You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@geode.apache.org by "Kirk Lund (JIRA)" <ji...@apache.org> on 2018/03/01 22:53:00 UTC

[jira] [Commented] (GEODE-4748) Geode transaction is not consistent in case of networking problems

    [ https://issues.apache.org/jira/browse/GEODE-4748?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16382806#comment-16382806 ] 

Kirk Lund commented on GEODE-4748:
----------------------------------

Thank you for reporting this issue.

Based on your steps to reproduce, this does not seem to be an issue with consistency in the face of networking problems. It looks like it is more likely a specific issue with throwing an exception from toData. Just throwing an exception from toData is not the same as a network or machine failure - in those cases the socket will be closed or not responding, which geode will handle differently. Also when I look at the stack trace that is generated from your steps to reproduce, I can see that there is actually special handling for errors in toData because there is a nested org.apache.geode.ToDataException.

Given that, I'm changing the description and severity of this issue to reflect the fact that this is inconsistency caused by a failure in toData. If you are experiencing issues in the case of actual network or machine failure and you can provide more information on that case please let us know.

FYI geode provides consistency in the case of member and network failures during commit in several ways:

* If a member fails, when it is restarted it will get the latest state from the surviving members, which will include the results of any in progress commit.
* If network connectivity is lost, Geode's heath monitoring in reach member will detract the partition, and members without quorum will shut down.
* If an individual tcp connection fails, geode will recreate the connection and resend the message.

> Geode transaction is not consistent in case of networking problems
> ------------------------------------------------------------------
>
>                 Key: GEODE-4748
>                 URL: https://issues.apache.org/jira/browse/GEODE-4748
>             Project: Geode
>          Issue Type: Bug
>          Components: transactions
>            Reporter: Vadim Lotarev
>            Priority: Critical
>
> Geode cache became inconsistent in case if networking problems occur at commit time. How to reproduce:
> # create any simple _replicated_ region
> # run two nodes
> # put some value in the region within a transaction
> # execute query on both nodes to check that the same value is returned (I used JMX for that)
> # emulate somehow temporary networking error (I changed the code throwing IOException from toData())
> # repeat [#3], exception should occur
> # repeat [#4] - you should see different values on different nodes
> It looks like errors occurred after {{TXState.applyChanges}} produce inconsistency - it is impossible to rollback applied local changes what leads to the state where local cache contains  changed data but other node(s) old data (before changes made in transaction).
> To me, consistency is a key property for the systems like Geode so I would consider this bug as a critical one.



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