@cryptotaxi247 / netdata-1 / commits / 7d6dcf48d

Fix race condition on shutdown (#21563)

* Initialize `discovery_thread` properly and adjust `exited` flag handling. Ensure thread-safe cleanup and synchronization primitive usage during thread creation. * Update src/collectors/cgroups.plugin/sys_fs_cgroup.c Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Use proper memory ordering * Do proper cleanup of mutex / cond var * Improve cleanup --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Stelios Fragkakis committed Jan 16, 2026 at 10:23 UTC 7d6dcf48d7493cf07900810a6426e82db8a7e181

2 files changed +24 -8

src/collectors/cgroups.plugin/cgroup-discovery.c

+1 -1

@@ -1356,5 +1356,5 @@ void cgroup_discovery_worker(void *ptr)

1356 cgroup_cleanup_ebpf_integration();

1357 worker_unregister();

1358 service_exits();

1359 - __atomic_store_n(&discovery_thread.exited,1,__ATOMIC_RELAXED);

1359 + __atomic_store_n(&discovery_thread.exited, 1, __ATOMIC_RELEASE);

1360 }

src/collectors/cgroups.plugin/sys_fs_cgroup.c

+23 -7

@@ -63,7 +63,9 @@ struct cgroups_systemd_config_setting cgroups_systemd_options[] = {

63 { .name = NULL, .setting = SYSTEMD_CGROUP_ERR },

64 };

66 -struct discovery_thread discovery_thread;

66 +struct discovery_thread discovery_thread = {

67 + .exited = 1, // Start as "exited" until properly initialized

68 +};

71 /* on Fed systemd is not in PATH for some reason */

@@ -1332,9 +1334,9 @@ static void cgroup_main_cleanup(void *pptr) {

1334

1335 usec_t max = 2 * USEC_PER_SEC, step = 50000;

1336

1335 - if (!__atomic_load_n(&discovery_thread.exited, __ATOMIC_RELAXED)) {

1337 + if (!__atomic_load_n(&discovery_thread.exited, __ATOMIC_ACQUIRE)) {

1338 collector_info("waiting for discovery thread to finish...");

1337 - while (!__atomic_load_n(&discovery_thread.exited, __ATOMIC_RELAXED) && max > 0) {

1339 + while (!__atomic_load_n(&discovery_thread.exited, __ATOMIC_ACQUIRE) && max > 0) {

1340 netdata_mutex_lock(&discovery_thread.mutex);

1341 netdata_cond_signal(&discovery_thread.cond_var);

1342 netdata_mutex_unlock(&discovery_thread.mutex);

@@ -1343,8 +1345,14 @@ static void cgroup_main_cleanup(void *pptr) {

1345 }

1346 }

1347 // We should be done, but just in case, avoid blocking shutdown

1346 - if (__atomic_load_n(&discovery_thread.exited, __ATOMIC_RELAXED))

1347 - (void) nd_thread_join(discovery_thread.thread);

1348 + // Only join and destroy synchronization primitives if thread has exited

1349 + if (__atomic_load_n(&discovery_thread.exited, __ATOMIC_ACQUIRE)) {

1350 + if (discovery_thread.thread) {

1351 + (void) nd_thread_join(discovery_thread.thread);

1352 + netdata_cond_destroy(&discovery_thread.cond_var);

1353 + netdata_mutex_destroy(&discovery_thread.mutex);

1354 + }

1355 + }

1356

1357 static_thread->enabled = NETDATA_MAIN_THREAD_EXITED;

1358 }

@@ -1390,21 +1398,29 @@ void cgroups_main(void *ptr) {

1398 // for the other nodes, the origin server should register it

1399 cgroup_netdev_link_init();

1400

1393 - discovery_thread.exited = 0;

1394 -

1401 if (netdata_mutex_init(&discovery_thread.mutex)) {

1402 collector_error("CGROUP: cannot initialize mutex for discovery thread");

1403 return;

1404 }

1405 if (netdata_cond_init(&discovery_thread.cond_var)) {

1406 collector_error("CGROUP: cannot initialize conditional variable for discovery thread");

1407 + netdata_mutex_destroy(&discovery_thread.mutex);

1408 return;

1409 }

1410

1411 + // Mark thread as "running" only after mutex/cond are initialized

1412 + // but before creating the thread. This ensures cleanup won't try

1413 + // to access uninitialized synchronization primitives.

1414 + // Use RELEASE ordering so readers with ACQUIRE see initialized mutex/cond.

1415 + __atomic_store_n(&discovery_thread.exited, 0, __ATOMIC_RELEASE);

1416 +

1417 discovery_thread.thread = nd_thread_create("CGDISCOVER", NETDATA_THREAD_OPTION_DEFAULT, cgroup_discovery_worker, NULL);

1418

1419 if (!discovery_thread.thread) {

1420 collector_error("CGROUP: cannot create thread worker");

1421 + __atomic_store_n(&discovery_thread.exited, 1, __ATOMIC_RELEASE); // Reset since thread wasn't created

1422 + netdata_cond_destroy(&discovery_thread.cond_var);

1423 + netdata_mutex_destroy(&discovery_thread.mutex);

1424 return;

1425 }

1426