mxqd: killall_over_memory: Send SIGKILL after sending SIGTERM

mariux64 · Oct 30, 2015 · 925a2ee · 925a2ee
1 parent 66c151f
commit 925a2ee
Showing 1 changed file with 10 additions and 2 deletions.
diff --git a/mxqd.c b/mxqd.c
@@ -1660,9 +1660,16 @@ int killall_over_memory(struct mxq_server *server)
 
             for (jlist = glist->jobs; jlist; jlist = jlist->next) {
                 unsigned long long int memory;
+                int signal;
 
                 job = &jlist->job;
 
+                /* sigterm has already been send last round ? */
+                if (jlist->max_sum_rss/1024 > group->job_memory)
+                    signal = SIGKILL;
+                else
+                    signal = SIGTERM;
+
                 pinfo = mx_proc_tree_proc_info(ptree, job->host_pid);
                 if (!pinfo) {
                     mx_log_warning("killall_over_memory(): Can't find process with pid %llu in process tree",
@@ -1678,16 +1685,17 @@ int killall_over_memory(struct mxq_server *server)
                 if (jlist->max_sum_rss/1024 <= group->job_memory)
                     continue;
 
-                mx_log_info("killall_over_memory(): used(%lluMiB) > requested(%lluMiB): Sending signal=TERM to job=%s(%d):%lu:%lu pid=%d",
+                mx_log_info("killall_over_memory(): used(%lluMiB) > requested(%lluMiB): Sending signal=%d to job=%s(%d):%lu:%lu pid=%d",
                     jlist->max_sum_rss/1024,
                     group->job_memory,
+                    signal,
                     group->user_name,
                     group->user_uid,
                     group->group_id,
                     job->job_id,
                     job->host_pid);
 
-                kill(job->host_pid, SIGTERM);
+                kill(job->host_pid, signal);
             }
         }
     }