[KUDU-2918] Rebalancer can fail when a service queue is full

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.11.0
• Component/s: CLI, ksck
• Labels: stability supportability
• Assignee: Unassigned
• Reporter: Adar Dembo
• Created: 15/Aug/19 00:54
• Updated: 03/Jun/20 15:38

Description

The various low-level RPCs issued by ksck aren't retried if the corresponding service queues are full. These include GetConsensusState, GetStatus, and ListTablets.

Without retries, ksck (and the rebalancer) can fail midway:

I0812 11:21:10.669682 42799 rebalancer.cc:831] tablet d729fb149e804696a0862adacb725d66: a0dca75bbbfb4de69616694834adf930 -> 24d0eb73b3c64a0f901ae092186b3439 move is abandoned: Remote error: Service unavailable: GetConsensusState request on kudu.consensus.ConsensusService from 10.17.182.15:50754 dropped due to backpressure. The service queue is full; it has 50 items.
I0812 11:21:10.871894 42799 rebalancer.cc:239] re-synchronizing cluster state
Illegal state: tablet server 0d88ff7360b74d1e81cd2ccd41fab8a5 (foo.bar.com:7050): unacceptable health status UNAVAILABLE

The helper classes in rpc/rpc.h may be useful here.