[ACCUMULO-2964] Unexpected ThriftSecurityException from BatchScanner

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Cannot Reproduce
• Component/s: client, tserver
• Assignee: Unassigned
• Reporter: Josh Elser
• Created: 01/Jul/14 05:08
• Updated: 14/Jul/16 19:36
• Resolved: 15/Jan/15 21:23

Description

This is something I've only seen a handful of times when writing/running tests that stop and restart tservers. After the tserver is restarted, there is a thread (typically running in the master) which is trying to read a table. As such, the thread will continue to poll until the tserver comes up.

Very infrequently, the client gets a ThriftSecurityException with a code of DEFAULT_SECURITY_ERROR and a message of Unknown security exception. There is no additional information in the client log (from the thrift call inside the batchscanner), and the tserver contains no error messages at all.

The error that the client saw.

2014-07-01 04:18:18,971 [impl.TabletServerBatchReaderIterator] DEBUG: Server : host:58090 msg : null
ThriftSecurityException(user:!SYSTEM, code:null)
        at org.apache.accumulo.core.tabletserver.thrift.TabletClientService$startMultiScan_result$startMultiScan_resultStandardScheme.read(TabletClientService.java:10045)
        at org.apache.accumulo.core.tabletserver.thrift.TabletClientService$startMultiScan_result$startMultiScan_resultStandardScheme.read(TabletClientService.java:10022)
        at org.apache.accumulo.core.tabletserver.thrift.TabletClientService$startMultiScan_result.read(TabletClientService.java:9961)
        at org.apache.thrift.TServiceClient.receiveBase(TServiceClient.java:78)
        at org.apache.accumulo.core.tabletserver.thrift.TabletClientService$Client.recv_startMultiScan(TabletClientService.java:313)
        at org.apache.accumulo.core.tabletserver.thrift.TabletClientService$Client.startMultiScan(TabletClientService.java:293)
        at org.apache.accumulo.core.client.impl.TabletServerBatchReaderIterator.doLookup(TabletServerBatchReaderIterator.java:632)
        at org.apache.accumulo.core.client.impl.TabletServerBatchReaderIterator.doLookup(TabletServerBatchReaderIterator.java:592)
        at org.apache.accumulo.core.metadata.MetadataLocationObtainer.lookupTablets(MetadataLocationObtainer.java:181)
        at org.apache.accumulo.core.client.impl.TabletLocatorImpl.processInvalidated(TabletLocatorImpl.java:667)
        at org.apache.accumulo.core.client.impl.TabletLocatorImpl.binRanges(TabletLocatorImpl.java:337)
        at org.apache.accumulo.core.client.impl.TabletLocatorImpl.processInvalidated(TabletLocatorImpl.java:660)
        at org.apache.accumulo.core.client.impl.TabletLocatorImpl._locateTablet(TabletLocatorImpl.java:610)
        at org.apache.accumulo.core.client.impl.TabletLocatorImpl.locateTablet(TabletLocatorImpl.java:440)
        at org.apache.accumulo.core.client.impl.ThriftScanner.scan(ThriftScanner.java:226)
        at org.apache.accumulo.core.client.impl.ScannerIterator$Reader.run(ScannerIterator.java:84)
        at org.apache.accumulo.core.client.impl.ScannerIterator.hasNext(ScannerIterator.java:177)
        at org.apache.accumulo.master.replication.DistributedWorkQueueWorkAssigner.createWork(DistributedWorkQueueWorkAssigner.java:161)
        at org.apache.accumulo.master.replication.DistributedWorkQueueWorkAssigner.assignWork(DistributedWorkQueueWorkAssigner.java:140)
        at org.apache.accumulo.master.replication.WorkDriver.run(WorkDriver.java:97)

The interesting part is that when the client saw this message, the new TabletServer was already started, and the old tabletserver appears to have been dead for 20s. So, the client in the master had been polling for 20s getting a ConnectException (connection refused) which is expected. I don't know why we got this exception after a length of time.

The infrequency in which I see this makes me wonder if the random ports in the new tabletserver are somehow re-grabbing the old tserver's thrift client service port and something is unexpectedly being interpreted as this ThriftSecurityException? That's the only thing that seems remotely possible to me.

Issue Links

• is related to: ACCUMULO-2963 ReplicationDriver daemon dies from RTE thrown out of BatchScanner
• is related to: ACCUMULO-4065 Oneway Thrift calls leave message on client's InputStream
• relates to: ACCUMULO-2746 VolumeIT.testDirtyReplaceVolume failed on AccumuloSecurityException