[YARN-2063] ZKRMStateStore: Better handling of operation failures

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 2.4.0
• Component/s: resourcemanager
• Target Version/s: 2.6.0
• Assignee: Karthik Kambatla
• Reporter: Karthik Kambatla
• Created: 15/May/14 02:19
• Updated: 28/Dec/14 18:51
• Resolved: 28/Dec/14 18:51

Description

Today, when a ZK operation fails, we handle connection-loss and operation-timeout the same way. This could definitely use some improvements:

1. Add special handling for other error codes
2. Connection-loss: Nullify zkClient, so a new connection is established
3. Operation-timeout: Retry a few times with exponential delay?

Issue Links

• Is contained by: YARN-2716 Refactor ZKRMStateStore retry code with Apache Curator