From 0fc4539e7a0e23f1eaab5c692ce33418c9f7768e Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Thu, 4 Apr 2013 12:24:13 -0700
Subject: [PATCH] Prevent assert failure in scheduling logic
This fixes a bug introduced in commit f1cf6d2df1abbcf2703e5ba81bc4792487a691d8
Without this change, a job with a partition related reason for not running
(e.g. MaxNodes) plus some other reason (e.g. dependency), could after satisfying
the second reason (e.g. dependency), have the select_nodes() function executed
and return SUCCESS on an allocation request, but it would not actually be
allocation resources since select_nodes() would interpret the request as a
test to see if it could ever run (e.g. will_run).
Bug 263
---
src/slurmctld/job_mgr.c | 12 ++++--------
src/slurmctld/job_scheduler.c | 15 ++++++++++-----
src/slurmctld/node_scheduler.c | 14 +-------------
3 files changed, 15 insertions(+), 26 deletions(-)
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index c79055a2077..b9f21108199 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -2727,6 +2727,8 @@ static int _select_nodes_parts(struct job_record *job_ptr, bool test_only,
}
list_iterator_destroy(iter);
} else {
+ if (job_limits_check(&job_ptr) != WAIT_NO_REASON)
+ test_only = true;
rc = select_nodes(job_ptr, test_only, select_node_bitmap);
}
@@ -2851,14 +2853,8 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
test_only = will_run || (allocate == 0);
no_alloc = test_only || too_fragmented ||
- (!top_prio) || (!independent);
- if (!no_alloc && !avail_front_end()) {
- debug("sched: job_allocate() returning, no front end nodes "
- "are available");
- error_code = ESLURM_NODES_BUSY;
- } else {
- error_code = _select_nodes_parts(job_ptr, no_alloc, NULL);
- }
+ (!top_prio) || (!independent) || !avail_front_end();
+ error_code = _select_nodes_parts(job_ptr, no_alloc, NULL);
if (!test_only) {
last_job_update = now;
slurm_sched_schedule(); /* work for external scheduler */
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index cf9c95d2cf4..4052a61f9de 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -200,13 +200,16 @@ static bool _job_runnable_test1(struct job_record *job_ptr, bool clear_start)
/* Job and partition tests for ability to run now */
static bool _job_runnable_test2(struct job_record *job_ptr)
{
- if (!part_policy_job_runnable_state(job_ptr)) {
- job_ptr->state_reason = job_limits_check(&job_ptr);
- xfree(job_ptr->state_desc);
+ int reason;
- if (job_ptr->state_reason != WAIT_NO_REASON)
- return false;
+ reason = job_limits_check(&job_ptr);
+ if ((reason != job_ptr->state_reason) &&
+ (part_policy_job_runnable_state(job_ptr))) {
+ job_ptr->state_reason = reason;
+ xfree(job_ptr->state_desc);
}
+ if (reason != WAIT_NO_REASON)
+ return false;
return true;
}
@@ -498,6 +501,8 @@ next_part: part_ptr = (struct part_record *)
continue;
}
}
+ if (job_limits_check(&job_ptr) != WAIT_NO_REASON)
+ continue;
/* Test for valid account, QOS and required nodes on each pass */
if (job_ptr->state_reason == FAIL_ACCOUNT) {
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index f7135bdcdb1..dec0a15982c 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -1426,7 +1426,6 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
struct node_set *node_set_ptr = NULL;
struct part_record *part_ptr = NULL;
uint32_t min_nodes, max_nodes, req_nodes;
- enum job_state_reason fail_reason;
time_t now = time(NULL);
bool configuring = false;
List preemptee_job_list = NULL;
@@ -1458,13 +1457,6 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
return ESLURM_JOB_HELD;
}
- /* Confirm that partition is up and has compatible nodes limits.
- * If not, we still continue to determine if the job can ever run.
- * For example, the geometry might be incompatible with BlueGene. */
- fail_reason = job_limits_check(&job_ptr);
- if (fail_reason != WAIT_NO_REASON)
- test_only = true;
-
/* build sets of usable nodes based upon their configuration */
error_code = _build_node_list(job_ptr, &node_set_ptr, &node_set_size);
if (error_code)
@@ -1552,7 +1544,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
}
}
}
- if (error_code || fail_reason) {
+ if (error_code) {
/* Fatal errors for job here */
if (error_code == ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE) {
/* Too many nodes requested */
@@ -1561,10 +1553,6 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
job_ptr->state_reason = WAIT_PART_NODE_LIMIT;
xfree(job_ptr->state_desc);
last_job_update = now;
- } else if (fail_reason) {
- job_ptr->state_reason = fail_reason;
- xfree(job_ptr->state_desc);
- last_job_update = now;
/* Non-fatal errors for job below */
} else if (error_code == ESLURM_NODE_NOT_AVAIL) {
--
GitLab