[HBASE-6227] SSH and cluster startup causes data loss

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.94.1, 0.95.0
• Component/s: master
• Hadoop Flags: Reviewed
• Assignee: Chunhui Shen
• Reporter: Chunhui Shen
• Created: 18/Jun/12 03:47
• Updated: 26/Feb/13 08:16
• Resolved: 27/Jun/12 17:18

Description

In AssignmentManager#processDeadServersAndRegionsInTransition, if servershutdownhandler is processing and master consider it cluster startup, master will assign all user regions, however, servershutdownhandler has not completed splitting logs.

Let me describe it in more detail.

Suppose there are two regionservers A1 and B1, A1 carried META and ROOT

1.master restart and completed assignRootAndMeta

2.A1 and B1 are both restarted, new regionservers are A2 and B2

3.SSH which processed for A1 completed assigning ROOT and META

4.master do rebuilding user regions and no regions added to master's region list

5.master consider it as a cluster startup, and assign all user regions

6.SSH which processed for B1 start to split B1's logs

7.All regions' data carried by B1 would loss.

Issue Links

• relates to: HBASE-5916 RS restart just before master intialization we make the cluster non operative