[jira] [Created] (AMQ-5658) ActiveMQ will not start after KahaDB Corruption due to "Protocol message contained an invalid tag (zero)" error

["Paul Manning (JIRA)" <ji...@apache.org>]

Paul Manning created AMQ-5658:
---------------------------------

             Summary: ActiveMQ will not start after KahaDB Corruption due to "Protocol message contained an invalid tag (zero)" error
                 Key: AMQ-5658
                 URL: https://issues.apache.org/jira/browse/AMQ-5658
             Project: ActiveMQ
          Issue Type: Bug
          Components: Message Store
    Affects Versions: 5.10.0
         Environment: Windows 7
            Reporter: Paul Manning


We experienced an ActiveMQ crash where the KahaDB data files where corrupted. The machine was powered down abruptly (pull the plug).

When the machine restarted, ActiveMQ would not start and the following entries were in the activemq.log:

2015-03-05 09:25:46,791 | INFO  | Corrupt journal records found in 'c:\work\09_git\vc-core\vc-server\build\data\kahadb\db-131.log' between offsets: 31054572..31231936 | org.apache.activemq.store.kahadb.disk.journal.Journal | WrapperSimpleAppMain

followed eventually by: 

2015-03-05 09:25:48,375 | ERROR | Failed to start Apache ActiveMQ ([broker-USATL-L-008043.americas.abb.com-0, null], org.apache.activemq.protobuf.InvalidProtocolBufferException: Protocol message contained an invalid tag (zero).) | org.apache.activemq.broker.BrokerService | WrapperSimpleAppMain

Removing the .data files and the corrupted db-131.log file allows ActiveMQ to restart. However, in that case, we experience message loss. 

Is it possible to only lose the corrupted record instead of the whole data file? 

Tracing through the code, it does not appear that there is any attempt to catch the InvalidProtocolBufferException exception and discard the corrupted record. The exception is raised from CodedInputStream.readTag() during the MessageDatabase.recover() process.

It is worth noting that we have not been able to reproduce this error. I imagine that this type of corruption is rare, but is there any way for a user to recover from this. Any tools, etc.?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)