Description
This might happen on heavy load in case of lagging HBase when sharing one HConnection by multiple threads:
2012-01-26 13:59:38,396 ERROR [http://*:8080-251-EventThread] zookeeper.ClientCnxn$EventThread(532): Error while calling watcher
java.lang.NullPointerException
at org.apache.hadoop.hbase.client.HConnectionManager$HConnectionImplementation.resetZooKeeperTrackers(HConnectionManager.java:533)
at org.apache.hadoop.hbase.client.HConnectionManager$HConnectionImplementation.abort(HConnectionManager.java:1536)
at org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher.connectionEvent(ZooKeeperWatcher.java:344)
at org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher.process(ZooKeeperWatcher.java:262)
at org.apache.zookeeper.ClientCnxn$EventThread.processEvent(ClientCnxn.java:530)
at org.apache.zookeeper.ClientCnxn$EventThread.run(ClientCnxn.java:506)
The following code is not protected against NPE:
private synchronized void resetZooKeeperTrackers() throws ZooKeeperConnectionException { LOG.info("Trying to reconnect to zookeeper"); masterAddressTracker.stop(); masterAddressTracker = null; rootRegionTracker.stop(); rootRegionTracker = null; clusterId = null; this.zooKeeper = null; setupZookeeperTrackers(); }
In some cases as proven by the log snippet above it might happen that either masterAddressTracker or rootRegionTracker might be null.
Because of the NPE the code can't reach setupZookeeperTrackers() call.
This should be fixed at least the way as shown in one of the patches in HBASE-5153
LOG.info("Trying to reconnect to zookeeper."); if (this.masterAddressTracker != null) { this.masterAddressTracker.stop(); this.masterAddressTracker = null; } if (this.rootRegionTracker != null) { this.rootRegionTracker.stop(); this.rootRegionTracker = null; }
Attachments
Issue Links
- is related to
-
HBASE-4893 HConnectionImplementation is closed but not deleted
-
- Closed
-
-
-
- Closed
-