[ACCUMULO-422] Bulk import failing when tablet server dies

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.4.0
• Labels: 14_qa_bug

Environment

10 node cluster running 1.4.0-SNAPSHOT

• Assignee: Unassigned
• Reporter: Keith Turner
• Created: 21/Feb/12 18:47
• Updated: 10/Feb/15 20:53
• Resolved: 28/Feb/12 18:38

Description

Saw this issue while running random walk test w/ agitation. The bulk import code picks random tablet servers and ask them to bulk load files. If a tablet server dies it takes 30 seconds for the master to see the zookeeper lock was lost. During this 30 second period the bulk import code will still try to use the tserver and fail. After it fails three times it will mark the file as a failure. This all happens within a second.

The bulk import code should probably catch TTransportException and black list the tablet server for that bulk import transaction.

Issue Links

• breaks: ACCUMULO-3574 preferCachedConnections=false is not respected