[HBASE-15703] Deadline scheduler needs to return to the client info about skipped calls, not just drop them - ASF JIRA

Voters

Watch issue

Watchers

Create sub-task

Link

Clone

Update Comment Author

Replace String in Comment

Update Comment Visibility

Delete Comments

XML

Word

Printable

JSON

Details

Type: Bug
Status: Resolved
Priority: Critical
Resolution: Fixed
Affects Version/s: 1.3.0
Fix Version/s: 1.3.0, 2.0.0
Component/s: IPC/RPC, Scheduler
Labels:
None

Hadoop Flags:

Reviewed
Release Note:

Hide
With previous deadline mode of RPC scheduling (the implementation in SimpleRpcScheduler, which is basically a FIFO except that long-running scans are de-prioritized) and FIFO-based RPC scheduler clients are getting CallQueueTooBigException when RPC call queue is full.

With this patch and when hbase.ipc.server.callqueue.type property is set to "codel" mode, clients will also be getting CallDroppedException, which means that the request was discarded by the server as it considers itself to be overloaded and starts to drop requests to avoid going down under the load. The clients will retry upon receiving this exception. It doesn't clear MetaCache with region locations.

Show
With previous deadline mode of RPC scheduling (the implementation in SimpleRpcScheduler, which is basically a FIFO except that long-running scans are de-prioritized) and FIFO-based RPC scheduler clients are getting CallQueueTooBigException when RPC call queue is full. With this patch and when hbase.ipc.server.callqueue.type property is set to "codel" mode, clients will also be getting CallDroppedException, which means that the request was discarded by the server as it considers itself to be overloaded and starts to drop requests to avoid going down under the load. The clients will retry upon receiving this exception. It doesn't clear MetaCache with region locations.

Description

In AdaptiveLifoCodelCallQueue we drop the calls when we think we're overloaded, we should instead return CallDroppedException to the cleent or something.