[ 
https://issues.apache.org/jira/browse/IGNITE-6011?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Andrew Mashenkov updated IGNITE-6011:
-------------------------------------
    Description: 
I'd expect that with FULL_SYNC write mode and TX cache data always be 
consistent.
And if EntryProcessor fails on primary (or backup) node and pass on backup (or 
primary) then whole transaction will be rolled back.
But I observe old value on node where EP has failed and new value on other 
nodes.

PFA repro attached.

Looks like we should apply EP on lock phase and fail TX if there is any 
failures.

UPD: We should also check if ignite has an adequate behavior if possible when 
Error (not Exception) occurs, e.g. assert in user code.

  was:
I'd expect that with FULL_SYNC write mode and TX cache data always be 
consistent.
And if EntryProcessor fails on primary (or backup) node and pass on backup (or 
primary) then whole transaction will be rolled back.

But I observe old value on node where EP has failed and new value on other 
nodes.

PFA repro attached.


Looks like we should apply EP on lock phase and fail TX if there is any 
failures.


> EntryProcessor can make data inconsistent if fails on TX cache.
> ---------------------------------------------------------------
>
>                 Key: IGNITE-6011
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6011
>             Project: Ignite
>          Issue Type: Bug
>          Components: cache
>            Reporter: Andrew Mashenkov
>         Attachments: EntryProcessorBug.java
>
>
> I'd expect that with FULL_SYNC write mode and TX cache data always be 
> consistent.
> And if EntryProcessor fails on primary (or backup) node and pass on backup 
> (or primary) then whole transaction will be rolled back.
> But I observe old value on node where EP has failed and new value on other 
> nodes.
> PFA repro attached.
> Looks like we should apply EP on lock phase and fail TX if there is any 
> failures.
> UPD: We should also check if ignite has an adequate behavior if possible when 
> Error (not Exception) occurs, e.g. assert in user code.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to