[
https://issues.apache.org/jira/browse/IGNITE-8676?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitriy Pavlov updated IGNITE-8676:
-----------------------------------
Fix Version/s: (was: 2.6)
2.7
> Possible data loss after stoping/starting several nodes at the same time
> ------------------------------------------------------------------------
>
> Key: IGNITE-8676
> URL: https://issues.apache.org/jira/browse/IGNITE-8676
> Project: Ignite
> Issue Type: Bug
> Components: persistence
> Affects Versions: 2.4
> Reporter: Andrey Aleksandrov
> Assignee: Stanislav Lukyanov
> Priority: Critical
> Fix For: 2.7
>
> Attachments: DataLossTest.zip, Ignite8676Test.java,
> image-2018-06-01-12-34-54-320.png, image-2018-06-01-13-12-47-218.png,
> image-2018-06-01-13-15-17-437.png
>
>
> Steps to reproduce:
> 1)Start 3 data (DN1, DN2, DN3) nodes with the configuration that contains the
> cache with node filter for only these three nodes and 1 backup. (see
> configuration from attachment)
> 2)Activate the cluster. Now you should have 3 nodes in BLT
> 3)Start new server node (SN). Now you should have 3 nodes in BLT and 1 node
> not in the baseline.
> 4)Using some node load about 10000 (or more) entities into the cache.
> 5)Start that number of primary partitions equals to backup partitions.
> !image-2018-06-01-12-34-54-320.png!
> 6)Now stop DN3 and SN. After that start them at the same time.
> 7)When DN3 and SN will be online, check that number of primary partitions
> (PN) equals to backup partitions (BN).
> 7.1)In a case if PN == BN => go to step 6)
> 7.2)In a case if PN != BN => go to step 8)
>
> !image-2018-06-01-13-12-47-218.png!
> 8)Deactivate the cluster with control.sh.
> 9)Activate the cluster with control.sh.
> Not you should see the data loss.
> !image-2018-06-01-13-15-17-437.png!
> Notes:
> 1)Stops/Starts should be done at the same time
> 2)Consistent Ids for nodes should be constant.
> Not you should see the data loss.
> Also, I provide the reproducer that often possible to reproduce this issue
> (not always). Free the working directory and restart reproducer in case if
> there is no data loss in this iteration.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
