Merge pull request #7441 from rhc54/topic/hack

Create a hack to protect against non-integer jobids
2020-02-21 11:28:51 -08:00 · 2020-02-21 11:28:51 -08:00 · f9643b84b9
--- a/ompi/runtime/ompi_rte.c
+++ b/ompi/runtime/ompi_rte.c
@ -65,6 +65,8 @@ pmix_process_info_t pmix_process_info = {0};
 bool pmix_proc_is_bound = false;
 bool ompi_singleton = false;

+static pmix_proc_t myprocid;
+
 static bool added_transport_keys = false;
 static bool added_num_procs = false;
 static bool added_app_ctx = false;
@ -498,7 +500,7 @@ int ompi_rte_init(int *pargc, char ***pargv)
    int ret;
    char *error = NULL;
    opal_process_name_t pname;
-    pmix_proc_t myproc, rproc;
+    pmix_proc_t rproc;
    int u32, *u32ptr;
    uint16_t u16, *u16ptr;
    char **peers=NULL;
@ -530,8 +532,11 @@ int ompi_rte_init(int *pargc, char ***pargv)
        goto error;
    }

+    /* setup our internal nspace hack */
+    opal_pmix_setup_nspace_tracker();
+
    /* initialize the selected module */
-    if (!PMIx_Initialized() && (PMIX_SUCCESS != (ret = PMIx_Init(&myproc, NULL, 0)))) {
+    if (!PMIx_Initialized() && (PMIX_SUCCESS != (ret = PMIx_Init(&myprocid, NULL, 0)))) {
        /* we cannot run - this could be due to being direct launched
         * without the required PMI support being built, so print
         * out a help message indicating it */
@ -539,8 +544,8 @@ int ompi_rte_init(int *pargc, char ***pargv)
        return OPAL_ERR_SILENT;
    }

-    /* setup the process name fields */
-    OPAL_PMIX_CONVERT_PROCT(rc, &pname, &myproc);
+    /* setup the process name fields - also registers the new nspace */
+    OPAL_PMIX_CONVERT_PROCT(rc, &pname, &myprocid);
    if (OPAL_SUCCESS != rc) {
        return rc;
    }
@ -548,6 +553,7 @@ int ompi_rte_init(int *pargc, char ***pargv)
    OPAL_PROC_MY_NAME.vpid = pname.vpid;
    pmix_process_info.my_name.jobid = OPAL_PROC_MY_NAME.jobid;
    pmix_process_info.my_name.vpid = OPAL_PROC_MY_NAME.vpid;
+
    /* set our hostname */
    OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, PMIX_HOSTNAME, &OPAL_PROC_MY_NAME,
                                   (char**)&ev1, PMIX_STRING);
@ -828,6 +834,10 @@ int ompi_rte_finalize(void)
    free (pmix_process_info.cpuset);
    pmix_process_info.cpuset = NULL;

+    /* cleanup our internal nspace hack */
+    opal_pmix_finalize_nspace_tracker();
+
+
    return OMPI_SUCCESS;
 }

--- a/opal/mca/pmix/base/pmix_base_fns.c
+++ b/opal/mca/pmix/base/pmix_base_fns.c
@ -77,6 +77,83 @@ int opal_pmix_base_exchange(pmix_info_t *indat,
    return opal_pmix_convert_status(rc);
 }

+typedef struct {
+    opal_list_item_t super;
+    pmix_nspace_t nspace;
+    opal_jobid_t jobid;
+} opal_nptr_t;
+static OBJ_CLASS_INSTANCE(opal_nptr_t,
+                          opal_list_item_t,
+                          NULL, NULL);
+
+static opal_list_t localnspaces;
+
+void opal_pmix_setup_nspace_tracker(void)
+{
+    /* check if we were launched by PRRTE */
+    if (NULL != getenv("PRRTE_LAUNCHED")) {
+        opal_process_info.nativelaunch = true;
+    }
+
+    OBJ_CONSTRUCT(&localnspaces, opal_list_t);
+}
+
+void opal_pmix_finalize_nspace_tracker(void)
+{
+    OPAL_LIST_DESTRUCT(&localnspaces);
+}
+
+int opal_pmix_convert_jobid(pmix_nspace_t nspace, opal_jobid_t jobid)
+{
+    opal_nptr_t *nptr;
+
+    /* zero out the nspace */
+    PMIX_LOAD_NSPACE(nspace, NULL);
+
+    if (opal_process_info.nativelaunch) {
+        opal_snprintf_jobid(nspace, PMIX_MAX_NSLEN, jobid);
+        return OPAL_SUCCESS;
+    } else {
+        /* cycle across our list of known jobids */
+        OPAL_LIST_FOREACH(nptr, &localnspaces, opal_nptr_t) {
+            if (jobid == nptr->jobid) {
+                PMIX_LOAD_NSPACE(nspace, nptr->nspace);
+                return OPAL_SUCCESS;
+            }
+        }
+    }
+    return OPAL_ERR_NOT_FOUND;
+}
+
+int opal_pmix_convert_nspace(opal_jobid_t *jobid, pmix_nspace_t nspace)
+{
+    opal_nptr_t *nptr;
+    opal_jobid_t jid;
+
+    /* set a default */
+    *jobid = OPAL_JOBID_INVALID;
+
+    if (opal_process_info.nativelaunch) {
+        return opal_convert_string_to_jobid(jobid, nspace);
+    } else {
+        /* cycle across our list of known jobids */
+        OPAL_LIST_FOREACH(nptr, &localnspaces, opal_nptr_t) {
+            if (PMIX_CHECK_NSPACE(nspace, nptr->nspace)) {
+                *jobid = nptr->jobid;
+                return OPAL_SUCCESS;
+            }
+        }
+        /* if we get here, we don't know this nspace */
+        OPAL_HASH_STR(nspace, jid);
+        *jobid = jid;
+        nptr = OBJ_NEW(opal_nptr_t);
+        nptr->jobid = jid;
+        PMIX_LOAD_NSPACE(nptr->nspace, nspace);
+        opal_list_append(&localnspaces, &nptr->super);
+    }
+    return OPAL_SUCCESS;
+}
+
 pmix_status_t opal_pmix_convert_rc(int rc)
 {
    switch (rc) {
--- a/opal/mca/pmix/pmix-internal.h
+++ b/opal/mca/pmix/pmix-internal.h
@ -1,6 +1,6 @@
 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
- * Copyright (c) 2014-2019 Intel, Inc.  All rights reserved.
+ * Copyright (c) 2014-2020 Intel, Inc.  All rights reserved.
 * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
 *                         reserved.
 * Copyright (c) 2019      Research Organization for Information Science
@ -442,9 +442,13 @@ OPAL_DECLSPEC pmix_proc_state_t opal_pmix_convert_state(int state);
 OPAL_DECLSPEC int opal_pmix_convert_pstate(pmix_proc_state_t);
 OPAL_DECLSPEC pmix_status_t opal_pmix_convert_rc(int rc);
 OPAL_DECLSPEC int opal_pmix_convert_status(pmix_status_t status);
+OPAL_DECLSPEC int opal_pmix_convert_jobid(pmix_nspace_t nspace, opal_jobid_t jobid);
+OPAL_DECLSPEC int opal_pmix_convert_nspace(opal_jobid_t *jobid, pmix_nspace_t nspace);
+OPAL_DECLSPEC void opal_pmix_setup_nspace_tracker(void);
+OPAL_DECLSPEC void opal_pmix_finalize_nspace_tracker(void);

 #define OPAL_PMIX_CONVERT_JOBID(n, j) \
-    (void)opal_snprintf_jobid((n), PMIX_MAX_NSLEN, (j))
+    opal_pmix_convert_jobid((n), (j))

 #define OPAL_PMIX_CONVERT_VPID(r, v)        \
    do {                                    \
@ -454,6 +458,7 @@ OPAL_DECLSPEC int opal_pmix_convert_status(pmix_status_t status);
            (r) = (v);                      \
        }                                   \
    } while(0)
+
 #define OPAL_PMIX_CONVERT_NAME(p, n)                        \
    do {                                                    \
        OPAL_PMIX_CONVERT_JOBID((p)->nspace, (n)->jobid);   \
@ -462,15 +467,17 @@ OPAL_DECLSPEC int opal_pmix_convert_status(pmix_status_t status);


 #define OPAL_PMIX_CONVERT_NSPACE(r, j, n)       \
-    (r) = opal_convert_string_to_jobid((j), (n))
+    (r) = opal_pmix_convert_nspace((j), (n))

-#define OPAL_PMIX_CONVERT_RANK(v, r)        \
-    do {                                    \
-        if (PMIX_RANK_WILDCARD == (r)) {    \
-            (v) = OPAL_VPID_WILDCARD;       \
-        } else {                            \
-            (v) = (r);                      \
-        }                                   \
+#define OPAL_PMIX_CONVERT_RANK(v, r)            \
+    do {                                        \
+        if (PMIX_RANK_WILDCARD == (r)) {        \
+            (v) = OPAL_VPID_WILDCARD;           \
+        } else if (PMIX_RANK_INVALID == (r)) {  \
+            (v) = OPAL_VPID_INVALID;            \
+        } else {                                \
+            (v) = (r);                          \
+        }                                       \
    } while(0)

 #define OPAL_PMIX_CONVERT_PROCT(r, n, p)                            \
--- a/opal/util/proc.c
+++ b/opal/util/proc.c
@ -28,6 +28,7 @@ opal_process_name_t opal_name_wildcard = {OPAL_JOBID_WILDCARD, OPAL_VPID_WILDCAR
 opal_process_name_t opal_name_invalid = {OPAL_JOBID_INVALID, OPAL_VPID_INVALID};

 opal_process_info_t opal_process_info = {
+    .nativelaunch = false,
    .nodename = NULL,
    .top_session_dir = NULL,
    .job_session_dir = NULL,
--- a/opal/util/proc.h
+++ b/opal/util/proc.h
@ -23,7 +23,6 @@
 #include "opal/types.h"
 #include "opal/dss/dss.h"

-
 #if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
 #include <arpa/inet.h>
 #endif
@ -105,6 +104,7 @@ typedef struct {
 OBJ_CLASS_DECLARATION(opal_namelist_t);

 typedef struct opal_process_info_t {
+    bool nativelaunch;                  /**< launched by mpirun */
    char *nodename;                     /**< string name for this node */
    char *top_session_dir;              /**< Top-level session directory */
    char *job_session_dir;              /**< Session directory for job */
--- a/2
+++ b/2
@ -1 +1 @@
-Subproject commit 53296629f9aae70a6cd2586c77306a499e5b573a
+Subproject commit c2e2231cc47c3df0fb3e40c130a9fecd1ca5cacf