You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@ignite.apache.org by "Alexey Scherbakov (Jira)" <ji...@apache.org> on 2023/11/28 12:26:00 UTC

[jira] [Created] (IGNITE-20985) Optimize WaiterImpl#notifyLocked threading model

Alexey Scherbakov created IGNITE-20985:
------------------------------------------

             Summary: Optimize WaiterImpl#notifyLocked threading model
                 Key: IGNITE-20985
                 URL: https://issues.apache.org/jira/browse/IGNITE-20985
             Project: Ignite
          Issue Type: Improvement
    Affects Versions: 3.0
            Reporter: Alexey Scherbakov
             Fix For: 3.0


There are two potential issues with current approach:
 # A next locker wait future is completed in the same thread. This means the whole lock queue will be processed by a single thread and can introduce starvation.
 # This method can be called from a sensitive thread (like jraft STM updating thread), worsening first case.

This can cause latency issues in contented scenarios. It looks more safe is to invoke the next locker asynchronously. Needs performance testing to figure out best behavior.



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