->procs, i))) {
-                continue;
+        if (pptr->name.jobid == jdata->jobid) {
+            if (pptr->name.vpid < vpid) {
+                vpid = pptr->name.vpid;
             }
-            if (pptr->name.jobid == jdata->jobid) {
-                if (pptr->name.vpid < vpid) {
-                    vpid = pptr->name.vpid;
-                }
-                /* go ahead and register this client */
-                if (OPAL_SUCCESS != (rc = opal_pmix.server_register_client(&pptr->name, uid, gid,
-                                                                           (void*)pptr, NULL, NULL))) {
-                    ORTE_ERROR_LOG(rc);
-                }
+            /* go ahead and register this client */
+            if (OPAL_SUCCESS != (rc = opal_pmix.server_register_client(&pptr->name, uid, gid,
+                                                                       (void*)pptr, NULL, NULL))) {
+                ORTE_ERROR_LOG(rc);
             }
         }
-        /* pass the local ldr */
-        kv = OBJ_NEW(opal_value_t);
-        kv->key = strdup(OPAL_PMIX_LOCALLDR);
-        kv->type = OPAL_VPID;
-        kv->data.name.vpid = vpid;
-        opal_list_append(info, &kv->super);
     }
+    /* pass the local ldr */
+    kv = OBJ_NEW(opal_value_t);
+    kv->key = strdup(OPAL_PMIX_LOCALLDR);
+    kv->type = OPAL_VPID;
+    kv->data.name.vpid = vpid;
+    opal_list_append(info, &kv->super);
 
     /* for each proc in this job, create an object that
      * includes the info describing the proc so the recipient has a complete
@@ -276,13 +262,11 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
         cpulist = NULL;
         peerlist = NULL;
         vpid = ORTE_VPID_MAX;
-        nlocalprocs = 0;
         for (i=0; i < node->procs->size; i++) {
             if (NULL == (pptr = (orte_proc_t*)opal_pointer_array_get_item(node->procs, i))) {
                 continue;
             }
             if (pptr->name.jobid == jdata->jobid) {
-                ++nlocalprocs;
                 opal_argv_append_nosize(&list, ORTE_VPID_PRINT(pptr->name.vpid));
                 if (pptr->name.vpid < vpid) {
                     vpid = pptr->name.vpid;
@@ -315,6 +299,26 @@ int orte_pmix_server_register_nspace(orte_job_t *jdata)
             procs = NULL;
         }
 
+        /* if this is me, then pass the peers and cpusets to myself
+         * in order to maintain backward compatibility for the non-pmix
+         * components in OPAL/pmix */
+        if (node == mynode) {
+            /* pass the list of peers */
+            kv = OBJ_NEW(opal_value_t);
+            kv->key = strdup(OPAL_PMIX_LOCAL_PEERS);
+            kv->type = OPAL_STRING;
+            kv->data.string = strdup(peerlist);
+            opal_list_append(info, &kv->super);
+
+            /* pass the list of cpusets */
+            kv = OBJ_NEW(opal_value_t);
+            kv->key = strdup(OPAL_PMIX_LOCAL_CPUSETS);
+            kv->type = OPAL_STRING;
+            kv->data.string = strdup(cpulist);
+            opal_list_append(info, &kv->super);
+
+        }
+
         /* now cycle across each proc on this node, passing all data that
          * varies by proc */
         for (i=0; i < node->procs->size; i++) {

Original file line number	Diff line number	Diff line change
`@@ -11,6 +11,8 @@`
`11`	`11`	`# $HEADER$`
`12`	`12`	`#`
`13`	`13`
	`14`	`+dist_pmixdata_DATA += server/help-pmix-server.txt`
	`15`	`+`
`14`	`16`	`headers += \`
`15`	`17`	`server/pmix_server_ops.h`
`16`	`18`