[HADOOP-14035] Reduce fair call queue backoff's impact on clients

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.7.0
• Fix Version/s: 2.9.0, 3.0.0-alpha4, 2.8.2
• Component/s: ipc
• Target Version/s: 2.8.1
• Hadoop Flags: Reviewed
• Assignee: Daryn Sharp
• Reporter: Daryn Sharp
• Created: 27/Jan/17 22:31
• Updated: 24/Apr/18 20:52
• Resolved: 06/Jun/17 17:36

Description

When fcq backoff is enabled and an abusive client overflows the call queue, its connection is closed, as well as subsequent good client connections. Disconnects are very disruptive, esp. to multi-threaded clients with multiple outstanding requests, or clients w/o a retry proxy (ex. datanodes).

Until the abusive user is downgraded to a lower priority queue, disconnect/reconnect mayhem occurs which significantly degrades performance. Server metrics look good despite horrible client latency.

The fcq should utilize selective ipc disconnects to avoid pushback disconnecting good clients.

Issue Links

• duplicates: HADOOP-14440 Add metrics for connections dropped