[HBASE-6908] Pluggable Call BlockingQueue for HBaseServer

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.5.0, 3.0.0-alpha-2, 2.4.6
• Component/s: IPC/RPC
• Hadoop Flags: Reviewed

Release Note

Can pass in a FQCN to load as the call queue implementation.

Standardized arguments to the constructor are the max queue length, the PriorityFunction, and the Configuration.

PluggableBlockingQueue abstract class provided to help guide the correct constructor signature.

Hard fails with PluggableRpcQueueNotFound if the class fails to load as a BlockingQueue<CallRunner>

Upstreaming on behalf of Hubspot, we are interested in defining our own custom RPC queue and don't want to get involved in necessarily upstreaming internal requirements/iterations.

• Assignee: Richard Marscher
• Reporter: James Taylor
• Created: 01/Oct/12 21:12
• Updated: 14/Jun/22 21:23
• Resolved: 16/Aug/21 19:10

Description

Allow the BlockingQueue implementation class to be specified in the HBase config to enable different behavior than a FIFO queue. The use case we have is around fairness and starvation for big scans that are parallelized on the client. It's easy to fill up the HBase server Call BlockingQueue when processing a single parallelized scan, leadng other scans to time out. Instead, doing round robin processesing on a dequeue through a different BlockingQueue implementation will prevent this from occurring.

Issue Links

• links to: GitHub Pull Request #3522
• links to: GitHub Pull Request #3587