[ 
https://issues.apache.org/jira/browse/IGNITE-18599?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Maksim Timonin updated IGNITE-18599:
------------------------------------
    Ignite Flags:   (was: Docs Required,Release Notes Required)

> Incremental snapshot ID in tx thread might be skipped if previous hasn't 
> finished yet
> -------------------------------------------------------------------------------------
>
>                 Key: IGNITE-18599
>                 URL: https://issues.apache.org/jira/browse/IGNITE-18599
>             Project: Ignite
>          Issue Type: Bug
>            Reporter: Maksim Timonin
>            Priority: Major
>              Labels: IEP-89, ise
>
> Transactions could be included into incremental snapshot while is excluded on 
> other nodes. Check this scenario:
>  # Incremental snapshot (A) started, processed, and be prepared for finish. 
> Coordinator sends FullMessage to clear snapshot resources (incSnpId, 
> clusterSnpReq, etc) on all nodes.
>  # Originated node (O) handles the FullMessage, clear resources for snapshot 
> (A), and it is available to start new incremental snapshot.
>  # User starts incremental snapshot (B) on this node (O), while other nodes 
> (N) still did not receive the FullMessage.
>  # Node (O) starts wrapping messages and send it (X) to node (N).
>  # Node (N) skips handling received incSnpId from IncSnpAwareMessage, as it 
> still has previous one.
>  # Node (N) received the FullMessage, and clear resources for snp (A)
>  # Node (N) received snapshot request for (B) by discovery.
> Then:
>  # Snapshot (B) didn't fail, and consider as correct.
>  # Transaction (X) on node (N) might miss transactions check required for 
> incremental snapshot. Then Node (N) might include this transaction to 
> snapshot (B) while other nodes might exclude it from this snapshot.
> There are some options to handle it:
>  # IncSnpAwareMessage might check previous snapshot, and await it finished.
>  # IncSnpAwareMessage might sign received ID as inconsistent, and after 
> receiving a snapshot request - stop it.



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

Reply via email to