[
https://issues.apache.org/jira/browse/SOLR-2894?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14096322#comment-14096322
]
Hoss Man commented on SOLR-2894:
--------------------------------
bq. If those numbers were reversed, i might guess maybe the discrepency was a
timing issue of some docs being visible on some replicas when the first
facet.pivot query was executed, but where later during the sanity check query –
but i can't imagine how the number of matching odcs would go down (at least not
in any way that wouldn't reproduce reliably)
1) i'm an idiot: _if_ one shard was slow on getting the updates, then it's
possible that shard might be the one consulted on the "verification" request,
and might have fewer matching docs then what the initial request included.
2) looking closer at mark's log (specifically: {{grep -v "path=/select"
~/tmp/pivotfail.log}} to ignore the query requests themselves) I see that after
we build up the whole index, there is definitely some randomized churn from the
SolrCloud test plumbing of adding new replicas, and peer sync, and
RecoveryStrategy.doRecovery taking place *after* the initial pivot query, while
the validation queries (which checks that the counts for each pivot constraint
match what you get with the equivalent "fq" params) are happening.
All of which leads me wondering if it's possible that this test has just
happened to trigger an unrelated bug where perhaps a new replica is getting put
into query rotation before it's compleletely in sync with it's leader?
[~markrmil...@gmail.com], [~thelabdude] ... does that sound like a viable
possibility given the recovery realted messages in mark's pivotfail.log ?
> Implement distributed pivot faceting
> ------------------------------------
>
> Key: SOLR-2894
> URL: https://issues.apache.org/jira/browse/SOLR-2894
> Project: Solr
> Issue Type: Improvement
> Reporter: Erik Hatcher
> Assignee: Hoss Man
> Fix For: 4.9, 5.0
>
> Attachments: SOLR-2894-mincount-minification.patch,
> SOLR-2894-reworked.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch, SOLR-2894.patch,
> SOLR-2894.patch, SOLR-2894.patch, SOLR-2894_cloud_test.patch,
> dateToObject.patch, pivot_mincount_problem.sh, pivotfail.log
>
>
> Following up on SOLR-792, pivot faceting currently only supports
> undistributed mode. Distributed pivot faceting needs to be implemented.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
