Merge pull request #2425 from rhc54/topic/stat

Fix debugger attach and cospawn of debugger daemons for the STAT debugger

Author: jsquyres

.gitignore

@@ -341,6 +341,7 @@ orte/mca/sstore/orte_sstore.7
 
 orte/test/mpi/abort
 orte/test/mpi/accept
+orte/test/mpi/attach
 orte/test/mpi/bad_exit
 orte/test/mpi/bcast_loop
 orte/test/mpi/concurrent_spawn

ompi/mca/rte/orte/rte_orte_module.c

@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2012-2013 Los Alamos National Security, LLC.
  *                         All rights reserved.
- * Copyright (c) 2013-2015 Intel, Inc. All rights reserved
+ * Copyright (c) 2013-2016 Intel, Inc.  All rights reserved.
  * Copyright (c) 2012-2014 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
@@ -104,6 +104,8 @@ void ompi_rte_wait_for_debugger(void)
 {
     int debugger;
     orte_rml_recv_cb_t xfer;
+    char *evar;
+    int time;
 
     /* See lengthy comment in orte/tools/orterun/debuggers.c about
        orte_in_parallel_debugger */
@@ -123,6 +125,12 @@ void ompi_rte_wait_for_debugger(void)
      */
     ompi_debugger_setup_dlls();
 
+    if (NULL != (evar = getenv("ORTE_TEST_DEBUGGER_SLEEP"))) {
+        time = strtol(evar, NULL, 10);
+        sleep(time);
+        return;
+    }
+
     if (orte_standalone_operation) {
         /* spin until debugger attaches and releases us */
         while (MPIR_debug_gate == 0) {

orte/mca/oob/base/oob_base_stubs.c

@@ -2,7 +2,7 @@
 /*
  * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2013-2015 Intel, Inc.  All rights reserved.
+ * Copyright (c) 2013-2016 Intel, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -117,9 +117,16 @@ void orte_oob_base_send_nb(int fd, short args, void *cbdata)
                  * this is a local proc we just haven't heard from
                  * yet due to a race condition. Check that situation */
                 if (ORTE_PROC_IS_DAEMON || ORTE_PROC_IS_HNP) {
-                    ORTE_OOB_SEND(msg);
-                    return;
+                    ++msg->retries;
+                    if (msg->retries < orte_rml_base.max_retries) {
+                        ORTE_OOB_SEND(msg);
+                        return;
+                    }
                 }
+                opal_output_verbose(5, orte_oob_base_framework.framework_output,
+                                    "%s CANNOT SEND TO %s: TAG %d",
+                                    ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                                    ORTE_NAME_PRINT(&msg->dst), msg->tag);
                 msg->status = ORTE_ERR_ADDRESSEE_UNKNOWN;
                 ORTE_RML_SEND_COMPLETE(msg);
                 return;
@@ -396,4 +403,3 @@ static void process_uri(char *uri)
     }
     opal_argv_free(uris);
 }
-

orte/mca/plm/base/plm_base_launch_support.c

@@ -13,7 +13,7 @@
  * Copyright (c) 2009      Institut National de Recherche en Informatique
  *                         et Automatique. All rights reserved.
  * Copyright (c) 2011-2012 Los Alamos National Security, LLC.
- * Copyright (c) 2013-2015 Intel, Inc. All rights reserved.
+ * Copyright (c) 2013-2016 Intel, Inc.  All rights reserved.
  * Copyright (c) 2014-2015 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2016      IBM Corporation.  All rights reserved.
@@ -757,9 +757,10 @@ void orte_plm_base_registered(int fd, short args, void *cbdata)
     }
 
  cleanup:
-    /* need to init_after_spawn for debuggers */
-    ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_READY_FOR_DEBUGGERS);
-
+    /* if this wasn't a debugger job, then need to init_after_spawn for debuggers */
+    if (!ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
+        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_READY_FOR_DEBUGGERS);
+    }
     OBJ_RELEASE(caddy);
 }
 

orte/mca/rml/base/base.h

@@ -12,7 +12,7 @@
  *                         All rights reserved.
  * Copyright (c) 2007-2014 Los Alamos National Security, LLC.  All rights
  *                         reserved.
- * Copyright (c) 2014      Intel, Inc. All rights reserved.
+ * Copyright (c) 2014-2016 Intel, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -84,6 +84,7 @@ ORTE_DECLSPEC void orte_rml_base_comm_stop(void);
 typedef struct {
     opal_list_t posted_recvs;
     opal_list_t unmatched_msgs;
+    int max_retries;
 #if OPAL_ENABLE_TIMING
     bool timing;
 #endif
@@ -123,6 +124,7 @@ typedef struct {
     orte_process_name_t origin;
     int status;                  // returned status on send
     orte_rml_tag_t tag;          // targeted tag
+    int retries;                 // #times we have tried to send it
 
     /* user's send callback functions and data */
     union {

orte/mca/rml/base/rml_base_frame.c

@@ -61,6 +61,14 @@ static int orte_rml_base_register(mca_base_register_flag_t flags)
                                    &orte_rml_base_wrapper);
     (void) mca_base_var_register_synonym(var_id, "orte", "rml",NULL,"wrapper", 0);
 
+    orte_rml_base.max_retries = 3;
+    mca_base_var_register("orte", "rml", "base", "max_retries",
+                           "Max #times to retry sending a message",
+                           MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                           OPAL_INFO_LVL_9,
+                           MCA_BASE_VAR_SCOPE_READONLY,
+                           &orte_rml_base.max_retries);
+
 #if OPAL_ENABLE_TIMING
     orte_rml_base.timing = false;
     (void) mca_base_var_register ("orte", "rml", "base", "timing",
@@ -259,6 +267,7 @@ void orte_rml_recv_callback(int status, orte_process_name_t* sender,
 /***   RML CLASS INSTANCES   ***/
 static void send_cons(orte_rml_send_t *ptr)
 {
+    ptr->retries = 0;
     ptr->cbdata = NULL;
     ptr->iov = NULL;
     ptr->buffer = NULL;
@@ -325,4 +334,3 @@ static void prq_des(orte_rml_recv_request_t *ptr)
 OBJ_CLASS_INSTANCE(orte_rml_recv_request_t,
                    opal_object_t,
                    prq_cons, prq_des);
-

orte/mca/routed/radix/routed_radix.c

@@ -6,7 +6,7 @@
  *                         reserved.
  * Copyright (c) 2011-2015 Los Alamos National Security, LLC.  All rights
  *                         reserved.
- * Copyright (c) 2013      Intel, Inc. All rights reserved.
+ * Copyright (c) 2013-2016 Intel, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -372,6 +372,10 @@ static orte_process_name_t get_route(orte_process_name_t *target)
     daemon.jobid = ORTE_PROC_MY_NAME->jobid;
     /* find out what daemon hosts this proc */
     if (ORTE_VPID_INVALID == (daemon.vpid = orte_get_proc_daemon_vpid(target))) {
+        opal_output_verbose(2, orte_routed_base_framework.framework_output,
+                            "%s ATTEMPTING TO SEND TO %s",
+                            ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                            ORTE_NAME_PRINT(target));
         ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
         ret = ORTE_NAME_INVALID;
         goto found;

orte/mca/schizo/base/schizo_base_stubs.c

@@ -1,6 +1,6 @@
 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
- * Copyright (c) 2015      Intel, Inc.  All rights reserved.
+ * Copyright (c) 2015-2016 Intel, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -22,7 +22,6 @@ int orte_schizo_base_parse_cli(char *personality,
     orte_schizo_base_active_module_t *mod;
 
     if (NULL == personality) {
-        opal_output(0, "NULL PERSONALITY");
         return ORTE_ERR_NOT_SUPPORTED;
     }
 
@@ -63,6 +62,11 @@ int orte_schizo_base_setup_fork(orte_job_t *jdata,
     int rc;
     orte_schizo_base_active_module_t *mod;
 
+    /* if no personality was specified, then nothing to do */
+    if (NULL == jdata->personality) {
+        return ORTE_SUCCESS;
+    }
+
     OPAL_LIST_FOREACH(mod, &orte_schizo_base.active_modules, orte_schizo_base_active_module_t) {
         if (0 == strcmp(jdata->personality, mod->component->mca_component_name)) {
             if (NULL != mod->module->setup_fork) {
@@ -81,6 +85,11 @@ int orte_schizo_base_setup_child(orte_job_t *jdata,
     int rc;
     orte_schizo_base_active_module_t *mod;
 
+    /* if no personality was specified, then nothing to do */
+    if (NULL == jdata->personality) {
+        return ORTE_SUCCESS;
+    }
+
     OPAL_LIST_FOREACH(mod, &orte_schizo_base.active_modules, orte_schizo_base_active_module_t) {
         if (0 == strcmp(jdata->personality, mod->component->mca_component_name)) {
             if (NULL != mod->module->setup_child) {

orte/mca/state/base/state_base_fns.c

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2011-2012 Los Alamos National Security, LLC.
- * Copyright (c) 2014      Intel, Inc.  All rights reserved.
+ * Copyright (c) 2014-2016 Intel, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -521,13 +521,13 @@ void orte_state_base_track_procs(int fd, short argc, void *cbdata)
         /* update the proc state */
         ORTE_FLAG_UNSET(pdata, ORTE_PROC_FLAG_ALIVE);
         pdata->state = state;
-	if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_LOCAL)) {
+        if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_LOCAL)) {
             /* Clean up the session directory as if we were the process
              * itself.  This covers the case where the process died abnormally
              * and didn't cleanup its own session directory.
              */
             orte_session_dir_finalize(proc);
-	}
+        }
         /* if we are trying to terminate and our routes are
          * gone, then terminate ourselves IF no local procs
          * remain (might be some from another job)
@@ -550,11 +550,11 @@ void orte_state_base_track_procs(int fd, short argc, void *cbdata)
         }
         /* return the allocated slot for reuse */
         cleanup_node(pdata);
-	/* track job status */
-	jdata->num_terminated++;
-	if (jdata->num_terminated == jdata->num_procs) {
+        /* track job status */
+        jdata->num_terminated++;
+        if (jdata->num_terminated == jdata->num_procs) {
             ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_TERMINATED);
-	}
+        }
     }
 
  cleanup:
@@ -752,10 +752,10 @@ void orte_state_base_check_all_complete(int fd, short args, void *cbdata)
                  * is maintained!
                  */
                 if (1 < j) {
-		    if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
-			/* this was a debugger daemon. notify that a debugger has detached */
-			ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DEBUGGER_DETACH);
-		    }
+                    if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
+                        /* this was a debugger daemon. notify that a debugger has detached */
+                        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DEBUGGER_DETACH);
+                    }
                     opal_pointer_array_set_item(orte_job_data, j, NULL);  /* ensure the array has a NULL */
                     OBJ_RELEASE(jdata);
                 }

orte/runtime/orte_mca_params.c

@@ -13,7 +13,7 @@
  * Copyright (c) 2009-2010 Oracle and/or its affiliates.  All rights reserved.
  * Copyright (c) 2012-2013 Los Alamos National Security, LLC.
  *                         All rights reserved
- * Copyright (c) 2013-2015 Intel, Inc. All rights reserved
+ * Copyright (c) 2013-2016 Intel, Inc.  All rights reserved.
  * Copyright (c) 2014      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
@@ -276,7 +276,7 @@ int orte_register_params(void)
                                   "Test debugger colaunch after debugger attachment",
                                   MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
                                   OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
-                                  &orte_debugger_test_daemon);
+                                  &orte_debugger_test_attach);
 
     orte_debugger_check_rate = 0;
     (void) mca_base_var_register ("orte", "orte", NULL, "debugger_check_rate",

orte/test/mpi/Makefile

@@ -1,4 +1,4 @@
-PROGS = mpi_no_op mpi_barrier hello hello_nodename abort multi_abort simple_spawn concurrent_spawn spawn_multiple mpi_spin delayed_abort loop_spawn loop_child bad_exit pubsub hello_barrier segv accept connect hello_output hello_show_help crisscross read_write ziatest slave reduce-hang ziaprobe ziatest bcast_loop parallel_w8 parallel_w64 parallel_r8 parallel_r64 sio sendrecv_blaster early_abort debugger singleton_client_server intercomm_create spawn_tree init-exit77 mpi_info info_spawn server client paccept pconnect ring hello.sapp binding badcoll
+PROGS = mpi_no_op mpi_barrier hello hello_nodename abort multi_abort simple_spawn concurrent_spawn spawn_multiple mpi_spin delayed_abort loop_spawn loop_child bad_exit pubsub hello_barrier segv accept connect hello_output hello_show_help crisscross read_write ziatest slave reduce-hang ziaprobe ziatest bcast_loop parallel_w8 parallel_w64 parallel_r8 parallel_r64 sio sendrecv_blaster early_abort debugger singleton_client_server intercomm_create spawn_tree init-exit77 mpi_info info_spawn server client paccept pconnect ring hello.sapp binding badcoll attach
 
 all: $(PROGS)
 

orte/test/mpi/attach.c

@@ -0,0 +1,30 @@
+/* -*- C -*-
+ *
+ * $HEADER$
+ *
+ * The most basic of MPI applications
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+int main(int argc, char* argv[])
+{
+    unsigned char fifo_cmd = 1;
+    int fd;
+
+    if (1 > argc) {
+        fprintf(stderr, "usage: attach <full-path-to-debugger-fifo-file>\n");
+        exit(1);
+    }
+
+    fd = open(argv[1], O_WRONLY);
+    write(fd, &fifo_cmd, sizeof(unsigned char));
+    close(fd);
+
+    return 0;
+}