[ 
https://issues.apache.org/jira/browse/SOLR-10594?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16005269#comment-16005269
 ] 

Mikhail Khludnev commented on SOLR-10594:
-----------------------------------------

[~noble.paul] 
Could you check the last patch? It should work for two concurrent updates as 
follows: 
- Zookeeper will order updates, assigns versions and send watcher events in 
order. 
- Since {{SolrCore.reload()}} is protected with the own synchronised section, 
cores are reloaded twice in the same correct order. 
Therefore, if we remove reload on GET from {{SolrConfigHandler}} and reloadLock 
from it, we will at least correctly handle two concurrent config updates. 
That's really enough I suppose, and I'm just not sure about ordering between 
the second and the third updates, but it might work too. 
The current code is probably exposed to the risk you mentioned, anyway. And 
it's not really possible to step back to pre 
https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;a=commitdiff;h=cbd3b02
 since that behaviour caused two core reload on single update and breaks a 
plenty of CLI/example tests. It was like this: test applied config change, got 
confirmation and stop the instance, while the second redundant reload occurs. 
And it seems like it's not easy to make reload safe for shutdown.   

Would you mind if I commit the last patch?

> Revise SolrConfigHandler.reloadLock after SolrCloudExampleTest fix
> ------------------------------------------------------------------
>
>                 Key: SOLR-10594
>                 URL: https://issues.apache.org/jira/browse/SOLR-10594
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Mikhail Khludnev
>         Attachments: SOLR-10594 no reload on config GET.patch, 
> SOLR-10594.patch
>
>
> SOLR-10588 fixed {{SolrCloudExampleTest}} by introducing ill dependency 
> {{SolrCore --> SolrConfigHandler.reloadLock}} let's think how we can design 
> it better. And it's also worth to think about overall flow: core reload is 
> triggered by Zk listener and SolrConfigHandler see SOLR-10588



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to