[
https://issues.apache.org/jira/browse/IGNITE-12741?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17050261#comment-17050261
]
Alexey Scherbakov commented on IGNITE-12741:
--------------------------------------------
[~avinogradov]
I think it's theoretically possible to do merges on latch sync because it's
bound to topology version.
1. recovery starts on discovery event, not inside the switch (subsequent
recovery does not wait for previous pme-free switch finish, so we already have
some kind of merge here),
I didn't get it. All PME's are processed sequentially by exchange-worker
thread, see
org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture#init
2. pme-free is fast and performs only latch sync (waits for recovery)
I agree. But I think merges can give some advantage if multiple nodes have
failed (for example rack failure) and unifies a code (reduces number of if's).
We have to do only one sync on latch.
This is not an urgent task and requires investigation.
> Allow exchange merges for PME free switch.
> ------------------------------------------
>
> Key: IGNITE-12741
> URL: https://issues.apache.org/jira/browse/IGNITE-12741
> Project: Ignite
> Issue Type: Improvement
> Affects Versions: 2.8
> Reporter: Alexey Scherbakov
> Priority: Major
> Fix For: 2.9
>
>
> Currently exchange merges are disabled if multiple baseline nodes left/failed.
> It's possible to have enabled merges together with enabled optimization.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
