[jira] [Resolved] (HBASE-29323) Use Priority Handler for all RegionServerStatus rpc at Master

["Nihal Jain (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-29323?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Nihal Jain resolved HBASE-29323.
--------------------------------
    Hadoop Flags: Reviewed
      Resolution: Fixed

Merged to code base. Pushed to branch-2.6+. Please let me know if should pull to branch-2.5 also?

 

Thank you [~umesh9414] for the contribution. Thank you [~zhangduo], [~vjasani], [~mnpoonia] for the reviews.

> Use Priority Handler for all RegionServerStatus rpc at Master
> -------------------------------------------------------------
>
>                 Key: HBASE-29323
>                 URL: https://issues.apache.org/jira/browse/HBASE-29323
>             Project: HBase
>          Issue Type: Improvement
>          Components: master
>    Affects Versions: 2.6.2
>            Reporter: Umesh Kumar Kumawat
>            Assignee: Umesh Kumar Kumawat
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 2.7.0, 3.0.0-beta-2, 2.6.3
>
>
> There are two main reasons to use the Priority Handler:
> 1. These reports like report for splitWalProcedure are critical for completing region movements and ensuring availability. Any delay in reporting can directly impact system availability.
> *2. Preventing Deadlocks on simultaneous Region Server Draining and serverCrash:*
> When draining a Region Server, the {{MoveWithAck}} API is used. This API waits for the {{TRSP}} to complete. If multiple Region Servers are drained simultaneously, all general handler threads can become occupied.
> If the draining server get crashed during this period, we will first do wal split. All the above moving rpc leaves no threads available to process `reportProcedureDone` for splitWALProcedure, resulting in a deadlock that blocks SCP from completing.
> Using a Priority Handler ensures that critical operations like state transitions are not blocked by general handler saturation.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)