diff --git a/NEWS b/NEWS
index edf0d067c75bb0c8b7f35c862decd0671c0c77eb..325cf7dcd02439ddcd381d5399a5f79f2b74d27d 100644
--- a/NEWS
+++ b/NEWS
@@ -62,6 +62,8 @@ documents those changes that are of interest to users and admins.
========================
-- Fix bug in mpi plugin to set the ID correctly
-- Accounting bug causing segv fixed (Andy Riebs, 14oct.jobacct.patch)
+ -- Fix for failed launch of a debugged job (e.g. bad executable name).
+ -- Wiki plugin fix for tracking allocated nodes (Ernest Artiaga, BSC).
* Changes in SLURM 0.6.8
========================
diff --git a/src/common/slurm_errno.c b/src/common/slurm_errno.c
index c2f9c1f8d256195f2aa68d6ab383849489c338cf..1ada2647222f70f1643279475634f9cb503ee285 100644
--- a/src/common/slurm_errno.c
+++ b/src/common/slurm_errno.c
@@ -211,9 +211,9 @@ static slurm_errtab_t slurm_errtab[] = {
{ ESLURMD_CANNOT_SPAWN_IO_THREAD,
"Slurmd could not spawn I/O thread" },
{ ESLURMD_FORK_FAILED,
- "Slurmd could not fork batch job" },
+ "Slurmd could not fork job" },
{ ESLURMD_EXECVE_FAILED,
- "Slurmd could not execve batch job" },
+ "Slurmd could not execve job" },
{ ESLURMD_IO_ERROR,
"Slurmd could not connect IO" },
{ ESLURMD_PROLOG_FAILED,
diff --git a/src/plugins/sched/maui/wiki/wiki_request.cpp b/src/plugins/sched/maui/wiki/wiki_request.cpp
index 7bf14331e104c3abfc37fb7eb00e9ef41e972b7f..bd1cc79ef10463b4d64e13db4cb5238b61a7e108 100644
--- a/src/plugins/sched/maui/wiki/wiki_request.cpp
+++ b/src/plugins/sched/maui/wiki/wiki_request.cpp
@@ -370,6 +370,7 @@ wiki_request_t::postproc_string( char * const field,
char * (*post_processor)( const char * const val );
} post_proc_map [] = {
{ JOB_FIELD_REQ_NODES, wiki_request_t::colonify_commas },
+ { JOB_FIELD_ALLOC_NODES,wiki_request_t::colonify_commas },
{ NULL, NULL }
};
@@ -472,6 +473,7 @@ wiki_getnodes_t::map_enum( char * const field,
{ NODE_STATE_LABEL_IDLE, "Idle" },
{ NODE_STATE_LABEL_DRAINED, "Draining" },
{ NODE_STATE_LABEL_DRAINING, "Draining" },
+ { NODE_STATE_LABEL_ALLOCATED, "Running" },
{ NODE_STATE_LABEL_COMPLETING, "Busy" },
{ NULL, NULL }
diff --git a/src/slurmd/slurmstepd/mgr.c b/src/slurmd/slurmstepd/mgr.c
index a3fe9c416f2d3a21d4dccce7c4933fd9c473eb04..defbe0f4f5354b25ae656944ff2a6c201ec4a047 100644
--- a/src/slurmd/slurmstepd/mgr.c
+++ b/src/slurmd/slurmstepd/mgr.c
@@ -470,11 +470,13 @@ job_manager(slurmd_job_t *job)
if (!job->batch &&
(interconnect_init(job->switch_job, job->uid) < 0)) {
/* error("interconnect_init: %m"); already logged */
+ rc = ESLURM_INTERCONNECT_FAILURE;
goto fail2;
}
if (_fork_all_tasks(job) < 0) {
debug("_fork_all_tasks failed");
+ rc = ESLURMD_EXECVE_FAILED;
goto fail2;
}
@@ -544,6 +546,7 @@ job_manager(slurmd_job_t *job)
static int
_fork_all_tasks(slurmd_job_t *job)
{
+ int rc = SLURM_SUCCESS;
int i;
int *writefds; /* array of write file descriptors */
int *readfds; /* array of read file descriptors */
@@ -659,12 +662,15 @@ _fork_all_tasks(slurmd_job_t *job)
* Prepare process for attach by parallel debugger
* (if specified and able)
*/
+ if (pdebug_trace_process(job, job->task[i]->pid)
+ == SLURM_ERROR)
+ rc = SLURM_ERROR;
pdebug_trace_process(job, job->task[i]->pid);
}
xfree(writefds);
xfree(readfds);
- return SLURM_SUCCESS;
+ return rc;
}
diff --git a/src/slurmd/slurmstepd/pdebug.c b/src/slurmd/slurmstepd/pdebug.c
index 67a1bd43df2afee3f9ed7aabd4618d2fe28aaf11..e84ac80ed98d6f78a2cbdacc9e74f37f40a1897c 100644
--- a/src/slurmd/slurmstepd/pdebug.c
+++ b/src/slurmd/slurmstepd/pdebug.c
@@ -27,8 +27,9 @@
/*
* Prepare task for parallel debugger attach
+ * Returns SLURM_SUCCESS or SLURM_ERROR.
*/
-void
+int
pdebug_trace_process(slurmd_job_t *job, pid_t pid)
{
/* If task to be debugged, wait for it to stop via
@@ -41,15 +42,34 @@ pdebug_trace_process(slurmd_job_t *job, pid_t pid)
if (job->task_flags & TASK_PARALLEL_DEBUG) {
int status;
waitpid(pid, &status, WUNTRACED);
- if ((pid > (pid_t) 0) && (kill(pid, SIGSTOP) < 0))
+ if (!WIFSTOPPED(status)) {
+ debug("pdebug_trace_process WIFSTOPPED false"
+ " for pid %lu", pid);
+ if (WIFEXITED(status)) {
+ debug("Process %lu exited \"normally\""
+ " with return code %d",
+ pid, WEXITSTATUS(status));
+ } else if (WIFSIGNALED(status)) {
+ debug("Process %lu kill by signal %d",
+ pid, WTERMSIG(status));
+ }
+ return SLURM_ERROR;
+ }
+ if ((pid > (pid_t) 0) && (kill(pid, SIGSTOP) < 0)) {
error("kill(%lu): %m", (unsigned long) pid);
+ return SLURM_ERROR;
+ }
+
#ifdef HAVE_AIX
- if (_PTRACE(PT_DETACH, pid, NULL, 0))
+ if (_PTRACE(PT_DETACH, pid, NULL, 0)) {
#else
- if (_PTRACE(PTRACE_DETACH, pid, NULL, 0))
+ if (_PTRACE(PTRACE_DETACH, pid, NULL, 0)) {
#endif
error("ptrace(%lu): %m", (unsigned long) pid);
+ return SLURM_ERROR;
+ }
}
+ return SLURM_SUCCESS;
}
/*
diff --git a/src/slurmd/slurmstepd/pdebug.h b/src/slurmd/slurmstepd/pdebug.h
index f92af015b47570a723b99c58bf56250f91f6056a..a4c62c25178ec3441f98cae45d2de3a231826a92 100644
--- a/src/slurmd/slurmstepd/pdebug.h
+++ b/src/slurmd/slurmstepd/pdebug.h
@@ -36,8 +36,9 @@
void pdebug_stop_current(slurmd_job_t *job);
/*
* Prepare task for parallel debugger attach
+ * Returns SLURM_SUCCESS or SLURM_ERROR.
*/
-void pdebug_trace_process(slurmd_job_t *job, pid_t pid);
+int pdebug_trace_process(slurmd_job_t *job, pid_t pid);
#ifdef HAVE_PTRACE64
# define _PTRACE(r,p,a,d) ptrace64((r),(long long)(p),(long long)(a),(d),NULL)
diff --git a/src/srun/msg.c b/src/srun/msg.c
index af25de6878ca19a5bfb9eedece237cb19cd23378..795fa4fb3b50b9314e50d58d21369aa1f8e94ed6 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -183,6 +183,8 @@ static void _dump_proctable(srun_job_t *job)
for (task_inx=0; task_inx<job->step_layout->tasks[node_inx]; task_inx++) {
taskid = job->step_layout->tids[node_inx][task_inx];
tv = &MPIR_proctable[taskid];
+ if (!tv)
+ break;
info("task:%d, host:%s, pid:%d",
taskid, tv->host_name, tv->pid);
}