[jira] [Commented] (IGNITE-10455) MVCC: Tx timeout can cause update counters inconsistency.
[
https://issues.apache.org/jira/browse/IGNITE-10455?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16759710#comment-16759710
]
Ivan Pavlukhin commented on IGNITE-10455:
-
After fixing the problem with active txs leak (linked ticket 8841) the test
reproduces an issue with inconsistent partition sizes when some MOVING
partitions are processed. Should be double checked after fixing broken sizes
(linked ticket 10082).
> MVCC: Tx timeout can cause update counters inconsistency.
> --
>
> Key: IGNITE-10455
> URL: https://issues.apache.org/jira/browse/IGNITE-10455
> Project: Ignite
> Issue Type: Bug
> Components: mvcc
>Reporter: Roman Kondakov
>Assignee: Ivan Pavlukhin
>Priority: Major
> Labels: mvcc_stabilization_stage_1, transactions
> Fix For: 2.8
>
>
> When transaction is rolled back on backup on prepare step, it could lead to
> update counters inconsistency between primary and backup. We need to fix
> backup counters update.
> Reproducer: {{TxWithSmallTimeoutAndContentionOneKeyTest#test}} with enabled
> MVCC.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (IGNITE-10455) MVCC: Tx timeout can cause update counters inconsistency.
[
https://issues.apache.org/jira/browse/IGNITE-10455?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16753930#comment-16753930
]
Ivan Pavlukhin commented on IGNITE-10455:
-
Test in the issue run many transactions with very small tx timeout. As a result
active snapshots leak occurs in {{MvccProcessorImpl}}: transaction is rolled
back when snapshot request is in fly, as a result transaction does not send an
acknowledge that it is completed to MVCC coordinator.
Checking that transaction is still active upon receiving snapshot and sending
an ack to coordinator if it is already aborted can fix the problem.
> MVCC: Tx timeout can cause update counters inconsistency.
> --
>
> Key: IGNITE-10455
> URL: https://issues.apache.org/jira/browse/IGNITE-10455
> Project: Ignite
> Issue Type: Bug
> Components: mvcc
>Reporter: Roman Kondakov
>Assignee: Ivan Pavlukhin
>Priority: Major
> Labels: mvcc_stabilization_stage_1, transactions
> Fix For: 2.8
>
>
> When transaction is rolled back on backup on prepare step, it could lead to
> update counters inconsistency between primary and backup. We need to fix
> backup counters update.
> Reproducer: {{TxWithSmallTimeoutAndContentionOneKeyTest#test}} with enabled
> MVCC.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (IGNITE-10455) MVCC: Tx timeout can cause update counters inconsistency.
[
https://issues.apache.org/jira/browse/IGNITE-10455?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16751121#comment-16751121
]
Ivan Pavlukhin commented on IGNITE-10455:
-
Checked on
https://github.com/apache/ignite/commit/4ea443ceffa4445003aa5c8448017eb430cedff3
Test failure reproduces with clam that hash sum is inconsistent between
partition replicas.
> MVCC: Tx timeout can cause update counters inconsistency.
> --
>
> Key: IGNITE-10455
> URL: https://issues.apache.org/jira/browse/IGNITE-10455
> Project: Ignite
> Issue Type: Bug
> Components: mvcc
>Reporter: Roman Kondakov
>Assignee: Ivan Pavlukhin
>Priority: Major
> Labels: mvcc_stabilization_stage_1, transactions
> Fix For: 2.8
>
>
> When transaction is rolled back on backup on prepare step, it could lead to
> update counters inconsistency between primary and backup. We need to fix
> backup counters update.
> Reproducer: {{TxWithSmallTimeoutAndContentionOneKeyTest#test}} with enabled
> MVCC.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
