Add healthcheck to ensure docker service is working

AMDResearch · Jun 26, 2024 · d31c67c · d31c67c
1 parent d78c9da
commit d31c67c
Showing 1 changed file with 6 additions and 0 deletions.
diff --git a/test/docker/slurm/compose.yaml b/test/docker/slurm/compose.yaml
@@ -36,6 +36,12 @@ services:
       - controller
     links:
       - controller
+    healthcheck:
+      test: curl --fail localhost:8000/metrics || exit 1
+      interval: 15s
+      timeout: 10s
+      retries: 5
+      start_period: 15s
 
 volumes:
   jobs_dir: