Github user danny0405 commented on a diff in the pull request:
https://github.com/apache/storm/pull/2433#discussion_r152716882
--- Diff:
storm-client/src/jvm/org/apache/storm/grouping/LoadAwareShuffleGrouping.java ---
@@ -291,7 +291,10 @@ private Scope calculateScope(Map<Integer, NodeInfo>
taskToNodePort, Map<String,
private Map<String, String> getHostToRackMapping(Map<Integer,
NodeInfo> taskToNodePort) {
Set<String> hosts = new HashSet();
for (int task: targetTasks) {
- hosts.add(taskToNodePort.get(task).get_node());
+ //if is a kill, taskToNodePort will be an empty map which is
refreshed by WorkerState
--- End diff --
i means that the taskToNodePort cache is refreshed by worker as a timer
task, if we kill the worker/or rebalance, the local task NodePort maybe a null,
so we should have a check here, or it may throw a NPE
---
