Major Description
So we had the following stack trace with a customer:
2017-10-04 11:25:30.339 ERROR (xxxx) [ ] o.a.s.c.Overseer Exception in Overseer main queue loop
org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode for /collections/xxxx/state.json
at org.apache.zookeeper.KeeperException.create(KeeperException.java:111)
at org.apache.zookeeper.KeeperException.create(KeeperException.java:51)
at org.apache.zookeeper.ZooKeeper.create(ZooKeeper.java:783)
at org.apache.solr.common.cloud.SolrZkClient$9.execute(SolrZkClient.java:391)
at org.apache.solr.common.cloud.SolrZkClient$9.execute(SolrZkClient.java:388)
at org.apache.solr.common.cloud.ZkCmdExecutor.retryOperation(ZkCmdExecutor.java:60)
at org.apache.solr.common.cloud.SolrZkClient.create(SolrZkClient.java:388)
at org.apache.solr.cloud.overseer.ZkStateWriter.writePendingUpdates(ZkStateWriter.java:235)
at org.apache.solr.cloud.overseer.ZkStateWriter.enqueueUpdate(ZkStateWriter.java:152)
at org.apache.solr.cloud.Overseer$ClusterStateUpdater.processQueueItem(Overseer.java:271)
at org.apache.solr.cloud.Overseer$ClusterStateUpdater.run(Overseer.java:199)
at java.lang.Thread.run(Thread.java:748)
I want to highlight:
at org.apache.solr.cloud.overseer.ZkStateWriter.enqueueUpdate(ZkStateWriter.java:152)
at org.apache.solr.cloud.Overseer$ClusterStateUpdater.processQueueItem(Overseer.java:271)
This ends up coming from Overseer:
while (data != null) {
final ZkNodeProps message = ZkNodeProps.load(data);
log.debug("processMessage: workQueueSize: {}, message = {}", workQueue.getStats().getQueueLength(), message);
// force flush to ZK after each message because there is no fallback if workQueue items
// are removed from workQueue but fail to be written to ZK
*clusterState = processQueueItem(message, clusterState, zkStateWriter, false, null);
workQueue.poll(); // poll-ing removes the element we got by peek-ing*
data = workQueue.peek();
}
Note: The processQueueItem comes before the poll, therefore upon a thrown exception the same node/message that won't process becomes stuck. This made a large cluster unable to come up on it's own without deleting the problem node.