From a64e499e2d2bf6eb0e519af5c36139225b49e18d Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Thu, 2 Dec 2010 19:50:25 +0000
Subject: [PATCH] In srun, log if a job allocation is revoked while waiting for
resources.
---
src/srun/allocate.c | 6 +++++-
testsuite/expect/test1.64 | 2 +-
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 49af546e3e1..4925466738e 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -126,6 +126,8 @@ static void _signal_while_allocating(int signo)
destroy_job = 1;
if (pending_job_id != 0) {
slurm_complete_job(pending_job_id, NO_VAL);
+ info("Job allocation %u has been revoked.", pending_job_id);
+
}
}
@@ -438,6 +440,8 @@ allocate_nodes(void)
/* create message thread to handle pings and such from slurmctld */
msg_thr = slurm_allocation_msg_thr_create(&j->other_port, &callbacks);
+ /* NOTE: Do not process signals in separate pthread. The signal will
+ * cause slurm_allocate_resources_blocking() to exit immediately. */
xsignal_unblock(sig_array);
for (i = 0; sig_array[i]; i++)
xsignal(sig_array[i], _signal_while_allocating);
@@ -448,7 +452,7 @@ allocate_nodes(void)
if (destroy_job) {
/* cancelled by signal */
break;
- } else if(!resp && !_retry()) {
+ } else if (!resp && !_retry()) {
break;
}
}
diff --git a/testsuite/expect/test1.64 b/testsuite/expect/test1.64
index 86e0d6bb311..e49a71e73e4 100755
--- a/testsuite/expect/test1.64
+++ b/testsuite/expect/test1.64
@@ -56,7 +56,7 @@ expect {
send_user "\nSent SIGINT\n"
exp_continue
}
- -re "Terminated job" {
+ -re "Job allocation ($number) has been revoked" {
incr matches
exp_continue
}
--
GitLab