You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Pavel Tupitsyn (Jira)" <ji...@apache.org> on 2025/01/30 16:28:00 UTC

[jira] [Commented] (IGNITE-24283) Safe time order is broken during concurrent raft command retry and FSM application

    [ https://issues.apache.org/jira/browse/IGNITE-24283?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17922429#comment-17922429 ] 

Pavel Tupitsyn commented on IGNITE-24283:
-----------------------------------------

Merged to main: [e9fd786910ec149d168bfbb08fc89798d5d85cb6|https://github.com/apache/ignite-3/commit/e9fd786910ec149d168bfbb08fc89798d5d85cb6]

> Safe time order is broken during concurrent raft command retry and FSM application
> ----------------------------------------------------------------------------------
>
>                 Key: IGNITE-24283
>                 URL: https://issues.apache.org/jira/browse/IGNITE-24283
>             Project: Ignite
>          Issue Type: Bug
>            Reporter: Alexey Scherbakov
>            Assignee: Alexey Scherbakov
>            Priority: Critical
>              Labels: ignite-3
>             Fix For: 3.0
>
>          Time Spent: 3.5h
>  Remaining Estimate: 0h
>
> When a raft command is retried (for example, due to timeout), it's patched again and the patch is applied to the command's shared state, causing wrong safe time order during command application in FSM.
> Happens only if a raft client is collocated with a raft leader.
> We need to eliminate shared state in raft pipeline.
> I was able to reproduce it with UpsertKvBenchmark using rocksdb storage, while investigating IGNITE-24115



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