fix vllm-disagg deadlock: stop router after rank 0 container exits

The vllm-router runs as a separate container on node 0. After node 0's
main container finishes the benchmark and exits, decode nodes remain
stuck waiting for the router port to close. The router cleanup in
job.slurm can't run until srun completes, but srun can't complete
because decode nodes are blocked — deadlock.

Fix: skip exec on rank 0 for vllm-disagg so the srun bash script
continues after docker exits and can stop the router container,
allowing decode nodes to detect the port closure and exit.

Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>

Author: ichbinblau

benchmarks/multi_node/amd_utils/job.slurm

@@ -427,7 +427,16 @@ if [[ \"$ENGINE\" == \"vllm-disagg\" && \"$ROUTER_TYPE\" == \"vllm-router\" && \
             --log-level info 2>&1 | tee /run_logs/slurm_job-${SLURM_JOB_ID}/vllm_router_\$(hostname).log \"
 fi
 
-exec \$DOCKER_CMD run \
+# Skip exec on vllm-disagg rank 0 so we can stop the router after the main
+# container exits.  Without this, decode nodes block forever waiting for the
+# router port to close (the router is a separate container).
+MAYBE_EXEC=exec
+if [[ \"$ENGINE\" == \"vllm-disagg\" && \"$ROUTER_TYPE\" == \"vllm-router\" && \"\$SLURM_PROCID\" == \"0\" ]]; then
+    MAYBE_EXEC=
+    set +e
+fi
+
+\$MAYBE_EXEC \$DOCKER_CMD run \
     --init \
     --stop-timeout 10 \
     --device /dev/dri \
@@ -468,11 +477,11 @@ exec \$DOCKER_CMD run \
         '"$RUN_FILE_FULL"' 2>&1 | tee /run_logs/slurm_job-'\"\$SLURM_JOB_ID\"'/server_\$(hostname).log
     '
 
+# Only reached when exec was skipped (vllm-disagg rank 0)
 DOCKER_EXIT_CODE=\$?
-if [[ \$DOCKER_EXIT_CODE -ne 0 ]]; then
-  echo \"ERROR: docker exited rc=\$DOCKER_EXIT_CODE on \$(hostname)\"
-  exit \$DOCKER_EXIT_CODE
-fi
+echo \"[rank 0] Main container exited (rc=\$DOCKER_EXIT_CODE). Stopping vllm-router...\"
+\$DOCKER_CMD rm -f \"$ROUTER_CONT_NAME\" 2>/dev/null || true
+exit \$DOCKER_EXIT_CODE
 "
 
 if [[ "${KEEP_CONTAINERS}" != "1" ]]; then