Add thread dump collection to report.sh

This patch adds thread dump collection to the report.sh tool. Unlike heap dumps, thread dumps are lightweight (~100ms), but still can help diagnose problems and better optimize application and JVM performance. We simply run kill -SIGQUIT 1 just before collecting the pod logs. Each output log will contain the thread dump at the end. Tested with both ZK and KRaft based clusters. Signed-off-by: Federico Valeri <[email protected]>
strimzi · Dec 2, 2024 · b68f595 · b68f595
1 parent 03e3045
commit b68f595
Showing 1 changed file with 3 additions and 0 deletions.
diff --git a/tools/report.sh b/tools/report.sh
@@ -221,6 +221,7 @@ for RES in "${CLUSTER_RESOURCES[@]}"; do
   get_nonnamespaced_yamls "$RES"
 done
 
+# when kill command is available, each log file includes the thread dump at the end
 get_pod_logs() {
   local pod="$1"
   local con="${2-}"
@@ -230,12 +231,14 @@ get_pod_logs() {
     local logs
     mkdir -p "$OUT_DIR"/reports/logs
     if [[ "$count" -eq 1 ]]; then
+      $KUBE_CLIENT -n "$NAMESPACE" exec -i "$pod" -- sh -c "kill -3 1 ||true"
       logs="$($KUBE_CLIENT -n "$NAMESPACE" logs "$pod" ||true)"
       if [[ -n $logs ]]; then printf "%s" "$logs" > "$OUT_DIR"/reports/logs/"$pod".log; fi
       logs="$($KUBE_CLIENT -n "$NAMESPACE" logs "$pod" -p 2>/dev/null ||true)"
       if [[ -n $logs ]]; then printf "%s" "$logs" > "$OUT_DIR"/reports/logs/"$pod".log.0; fi
     fi
     if [[ "$count" -gt 1 && -n "$con" && "$names" == *"$con"* ]]; then
+      $KUBE_CLIENT -n "$NAMESPACE" exec -i "$pod" -c "$con" -- sh -c "kill -3 1 ||true"
       logs="$($KUBE_CLIENT -n "$NAMESPACE" logs "$pod" -c "$con" ||true)"
       if [[ -n $logs ]]; then printf "%s" "$logs" > "$OUT_DIR"/reports/logs/"$pod"-"$con".log; fi
       logs="$($KUBE_CLIENT -n "$NAMESPACE" logs "$pod" -p -c "$con" 2>/dev/null ||true)"