[jira] [Commented] (IGNITE-10972) MERGE INTO hangs in MVCC mode with unsorted keys

["Vladimir Ozerov (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-10972?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16747416#comment-16747416 ] 

Vladimir Ozerov commented on IGNITE-10972:
------------------------------------------

[~ilyak], this is incorrect assumption - deadlocks are equally possible in MVCC mode. 

> MERGE INTO hangs in MVCC mode with unsorted keys
> ------------------------------------------------
>
>                 Key: IGNITE-10972
>                 URL: https://issues.apache.org/jira/browse/IGNITE-10972
>             Project: Ignite
>          Issue Type: Bug
>          Components: mvcc, sql
>    Affects Versions: 2.7
>            Reporter: Ilya Kasnacheev
>            Priority: Major
>         Attachments: CacheMvccMergeConflictTest.java
>
>
> right now, if you perform repeatedly
> MERGE INTO T(K, V) VALUES(k1, v1), (k2, v2), (k3, v3);
> and in parallel
> MERGE INTO T(K, V) VALUES(k2, v2), (k1, v1);
> you will eventually see a deadlock. This is expected behavior as per old putAll behavior, but the expectation is that you should see "Cannot serialize transaction" errors instead of deadlock when using MVCC.
> When doing MERGE INTO with sorted keys you will not get deadlock but will see a lot of "Cannot serialize transaction" exception with expectation that such statements to not conflict instead since they are ordered.
> Please see attached test and userlist discussion.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)