[ 
https://issues.apache.org/jira/browse/SOLR-10605?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15996885#comment-15996885
 ] 

Erick Erickson commented on SOLR-10605:
---------------------------------------

The spin in SolrCloudExampleTest is better than what used to be there, but not 
all that great. It handles insuring that each replica's view of solrconfig.xml 
is consistent rather than just seeing that ZK got the right thing, but there's 
still a weakness here if _two_ reloads are happening. The first succeeds and 
the property changes, then the second one comes along and interferes with, say, 
a subsequent delete. I'd vastly prefer option <2>, let's not return until the 
operation is done if at all possible.



> TestSolrCLIRunExample.testInteractiveSolrCloudExample() ObjectTracker failure
> -----------------------------------------------------------------------------
>
>                 Key: SOLR-10605
>                 URL: https://issues.apache.org/jira/browse/SOLR-10605
>             Project: Solr
>          Issue Type: Bug
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Mikhail Khludnev
>
> This case arose during beasting TestSolrCLIRunExample under SOLR-9867.
> h3. The race is
> * {{RunExampleTool.runCloudExample()}} updates softCommit time config
> * but it waits only for Zk change, but not till cores are reloaded. 
> * hence test goes further and deletes collection that breaks a core which 
> reload the core. see [logs 
> excerpt|https://issues.apache.org/jira/browse/SOLR-9867?focusedCommentId=15995734&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15995734]
> h3. The possible measures are 
> * probably {{RunExampleTool.runCloudExample()}} should spin wait until cores' 
> config has changed like it's done in 
> {{SolrCloudExampleTest.doTestConfigUpdate()}}
> * Core UNLOAD might wait to accuire {{SolrConfigHandler.reloadLock}}, which I 
> should be polished itself under SOLR-10594 and that similar to what's done 
> under SOLR-9867, where {{SolrCores.close()}} waits until all cores are loaded 
> (not reloaded).
> Which of these do you like more?  
>   



--
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