lrun: detect OOM directly

This should work more reliably than testing cg.memory_peak() >= config.memory_limit.
quark-zju · Jan 9, 2019 · 76063ef · 76063ef
1 parent f93775f
commit 76063ef
Show file tree

Hide file tree

Showing 3 changed files with 18 additions and 1 deletion.
diff --git a/src/cgroup.cc b/src/cgroup.cc
@@ -422,6 +422,12 @@ long long Cgroup::memory_limit() const {
     return strconv::to_longlong(limit);
 }
 
+bool Cgroup::is_under_oom() const {
+    string content = get(CG_MEMORY, "memory.oom_control");
+    return content.find("under_oom 1") != string::npos;
+
+}
+
 long long Cgroup::set_memory_limit(long long bytes) {
     int e = 1;
 

diff --git a/src/cgroup.h b/src/cgroup.h
@@ -226,6 +226,13 @@ namespace lrun {
              */
             long long memory_limit() const;
 
+            /**
+             * test if the cgroup is under OOM
+             * @return  true   if under OOM
+             *          false  otherwise
+             */
+            bool is_under_oom() const;
+
             /**
              * get cpu usage
              * @return  cpu usage in seconds

diff --git a/src/lrun.cc b/src/lrun.cc
@@ -319,7 +319,11 @@ static int run_command() {
         }
 
         // check memory limit
-        if (cg.memory_peak() >= config.memory_limit && config.memory_limit > 0) {
+        if (cg.is_under_oom()) {
+            INFO("OOM detected");
+            exceeded_limit = "MEMORY";
+            break;
+        } else if (cg.memory_peak() >= config.memory_limit && config.memory_limit > 0) {
             exceeded_limit = "MEMORY";
             break;
         }