Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Won't Fix
-
None
-
None
-
None
-
None
Description
PENDING_OPEN currently is a murky state. Its a master in-memory state with no corresponding znode state that sits between OFFLINE and OPENING states.
The OFFLINE state is set by the master when it goes to open a region. OPENING is set by the regionserver after its assumed control of a region and is moving it through the OPENING process. PENDING_OPEN currently spans the open rpc invocation. This state is in place pre-open-rpc-invocation, during open-rpc-invocation, and post-rpc-invocation until we get the OPENING callback. That PENDING_OPEN covers this many different conditions effectively makes it unactionable.
This issue proposes PENDING_OPEN only be in place post-rpc-invocation. Now its meaning is clear as the space between rpc-open-invocation and our receiving the callback which sets RegionState to OPENING. PENDING_OPEN becomes actionable too in that if a regionserver dies post rpc-open-invocation, we know that we can reassign the region.
See https://issues.apache.org/jira/browse/HBASE-6060?focusedCommentId=13292646&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13292646 for more discussion.
Attachments
Attachments
Issue Links
- is related to
-
HBASE-6060 Regions's in OPENING state from failed regionservers takes a long time to recover
- Closed