You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Andrew Mashenkov (JIRA)" <ji...@apache.org> on 2018/01/31 16:15:00 UTC
[jira] [Commented] (IGNITE-7591) Backup node crash causes incorrect
transaction failover recovery
[ https://issues.apache.org/jira/browse/IGNITE-7591?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16347075#comment-16347075 ]
Andrew Mashenkov commented on IGNITE-7591:
------------------------------------------
You expect localPeek will return '4' after backup node stop.
I'd think correct way will be
# to ensure thread that set '4' is the last thread that tries to get a lock. Sleep(1s) doen't give such guarantee.
# wait for all transactions finished before calling localPeek. AFAIK, local peek ignores entry locks and may read stale value.
> Backup node crash causes incorrect transaction failover recovery
> ----------------------------------------------------------------
>
> Key: IGNITE-7591
> URL: https://issues.apache.org/jira/browse/IGNITE-7591
> Project: Ignite
> Issue Type: Bug
> Reporter: Alexey Kuznetsov
> Priority: Major
> Attachments: TxBackupFailoverTest.java
>
>
> reproduction:
> Start multiple pessimistic transactions, changing the same key.Prepare message from backup node must be blocked(GridDhtTxPrepareResponse). Then backup node crashes.
> All transactions must finish and keys must be committed.
> But not all keys are committed. Reproducer attached.
> Test is flaky.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)