PMIx_Connect usage: add optional timeout

Add an MCA parameter that can be used to set a timeot on the PMIx_Connect operation used to support MPI_Comm_accept/connect and relatives. Related to open-mpi#8958 Signed-off-by: Howard Pritchard <[email protected]>
hppritcha · May 13, 2021 · 038291a · 038291a
1 parent 0e76855
commit 038291a
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 2 deletions.
diff --git a/ompi/dpm/dpm.c b/ompi/dpm/dpm.c
@@ -21,6 +21,8 @@
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2018      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2021      Nanook Consulting.  All rights reserved.
+ * Copyright (c) 2021      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -104,7 +106,7 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
     bool dense, isnew;
     opal_process_name_t pname;
     opal_list_t ilist, mlist, rlist;
-    pmix_info_t info;
+    pmix_info_t info, tinfo;
     pmix_value_t pval;
     pmix_pdata_t pdat;
     pmix_proc_t *procs, pxproc;
@@ -373,7 +375,10 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
     /* tell the host RTE to connect us - this will download
      * all known data for the nspace's of participating procs
      * so that add_procs will not result in a slew of lookups */
-    pret = PMIx_Connect(procs, nprocs, NULL, 0);
+    PMIX_INFO_CONSTRUCT(&tinfo);
+    PMIX_INFO_LOAD(&tinfo, PMIX_TIMEOUT, &ompi_pmix_connect_timeout, PMIX_UINT32);
+    pret = PMIx_Connect(procs, nprocs, &tinfo, 1);
+    PMIX_INFO_DESTRUCT(&tinfo);
     PMIX_PROC_FREE(procs, nprocs);
     rc = opal_pmix_convert_status(pret);
     if (OPAL_SUCCESS != rc) {

diff --git a/ompi/runtime/ompi_mpi_params.c b/ompi/runtime/ompi_mpi_params.c
@@ -20,6 +20,8 @@
  *                         All rights reserved.
  * Copyright (c) 2016-2019 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2021      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -83,6 +85,7 @@ bool ompi_mpi_compat_mpi3 = false;
 
 char *ompi_mpi_spc_attach_string = NULL;
 bool ompi_mpi_spc_dump_enabled = false;
+uint32_t ompi_pmix_connect_timeout;
 
 static bool show_default_mca_params = false;
 static bool show_file_mca_params = false;
@@ -391,6 +394,13 @@ int ompi_mpi_register_params(void)
                                  &ompi_mpi_spc_dump_enabled);
 #endif // SPC_ENABLE
 
+    ompi_pmix_connect_timeout = 0; /* infinite timeout - see PMIx standard */
+    (void) mca_base_var_register ("ompi", "mpi", NULL, "pmix_connect_timeout",
+                                  "Timeout(secs) for calls to PMIx_Connect. Default is no timeout.",
+                                  MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL,
+                                  0, 0, OPAL_INFO_LVL_3, MCA_BASE_VAR_SCOPE_LOCAL,
+                                  &ompi_pmix_connect_timeout);
+
     return OMPI_SUCCESS;
 }
 

diff --git a/ompi/runtime/params.h b/ompi/runtime/params.h
@@ -16,6 +16,8 @@
  * Copyright (c) 2010-2012 Oak Ridge National Labs.  All rights reserved.
  * Copyright (c) 2013      NVIDIA Corporation.  All rights reserved.
  * Copyright (c) 2013      Intel, Inc. All rights reserved
+ * Copyright (c) 2021      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -172,6 +174,10 @@ OMPI_DECLSPEC extern char * ompi_mpi_spc_attach_string;
  */
 OMPI_DECLSPEC extern bool ompi_mpi_spc_dump_enabled;
 
+/**
+ * Timeout for calls to PMIx_Connect(defaut 0, no timeout)
+ */
+OMPI_DECLSPEC extern uint32_t ompi_pmix_connect_timeout;
 
 /**
  * Register MCA parameters used by the MPI layer.