You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Sergey Chugunov (Jira)" <ji...@apache.org> on 2025/07/29 14:46:00 UTC
[jira] [Updated] (IGNITE-26060) FailureHandler isn't called on TX coordinator node in certain scenarios
[ https://issues.apache.org/jira/browse/IGNITE-26060?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sergey Chugunov updated IGNITE-26060:
-------------------------------------
Description:
Tests implemented for IGNITE-25641 have shown that a server node acting as a TX coordinator may not fail by FailureHandler (FH) when the following conditions are met:
# The node hosts a primary or backup partition involved into the transaction.
# An error occurred during commit phase of the transaction on that node..
At the same time "regular" primary or backup nodes that doesn't serve as TX coordinators correctly call FH and take proper action.
It looks like that leaving the node alive and not calling FH in this situation may lead to a data integrity violation when the same keys may have different values on different nodes.
When need to fix this carefully to avoid calling FH on nodes that act like TX coordinators but don't store any data related to that transaction (e.g. if a transaction started for some reason on an empty server node or it was a thick client node coordinating the tx).
was:
Tests implemented for IGNITE-25641 have shown that a server node acting as a TX coordinator may not fail by FailureHandler (FH) when the following conditions are met:
# The node hosts a primary or backup partition involved into the transaction.
# An error occurred during commit phase of the transaction.
At the same time "regular" primary or backup nodes that doesn't serve as TX coordinators correctly call FH and take proper action.
It looks like that leaving the node alive and not calling FH in this situation may lead to a data integrity violation when the same keys may have different values on different nodes.
When need to fix this carefully to avoid calling FH on nodes that act like TX coordinators but don't store any data related to that transaction (e.g. if a transaction started for some reason on an empty server node or it was a thick client node coordinating the tx).
> FailureHandler isn't called on TX coordinator node in certain scenarios
> -----------------------------------------------------------------------
>
> Key: IGNITE-26060
> URL: https://issues.apache.org/jira/browse/IGNITE-26060
> Project: Ignite
> Issue Type: Bug
> Reporter: Sergey Chugunov
> Priority: Major
>
> Tests implemented for IGNITE-25641 have shown that a server node acting as a TX coordinator may not fail by FailureHandler (FH) when the following conditions are met:
> # The node hosts a primary or backup partition involved into the transaction.
> # An error occurred during commit phase of the transaction on that node..
> At the same time "regular" primary or backup nodes that doesn't serve as TX coordinators correctly call FH and take proper action.
> It looks like that leaving the node alive and not calling FH in this situation may lead to a data integrity violation when the same keys may have different values on different nodes.
> When need to fix this carefully to avoid calling FH on nodes that act like TX coordinators but don't store any data related to that transaction (e.g. if a transaction started for some reason on an empty server node or it was a thick client node coordinating the tx).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)