You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Hoss Man (JIRA)" <ji...@apache.org> on 2014/08/15 00:23:18 UTC
[jira] [Updated] (SOLR-6379) new replicas seem to be nvolved in
queries before they are ready
[ https://issues.apache.org/jira/browse/SOLR-6379?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Hoss Man updated SOLR-6379:
---------------------------
Attachment: SOLR-6379.patch
Attaching a new test that seems to recreate teh general problem w/o involving any pivot code -- just a basic search and check of the numFound count.
the test flow in a nutshell:
* build a random index of at least 500 docs
* pick 20 (simple) random queries and record the numFound for each
* loop 100 times over this "static" collection, executing each of the 20 queries and checking the count
* loop 100 times over the collection again, but this time randomly inject some async "addreplica" calls (max 5 per shard, TODO: randomly "deletereplica" occasionally)
On my system, the first 100 iterations work fine, but in the second iteration, when random addreplica calls are included, i get failures where
Seed never seems to matter: at some point during the second loop, there is always a failure where a search returns 0 docs instead of the expected count.
> new replicas seem to be nvolved in queries before they are ready
> ----------------------------------------------------------------
>
> Key: SOLR-6379
> URL: https://issues.apache.org/jira/browse/SOLR-6379
> Project: Solr
> Issue Type: Bug
> Reporter: Hoss Man
> Attachments: SOLR-6379.patch
>
>
> spin off of SOLR-2894 where sarowe & miller were getting failures from TestCloudPivot that seemed unrelated to any of hte distrib pivot logic itself.
> in particular: adding a call to "waitForThingsToLevelOut" at the start of the test, even before indexing any docs, seemed to work around the problem -- but even if all replicas aren't yet up when the test starts, we should either get a failure when adding docs (ie: no replica hosting the target shard) or queries should only be routed to the replicas that are up and fully caught up with the rest of the collection.
> (NOTE: we're specifically talking about a situation where the set of docs in the collection is static during the query request)
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org