[
https://issues.apache.org/jira/browse/IGNITE-1027?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14708944#comment-14708944
]
Semen Boikov commented on IGNITE-1027:
--------------------------------------
>From rebalance code I see that SYNC rebalance is broken if multiple nodes
>start concurrently:
- method 'GridDhtPartitionDemandPool.assign' returns empty assigments if there
are pending exchanges
- DemandWorkers receive empty assigns, finish loop and complete SyncFuture
- Ignite exits from start method before rebalance was really finished
Also DemandWorkers can stop rebalance process and complete SyncFuture if
topology changed during rebalancing (see usages of
DemandWorker.topologyChanged()).
> Possible data loss in replicated cache on unstable topology.
> ------------------------------------------------------------
>
> Key: IGNITE-1027
> URL: https://issues.apache.org/jira/browse/IGNITE-1027
> Project: Ignite
> Issue Type: Bug
> Reporter: Sergi Vladykin
> Assignee: Semen Boikov
> Fix For: ignite-1.4
>
>
> In test IgniteCacheClientQueryReplicatedNodeRestartSelfTest we have 4 data
> nodes with replicated caches and single client-only node, which runs SQL
> queries against those data nodes. Background threads restarting data nodes.
> When we restart 2 of 4 data nodes everything is fine, when 3 of 4 then
> eventually query returns inconsistent result and cache size returns smaller
> values than expected. Since we use rebalance mode SYNC, such a data loss
> should not happen.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
