From f648c82c35bbfccfeab38fec34b140bcf41a9817 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Thu, 2 Apr 2009 20:24:05 +0000
Subject: [PATCH] svn merge -r17122:17124
https://eris.llnl.gov/svn/slurm/branches/slurm-1.3
---
.../select/bluegene/plugin/bg_block_info.c | 2 ++
src/plugins/select/bluegene/plugin/bg_job_run.c | 16 +++++++++++++---
src/plugins/select/bluegene/plugin/state_test.c | 2 +-
3 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/src/plugins/select/bluegene/plugin/bg_block_info.c b/src/plugins/select/bluegene/plugin/bg_block_info.c
index e5f3bba9972..21bd998f064 100644
--- a/src/plugins/select/bluegene/plugin/bg_block_info.c
+++ b/src/plugins/select/bluegene/plugin/bg_block_info.c
@@ -198,6 +198,8 @@ extern int block_ready(struct job_record *job_ptr)
xfree(block_id);
} else
rc = READY_JOB_ERROR;
+/* info("returning %d for job %u %d %d", */
+/* rc, job_ptr->job_id, READY_JOB_ERROR, READY_JOB_FATAL); */
return rc;
}
diff --git a/src/plugins/select/bluegene/plugin/bg_job_run.c b/src/plugins/select/bluegene/plugin/bg_job_run.c
index 9bee222984a..f9a970183db 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_run.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_run.c
@@ -171,18 +171,28 @@ static int _remove_job(db_job_id_t job_id)
return STATUS_OK;
}
- rc = bridge_cancel_job(job_id);
+ /* we have been told the next 2 lines do the same
+ * thing, but I don't believe it to be true. In most
+ * cases when you do a signal of SIGTERM the mpirun
+ * process gets killed with a SIGTERM. In the case of
+ * bridge_cancel_job it always gets killed with a
+ * SIGKILL. From IBM's point of view that is a bad
+ * deally, so we are going to use signal ;).
+ */
+
+// rc = bridge_cancel_job(job_id);
+ rc = bridge_signal_job(job_id, SIGTERM);
if (rc != STATUS_OK) {
if (rc == JOB_NOT_FOUND) {
debug("job %d removed from MMCS", job_id);
return STATUS_OK;
- }
+ }
if(rc == INCOMPATIBLE_STATE)
debug("job %d is in an INCOMPATIBLE_STATE",
job_id);
else
- error("bridge_cancel_job(%d): %s", job_id,
+ error("bridge_cancel_job(%d): %s", job_id,
bg_err_str(rc));
}
}
diff --git a/src/plugins/select/bluegene/plugin/state_test.c b/src/plugins/select/bluegene/plugin/state_test.c
index 5264caa798c..e481b029329 100644
--- a/src/plugins/select/bluegene/plugin/state_test.c
+++ b/src/plugins/select/bluegene/plugin/state_test.c
@@ -339,7 +339,7 @@ static void _test_down_nodes(my_bluegene_t *my_bg)
int bp_num, i, rc;
rm_BP_t *my_bp;
- debug("Running _test_down_nodes");
+ debug2("Running _test_down_nodes");
if ((rc = bridge_get_data(my_bg, RM_BPNum, &bp_num)) != STATUS_OK) {
error("bridge_get_data(RM_BPNum): %s", bg_err_str(rc));
bp_num = 0;
--
GitLab