From ecbfb937a417a5bbfb2eda850ce715d1cc1f5dc9 Mon Sep 17 00:00:00 2001
From: Ryan Cox <ryan_cox@byu.edu>
Date: Fri, 8 Nov 2013 15:21:39 -0700
Subject: [PATCH] use mem and memsw failcnt, check for existence
---
src/plugins/task/cgroup/task_cgroup_memory.c | 46 +++++++++++++++-----
1 file changed, 34 insertions(+), 12 deletions(-)
diff --git a/src/plugins/task/cgroup/task_cgroup_memory.c b/src/plugins/task/cgroup/task_cgroup_memory.c
index 360df5f6d1b..abed8fb621c 100644
--- a/src/plugins/task/cgroup/task_cgroup_memory.c
+++ b/src/plugins/task/cgroup/task_cgroup_memory.c
@@ -451,6 +451,22 @@ extern int task_cgroup_memory_attach_task(slurmd_job_t *job)
return fstatus;
}
+/* return 1 if failcnt file exists and is > 0 */
+int failcnt_non_zero(xcgroup_t* cg, char* param)
+{
+ int fstatus = XCGROUP_ERROR;
+ uint64_t value;
+ fstatus = xcgroup_get_uint64_param(cg,
+ param,
+ &value);
+ if(fstatus != XCGROUP_SUCCESS) {
+ debug2("unable to read '%s' from '%s'", param, cg->path);
+ return 0;
+ }
+ else
+ return value > 0;
+}
+
extern int task_cgroup_memory_check_oom(slurmd_job_t *job)
{
xcgroup_t memory_cg;
@@ -463,20 +479,26 @@ extern int task_cgroup_memory_check_oom(slurmd_job_t *job)
* for a step and vice versa...
* can't tell which is which so we'll treat
* them the same */
- xcgroup_get_uint64_param(&step_memory_cg,
- "memory.memsw.failcnt",
- &memory_memsw_failcnt);
- if (memory_memsw_failcnt > 0)
+ if(failcnt_non_zero(&step_memory_cg,
+ "memory.memsw.failcnt"))
+ error("Exceeded step memory limit at some "
+ "point. oom-killer likely killed a "
+ "process.");
+ else if(failcnt_non_zero(&step_memory_cg,
+ "memory.failcnt"))
error("Exceeded step memory limit at some "
- "point. oom-killer likely "
- "killed a process.");
- xcgroup_get_uint64_param(&job_memory_cg,
- "memory.memsw.failcnt",
- &memory_memsw_failcnt);
- if (memory_memsw_failcnt > 0)
+ "point. Step may have been partially "
+ "swapped out to disk.");
+ if(failcnt_non_zero(&job_memory_cg,
+ "memory.memsw.failcnt"))
+ error("Exceeded job memory limit at some "
+ "point. oom-killer likely killed a "
+ "process.");
+ else if(failcnt_non_zero(&job_memory_cg,
+ "memory.failcnt"))
error("Exceeded job memory limit at some "
- "point. oom-killer likely "
- "killed a process.");
+ "point. Job may have been partially "
+ "swapped out to disk.");
xcgroup_unlock(&memory_cg);
} else
error("task/cgroup task_cgroup_memory_check_oom: "
--
GitLab