From a0e3e5de36e94b3878970ea0ee8766881be30400 Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Tue, 23 Feb 2016 16:37:25 -0800
Subject: [PATCH] Improve some step allocation logs
Include warning for Cray simulation as reminder for developers to
change code as needed.
---
src/slurmctld/step_mgr.c | 28 ++++++++++++++++++++--------
1 file changed, 20 insertions(+), 8 deletions(-)
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index f8770f527d9..9c7a3b08b5b 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -1925,8 +1925,8 @@ extern void step_alloc_lps(struct step_record *step_ptr)
if (slurmctld_conf.debug_flags & DEBUG_FLAG_CPU_BIND)
_dump_step_layout(step_ptr);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_STEPS) {
- info("step alloc of %s procs: %u of %u",
- node_record_table_ptr[i_node].name,
+ info("step alloc on job node %d (%s) used %u of %u CPUs",
+ job_node_inx, node_record_table_ptr[i_node].name,
job_resrcs_ptr->cpus_used[job_node_inx],
job_resrcs_ptr->cpus[job_node_inx]);
}
@@ -1979,6 +1979,9 @@ static void _dump_step_layout(struct step_record *step_ptr)
static void _step_dealloc_lps(struct step_record *step_ptr)
{
+#if !defined(HAVE_NATIVE_CRAY) && !defined(HAVE_CRAY_NETWORK)
+ static bool cray_simulate_logged = false;
+#endif
struct job_record *job_ptr = step_ptr->job_ptr;
job_resources_t *job_resrcs_ptr = job_ptr->job_resrcs;
int cpus_alloc;
@@ -2023,12 +2026,21 @@ static void _step_dealloc_lps(struct step_record *step_ptr)
cpus_alloc = step_ptr->step_layout->tasks[step_node_inx] *
step_ptr->cpus_per_task;
#endif
- if (job_resrcs_ptr->cpus_used[job_node_inx] >= cpus_alloc)
+ if (job_resrcs_ptr->cpus_used[job_node_inx] >= cpus_alloc) {
job_resrcs_ptr->cpus_used[job_node_inx] -= cpus_alloc;
- else {
- error("_step_dealloc_lps: cpu underflow for %u.%u",
- job_ptr->job_id, step_ptr->step_id);
+ } else {
+ error("%s: CPU underflow for %u.%u (%u<%u on job node %d)",
+ __func__, job_ptr->job_id, step_ptr->step_id,
+ job_resrcs_ptr->cpus_used[job_node_inx],
+ cpus_alloc, job_node_inx);
job_resrcs_ptr->cpus_used[job_node_inx] = 0;
+#if !defined(HAVE_NATIVE_CRAY) && !defined(HAVE_CRAY_NETWORK)
+ if (!cray_simulate_logged) {
+ error("Remember to comment out post_job_step() "
+ "call in slurmctld/step_mgr.c");
+ cray_simulate_logged = true;
+ }
+#endif
}
if (step_ptr->pn_min_memory && _is_mem_resv()) {
uint32_t mem_use = step_ptr->pn_min_memory;
@@ -2048,8 +2060,8 @@ static void _step_dealloc_lps(struct step_record *step_ptr)
}
}
if (slurmctld_conf.debug_flags & DEBUG_FLAG_STEPS) {
- info("step dealloc of %s procs: %u of %u",
- node_record_table_ptr[i_node].name,
+ info("step dealloc on job node %d (%s) used: %u of %u CPUs",
+ job_node_inx, node_record_table_ptr[i_node].name,
job_resrcs_ptr->cpus_used[job_node_inx],
job_resrcs_ptr->cpus[job_node_inx]);
}
--
GitLab