Checkpoint

This commit was SVN r5814.

src/mca/rds/resfile/rds_resfile.c

@@ -32,7 +32,6 @@ static int orte_rds_resfile_parse_site(char *site, FILE *fp);
 
 static int orte_rds_resfile_parse_resource(orte_rds_cell_desc_t *cell, FILE *fp);
 
-
 static int orte_rds_resfile_parse_resource(orte_rds_cell_desc_t *cell, FILE *fp)
 {
     char *line;
@@ -191,6 +190,12 @@ int orte_rds_resfile_query(void)
     
     OMPI_LOCK(&mca_rds_resfile_component.lock);
 
+    if (orte_rds_resfile_queried) {
+       OMPI_UNLOCK(&mca_rds_resfile_component.lock);
+       return ORTE_SUCCESS;
+    }
+    orte_rds_resfile_queried = true;
+    
     /* get the resource filename */
     fileid = mca_base_param_register_string("rds", "resfile", "name", NULL, NULL);
     mca_base_param_lookup_string(fileid, &resfile);

src/mca/rds/resfile/rds_resfile.h

@@ -73,6 +73,7 @@ typedef struct orte_rds_resfile_component_t orte_rds_resfile_component_t;
 OMPI_COMP_EXPORT extern orte_rds_resfile_component_t mca_rds_resfile_component;
 OMPI_COMP_EXPORT extern orte_rds_base_module_t orte_rds_resfile_module;
 
+extern bool orte_rds_resfile_queried;
 
 #if defined(c_plusplus) || defined(__cplusplus)
 }

src/mca/rds/resfile/rds_resfile_component.c

@@ -66,6 +66,11 @@ orte_rds_base_module_t orte_rds_resfile_module = {
     orte_rds_resfile_finalize
 };
 
+/*
+ * Instantiate component variables
+ */
+bool orte_rds_resfile_queried;
+
 /**
  *  Convience functions to lookup MCA parameter values.
  */
@@ -99,6 +104,7 @@ static int orte_rds_resfile_open(void)
     OBJ_CONSTRUCT(&mca_rds_resfile_component.lock, ompi_mutex_t);
     mca_rds_resfile_component.debug = orte_rds_resfile_param_register_int("debug",1);
     mca_rds_resfile_component.filename = orte_rds_resfile_param_register_string("file", NULL);
+    orte_rds_resfile_queried = false;
     
     return ORTE_SUCCESS;
 }

src/runtime/orte_init_stage1.c

@@ -195,7 +195,7 @@ int orte_init_stage1(void)
                 free(orte_universe_info.name);
                 orte_universe_info.name = NULL;
                 pid = getpid();
-                if (0 > asprintf(&orte_universe_info.name, "%s-%d", universe, pid)) {
+                if (0 > asprintf(&orte_universe_info.name, "%s-%d", universe, (int)pid)) {
                     ompi_output(0, "orte_init: failed to create unique universe name");
                     return ret;
                 }

src/runtime/orte_setup_hnp.c

@@ -44,6 +44,7 @@
 #include "threads/condition.h"
 #include "runtime/orte_wait.h"
 #include "util/argv.h"
+#include "util/ompi_environ.h"
 #include "util/output.h"
 #include "util/path.h"
 #include "util/univ_info.h"
@@ -79,6 +80,10 @@ typedef struct {
  */
 static ompi_mutex_t orte_setup_hnp_mutex;
 static ompi_condition_t orte_setup_hnp_condition;
+/* Local return code */
+static int orte_setup_hnp_rc;
+/* Local uri storage */
+static char *orte_setup_hnp_orted_uri;
 
 static orte_setup_hnp_cb_data_t orte_setup_hnp_cbdata = {NULL, NULL, NULL, 0};
 
@@ -421,6 +426,7 @@ MOVEON:
     }
     
     /* fork a child to exec the rsh/ssh session */
+    orte_setup_hnp_rc = ORTE_SUCCESS;
     pid = fork();
     if (pid < 0) {
         ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
@@ -446,7 +452,44 @@ MOVEON:
         OMPI_THREAD_LOCK(&orte_setup_hnp_mutex);
         ompi_condition_timedwait(&orte_setup_hnp_condition, &orte_setup_hnp_mutex, &ts);
         OMPI_THREAD_UNLOCK(&orte_setup_hnp_mutex);
-        return ORTE_SUCCESS;
+
+        if (ORTE_SUCCESS == orte_setup_hnp_rc) {
+            /* need to restart the local system so it can connect to the remote daemon.
+             * we only want to clear the run-time itself - we cannot close the OPAL
+             * utilities, though, or we will lose all of our MCA parameters
+             */
+            orte_system_finalize();
+            /*
+             * now set the relevant MCA parameters to point us at the remote daemon...
+             */
+            if (ORTE_SUCCESS != (rc = ompi_setenv("OMPI_MCA_gpr_replica_uri",
+                        orte_setup_hnp_orted_uri, true, &environ))) {
+                fprintf(stderr, "orte_setup_hnp: could not set gpr_replica_uri in environ\n");
+                return rc;
+            }
+            
+            if (ORTE_SUCCESS != (rc = ompi_setenv("OMPI_MCA_ns_replica_uri",
+                        orte_setup_hnp_orted_uri, true, &environ))) {
+                fprintf(stderr, "orte_setup_hnp: could not set ns_replica_uri in environ\n");
+                return rc;
+            }
+
+            ompi_unsetenv("OMPI_MCA_seed", &environ);
+            
+            /*
+             * ...re-init ourselves...
+             */
+            if (ORTE_SUCCESS != (rc = orte_system_init())) {
+                ORTE_ERROR_LOG(rc);
+                return rc;
+            }
+            /*
+             * ...and we are now ready to go!
+             */
+            return ORTE_SUCCESS;
+        }
+        
+        return orte_setup_hnp_rc;
     }
     
 CLEANUP:
@@ -463,15 +506,19 @@ static void orte_setup_hnp_recv(int status, orte_process_name_t* sender,
                                 orte_buffer_t* buffer, orte_rml_tag_t tag,
                                 void* cbdata)
 {
-    char *orted_uri;
     size_t n=1;
     int rc;
     
     OMPI_THREAD_LOCK(&orte_setup_hnp_mutex);
-    if (ORTE_SUCCESS != (rc = orte_dps.unpack(buffer, &orted_uri, &n, ORTE_STRING))) {
+    if (ORTE_SUCCESS != (rc = orte_dps.unpack(buffer, &orte_setup_hnp_orted_uri, &n, ORTE_STRING))) {
         ORTE_ERROR_LOG(rc);
+        orte_setup_hnp_rc = rc;
+        ompi_condition_signal(&orte_setup_hnp_condition);
+        OMPI_THREAD_UNLOCK(&orte_setup_hnp_mutex);
+        return;
     }
-ompi_output(0, "orteprobe: received uri %s", orted_uri);
+ompi_output(0, "orteprobe: received uri %s", orte_setup_hnp_orted_uri);
+    orte_setup_hnp_rc = ORTE_SUCCESS;
     ompi_condition_signal(&orte_setup_hnp_condition);
     OMPI_THREAD_UNLOCK(&orte_setup_hnp_mutex);
 }

test/util/Makefile.am

@@ -23,9 +23,10 @@ check_PROGRAMS = \
 	orte_os_path \
 	orte_sys_info \
 	orte_os_create_dirpath \
-        orte_session_dir \
+    orte_session_dir \
 	ompi_argv \
-        ompi_basename
+    ompi_basename \
+    orte_universe_setup_file_io
 
 TESTS = \
 	$(check_PROGRAMS)
@@ -79,3 +80,9 @@ orte_session_dir_LDADD = \
         $(top_builddir)/test/support/libsupport.a 
 orte_session_dir_DEPENDENCIES = $(orte_session_dir_LDADD)
 
+orte_universe_setup_file_io_SOURCES = orte_universe_setup_file_io.c
+orte_universe_setup_file_io_LDADD = \
+        $(top_builddir)/src/libmpi.la \
+        $(top_builddir)/test/support/libsupport.a 
+orte_universe_setup_file_io_DEPENDENCIES = $(orte_universe_setup_file_io_LDADD)
+

test/util/orte_universe_setup_file_io.c

@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 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.
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#include "orte_config.h"
+#include "include/orte_constants.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#ifdef HAVE_SYS_PARAM_H
+#include <sys/param.h>
+#endif
+
+#include "util/univ_info.h"
+#include "util/universe_setup_file_io.h"
+#include "support.h"
+
+static bool test1(void);   /* verify it returns info */
+static bool test2(void);   /* test second time through */
+
+orte_universe_t orte_universe_info;
+
+int main(int argc, char* argv[])
+{
+    test_init("universe_setup_file_io");
+
+    /* initialize the univ_info structure */
+    orte_universe_info.name = strdup("default-universe");
+    orte_universe_info.host = strdup("test-host.domain.org");
+    orte_universe_info.uid = strdup("this-is-me");
+    orte_universe_info.persistence = true;
+    orte_universe_info.scope = strdup("private");
+    orte_universe_info.console = false;
+    orte_universe_info.seed_uri = strdup("0.0.0;tcp://128.165.148.81:52424");
+    orte_universe_info.console_connected = false;
+    orte_universe_info.scriptfile = NULL;
+    
+    if (test1()) {
+        test_success();
+    }
+    else {
+      test_failure("universe_setup_file_io write failed");
+    }
+
+    if (test2()) {
+        test_success();
+    }
+    else {
+      test_failure("universe_setup_file_io read failed");
+    }
+
+    test_finalize();
+    return 0;
+}
+
+
+static bool test1(void)
+{
+    int rc;
+    
+    /* Test write */
+
+    if (ORTE_SUCCESS != (rc = orte_write_universe_setup_file("test-file", &orte_universe_info))) {
+        fprintf(stderr, "universe_setup_file_io: failed write with code %d\n", rc);
+        return false;
+    }
+    
+    return true;
+}
+
+
+static bool test2(void)
+{
+    int rc;
+    orte_universe_t univ;
+    
+    /* Test read */
+
+    if (ORTE_SUCCESS != (rc = orte_read_universe_setup_file("test-file", &univ))) {
+        fprintf(stderr, "universe_setup_file_io: failed read with code %d\n", rc);
+        return false;
+    }
+    
+    if (0 != strcmp(orte_universe_info.name, univ.name) ||
+        0 != strcmp(orte_universe_info.host, univ.host) ||
+        0 != strcmp(orte_universe_info.uid, univ.uid) ||
+        orte_universe_info.persistence != univ.persistence ||
+        0 != strcmp(orte_universe_info.scope, univ.scope) ||
+        orte_universe_info.console != univ.console ||
+        0 != strcmp(orte_universe_info.seed_uri, univ.seed_uri) ||
+        orte_universe_info.console_connected != univ.console_connected ||
+        orte_universe_info.scriptfile != univ.scriptfile) {
+        fprintf(stderr, "universe_setup_file_io: read mismatch\n");
+        return false;
+    }
+    
+    return true;
+}