Merge pull request #96 from rhc54/topic/dbgr

Restore debugger launch of apps and then daemons
openpmix · Sep 26, 2018 · 4301061 · 4301061
2 parents f0c1262 + cb3926a
commit 4301061
Show file tree

Hide file tree

Showing 8 changed files with 54 additions and 21 deletions.
diff --git a/config/opal_check_pmi.m4 b/config/opal_check_pmi.m4
@@ -176,6 +176,13 @@ AC_DEFUN([OPAL_CHECK_PMIX],[
            LDFLAGS="-L$pmix_ext_install_libdir $LDFLAGS"])
 
     LIBS="$LIBS -lpmix"
+
+    CPPFLAGS="$CPPFLAGS -I$pmix_ext_install_dir/include"
+    OPAL_WRAPPER_FLAGS_ADD([CPPFLAGS], [-I$pmix_ext_install_dir/include])
+
+    LDFLAGS="$LDFLAGS -L$pmix_ext_install_libdir"
+    OPAL_WRAPPER_FLAGS_ADD([LDFLAGS], [-L$pmix_ext_install_libdir])
+
     opal_external_pmix_happy=yes
 
     AC_DEFINE_UNQUOTED([OPAL_PMIX_VERSION], [$opal_external_pmix_version_found],

diff --git a/examples/debugger.c b/examples/debugger.c
@@ -121,6 +121,7 @@ static void cbfunc(pmix_status_t status,
         PMIX_INFO_CREATE(mq->info, ninfo);
         mq->ninfo = ninfo;
         for (n=0; n < ninfo; n++) {
+            fprintf(stderr, "Key %s Type %s(%d)\n", info[n].key, PMIx_Data_type_string(info[n].value.type), info[n].value.type);
             PMIX_INFO_XFER(&mq->info[n], &info[n]);
         }
     }
@@ -272,7 +273,6 @@ static pmix_status_t spawn_debugger(char *appspace, myrel_t *myrel)
         PMIX_APP_FREE(debugger, 1);
         return rc;
     }
-    fprintf(stderr, "SPAWNED DEBUGGERD\n");
     /* cleanup */
     PMIX_INFO_FREE(dinfo, dninfo);
     PMIX_APP_FREE(debugger, 1);
@@ -438,7 +438,7 @@ int main(int argc, char **argv)
         app[0].cwd = strdup(cwd);
         app[0].maxprocs = 1;
         /* provide job-level directives so the apps do what the user requested */
-        ninfo = 5;
+        ninfo = 6;
         PMIX_INFO_CREATE(info, ninfo);
         PMIX_INFO_LOAD(&info[0], PMIX_MAPBY, "slot", PMIX_STRING);  // map by slot
         asprintf(&tmp, "%s:%d", myproc.nspace, myproc.rank);
@@ -450,6 +450,7 @@ int main(int argc, char **argv)
         PMIX_INFO_LOAD(&info[2], PMIX_FWD_STDOUT, &cospawn, PMIX_BOOL);  // forward stdout to me
         PMIX_INFO_LOAD(&info[3], PMIX_FWD_STDERR, &cospawn, PMIX_BOOL);  // forward stderr to me
         PMIX_INFO_LOAD(&info[4], PMIX_NOTIFY_COMPLETION, NULL, PMIX_BOOL); // notify us when the job completes
+        PMIX_INFO_LOAD(&info[5], PMIX_LAUNCHER_RENDEZVOUS_FILE, "dbgr.rndz.txt", PMIX_STRING);  // have it output a specific rndz file
 
         /* spawn the job - the function will return when the launcher
          * has been launched. Note that this doesn't tell us anything
@@ -463,7 +464,6 @@ int main(int argc, char **argv)
             fprintf(stderr, "Application failed to launch with error: %s(%d)\n", PMIx_Error_string(rc), rc);
             goto done;
         }
-        fprintf(stderr, "Spawn complete\n\n");
 
         /* wait here for the launcher to declare itself ready */
         DEBUG_WAIT_THREAD(&launcher_ready.lock);
@@ -481,7 +481,7 @@ int main(int argc, char **argv)
             fprintf(stderr, "Failed to connect to %s server: %s(%d)\n", argv[1], PMIx_Error_string(rc), rc);
             goto done;
         }
-        fprintf(stderr, "Connection completed\n");
+
         /* send the launch directives */
         ninfo = 3;
         PMIX_INFO_CREATE(info, ninfo);
@@ -527,7 +527,6 @@ int main(int argc, char **argv)
         myquery_data.info = NULL;
         myquery_data.ninfo = 0;
         /* execute the query */
-        fprintf(stderr, "Debugger: querying capabilities\n");
         if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) {
             fprintf(stderr, "PMIx_Query_info failed: %d\n", rc);
             goto done;
@@ -574,7 +573,7 @@ int main(int argc, char **argv)
                 }
             }
         }
-
+        cospawn = false;
         /* if cospawn is true, then we can launch both the app and the debugger
          * daemons at the same time */
         if (cospawn) {
@@ -630,22 +629,24 @@ int main(int argc, char **argv)
             /* wait to get a response */
             DEBUG_WAIT_THREAD(&myquery_data.lock);
             DEBUG_DESTRUCT_LOCK(&myquery_data.lock);
-
             /* we should have gotten a response */
             if (PMIX_SUCCESS != myquery_data.status) {
                 fprintf(stderr, "Debugger[%s:%d] Proctable query failed: %s\n",
                         myproc.nspace, myproc.rank, PMIx_Error_string(myquery_data.status));
                 goto done;
             }
-            /* there should hvae been data */
+            /* there should have been data */
             if (NULL == myquery_data.info || 0 == myquery_data.ninfo) {
                 fprintf(stderr, "Debugger[%s:%d] Proctable query return no results\n",
                         myproc.nspace, myproc.rank);
                 goto done;
             }
             /* the query should have returned a data_array */
             if (PMIX_DATA_ARRAY != myquery_data.info[0].value.type) {
-                fprintf(stderr, "Debugger[%s:%d] Query returned incorrect data type: %s\n", PMIx_Data_type_string(myquery_data.info[0].value.type));
+                fprintf(stderr, "Debugger[%s:%d] Query returned incorrect data type: %s(%d)\n",
+                        myproc.nspace, myproc.rank,
+                        PMIx_Data_type_string(myquery_data.info[0].value.type),
+                        (int)myquery_data.info[0].value.type);
                 return -1;
             }
             if (NULL == myquery_data.info[0].value.data.darray->array) {
@@ -664,7 +665,6 @@ int main(int argc, char **argv)
              *     int exit_code;
              *     pmix_proc_state_t state;
              */
-            fprintf(stderr, "Received %d array elements\n", (int)myquery_data.info[0].value.data.darray->size);
 
             /* now launch the debugger daemons */
             if (PMIX_SUCCESS != (rc = spawn_debugger(clientspace, &dbrel))) {

diff --git a/examples/debuggerd.c b/examples/debuggerd.c
@@ -273,9 +273,13 @@ int main(int argc, char **argv)
     DEBUG_DESTRUCT_LOCK(&mylock);
 
     /* get the nspace of the job we are to debug - it will be in our JOB info */
+#ifdef PMIX_LOAD_PROCID
+    PMIX_LOAD_PROCID(&proc, myproc.nspace, PMIX_RANK_WILDCARD);
+#else
     PMIX_PROC_CONSTRUCT(&proc);
     (void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_KEYLEN);
     proc.rank = PMIX_RANK_WILDCARD;
+#endif
     if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) {
         fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - error %s\n",
                 myproc.nspace, myproc.rank,
@@ -384,7 +388,7 @@ int main(int argc, char **argv)
     PMIX_INFO_LOAD(&info[1], PMIX_EVENT_NON_DEFAULT, NULL, PMIX_BOOL);  // deliver to the target nspace
     fprintf(stderr, "[%s:%u:%lu] Sending release\n", myproc.nspace, myproc.rank, (unsigned long)pid);
     rc = PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE,
-                           NULL, PMIX_RANGE_LOCAL,
+                           NULL, PMIX_RANGE_CUSTOM,
                            info, ninfo, NULL, NULL);
     if (PMIX_SUCCESS != rc) {
         fprintf(stderr, "%s[%s:%u:%lu] Sending release failed with error %s(%d)\n",

diff --git a/orte/mca/rmaps/base/base.h b/orte/mca/rmaps/base/base.h
@@ -123,7 +123,8 @@ ORTE_DECLSPEC int orte_rmaps_base_filter_nodes(orte_app_context_t *app,
                                                opal_list_t *nodes,
                                                bool remove);
 
-ORTE_DECLSPEC int orte_rmaps_base_set_mapping_policy(orte_mapping_policy_t *policy,
+ORTE_DECLSPEC int orte_rmaps_base_set_mapping_policy(orte_job_t *jdata,
+                                                     orte_mapping_policy_t *policy,
                                                      char **device, char *spec);
 ORTE_DECLSPEC int orte_rmaps_base_set_ranking_policy(orte_ranking_policy_t *policy,
                                                      orte_mapping_policy_t mapping,

diff --git a/orte/mca/rmaps/base/rmaps_base_frame.c b/orte/mca/rmaps/base/rmaps_base_frame.c
@@ -296,7 +296,7 @@ static int orte_rmaps_base_open(mca_base_open_flag_t flags)
                        "rmaps_base_cpus_per_proc", "rmaps_base_mapping_policy=<obj>:PE=N, default <obj>=NUMA");
     }
 
-    if (ORTE_SUCCESS != (rc = orte_rmaps_base_set_mapping_policy(&orte_rmaps_base.mapping,
+    if (ORTE_SUCCESS != (rc = orte_rmaps_base_set_mapping_policy(NULL, &orte_rmaps_base.mapping,
                                                                  &orte_rmaps_base.device,
                                                                  rmaps_base_mapping_policy))) {
         return rc;
@@ -593,7 +593,8 @@ static int check_modifiers(char *ck, orte_mapping_policy_t *tmp)
     return ORTE_ERR_TAKE_NEXT_OPTION;
 }
 
-int orte_rmaps_base_set_mapping_policy(orte_mapping_policy_t *policy,
+int orte_rmaps_base_set_mapping_policy(orte_job_t *jdata,
+                                       orte_mapping_policy_t *policy,
                                        char **device, char *inspec)
 {
     char *ck;
@@ -681,7 +682,11 @@ int orte_rmaps_base_set_mapping_policy(orte_mapping_policy_t *policy,
                     }
                 }
                 /* now save the pattern */
-                orte_rmaps_base.ppr = strdup(ck);
+                if (NULL == jdata || NULL == jdata->map) {
+                    orte_rmaps_base.ppr = strdup(ck);
+                } else {
+                    jdata->map->ppr = strdup(ck);
+                }
                 ORTE_SET_MAPPING_POLICY(tmp, ORTE_MAPPING_PPR);
                 ORTE_SET_MAPPING_DIRECTIVE(tmp, ORTE_MAPPING_GIVEN);
                 free(spec);
@@ -747,7 +752,11 @@ int orte_rmaps_base_set_mapping_policy(orte_mapping_policy_t *policy,
     }
 
  setpolicy:
-    *policy = tmp;
+    if (NULL == jdata || NULL == jdata->map) {
+        *policy = tmp;
+    } else {
+        jdata->map->mapping = tmp;
+    }
 
     return ORTE_SUCCESS;
 }

diff --git a/orte/orted/pmix/pmix_server_dyn.c b/orte/orted/pmix/pmix_server_dyn.c
@@ -340,7 +340,7 @@ static void interim(int sd, short args, void *cbdata)
                 rc = ORTE_ERR_BAD_PARAM;
                 goto complete;
             }
-            rc = orte_rmaps_base_set_mapping_policy(&jdata->map->mapping,
+            rc = orte_rmaps_base_set_mapping_policy(jdata, &jdata->map->mapping,
                                                     NULL, info->value.data.string);
             if (ORTE_SUCCESS != rc) {
                 goto complete;

diff --git a/orte/orted/pmix/pmix_server_gen.c b/orte/orted/pmix/pmix_server_gen.c
@@ -565,7 +565,7 @@ static void _query(int sd, short args, void *cbdata)
     pmix_data_array_t *darray;
     pmix_proc_info_t *procinfo;
     pmix_info_t *info;
-    pmix_status_t ret;
+    pmix_status_t ret = PMIX_SUCCESS;
     opal_ds_info_t *kv;
     orte_jobid_t jobid;
     orte_job_t *jdata;
@@ -785,14 +785,14 @@ static void _query(int sd, short args, void *cbdata)
                     }
                 }
                 if (ORTE_JOBID_INVALID == jobid) {
-                    rc = ORTE_ERR_BAD_PARAM;
+                    ret = PMIX_ERR_NOT_FOUND;
                     goto done;
                 }
                 /* construct a list of values with opal_proc_info_t
                  * entries for each proc in the indicated job */
                 jdata = orte_get_job_data_object(jobid);
                 if (NULL == jdata) {
-                    rc = ORTE_ERR_NOT_FOUND;
+                    ret = PMIX_ERR_NOT_FOUND;
                     goto done;
                 }
                 /* setup the reply */
@@ -897,9 +897,13 @@ static void _query(int sd, short args, void *cbdata)
             /* convert the list of results to an info array */
             rcd->ninfo = opal_list_get_size(results);
             PMIX_INFO_CREATE(rcd->info, rcd->ninfo);
+            n=0;
+            OPAL_LIST_FOREACH(kv, results, opal_ds_info_t) {
+                PMIX_INFO_XFER(&rcd->info[n], kv->info);
+                n++;
+            }
         }
     }
-
     cd->infocbfunc(ret, rcd->info, rcd->ninfo, cd->cbdata, qrel, rcd);
 }
 

diff --git a/orte/tools/prun/prun.c b/orte/tools/prun/prun.c
@@ -649,6 +649,14 @@ int prun(int argc, char *argv[])
         opal_list_append(&tinfo, &ds->super);
     }
 
+    /* check for request to drop a rendezvous file */
+    if (NULL != (param = getenv("PMIX_LAUNCHER_RENDEZVOUS_FILE"))) {
+        ds = OBJ_NEW(opal_ds_info_t);
+        PMIX_INFO_CREATE(ds->info, 1);
+        PMIX_INFO_LOAD(ds->info, PMIX_LAUNCHER_RENDEZVOUS_FILE, param, PMIX_STRING);
+        opal_list_append(&tinfo, &ds->super);
+    }
+
     /* convert to array of info */
     ninfo = opal_list_get_size(&tinfo);
     PMIX_INFO_CREATE(iptr, ninfo);