[KUDU-699] PeerManager::Close shouldn't block on requests

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Private Beta
• Fix Version/s: 1.3.0
• Component/s: consensus
• Target Version/s: 1.3.0
• Code Review: http://gerrit.cloudera.org:8080/5490
• Assignee: Todd Lipcon
• Reporter: Todd Lipcon
• Created: 08/Apr/15 19:19
• Updated: 19/Dec/16 09:11
• Resolved: 19/Dec/16 09:11

Description

PeerManager::Close() currently blocks on any outstanding requests before proceeding. This can cause delays in a leader stepping down to become a replica, which also means it can delay vote responses and cause a slower (or flappy) election process.

Issue Links

• is related to: KUDU-1564 Deadlock on raft notification ThreadPool
• is related to: KUDU-1635 Mitigate leader election storms when consensus metadata disk is slow
• is related to: KUDU-1788 Raft UpdateConsensus retry behavior on timeout is counter-productive
• relates to: KUDU-649 Deadlock found with YCSB when creating the table