You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Igor Seliverstov (JIRA)" <ji...@apache.org> on 2018/10/24 09:29:00 UTC

[jira] [Commented] (IGNITE-9928) MVCC TX: Late affinity assignment support.

    [ https://issues.apache.org/jira/browse/IGNITE-9928?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16661996#comment-16661996 ] 

Igor Seliverstov commented on IGNITE-9928:
------------------------------------------

Another important point is what we also should support partitions invalidations. That means remote transactions:
 # should track touched partitions and evict CQ/DR caches as soon as corresponding partitions are evicted,
 # should rollback themselves as soon as all involved partitions become invalid
 # notify DHT transactions about partitions invalidations, recalculated mappings and rollbacks of needless transactions

> MVCC TX: Late affinity assignment support.
> ------------------------------------------
>
>                 Key: IGNITE-9928
>                 URL: https://issues.apache.org/jira/browse/IGNITE-9928
>             Project: Ignite
>          Issue Type: Bug
>          Components: mvcc
>            Reporter: Roman Kondakov
>            Assignee: Igor Seliverstov
>            Priority: Major
>             Fix For: 2.7
>
>
> On unstable topology SQL queries are mapped to random partitions in OWNED state (regardless whether it's going to be evicted or not) but updates take into consideration current ideal affinity assignment only (in other words updates are sent to current backups only and don't touch partitions which are going to be evicted) what causes a situation when subsequent reads may not see previous updates. 



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