[
https://issues.apache.org/jira/browse/SOLR-15096?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17271647#comment-17271647
]
Mike Drob commented on SOLR-15096:
----------------------------------
Some more specific logging to highlight the differences with a few observations:
{noformat:title=Java 8 RELOAD}
2021-01-25 19:45:10.804 INFO
(OverseerThreadFactory-18-thread-3-processing-n:10.0.0.160:8983_solr) [ ]
o.a.s.c.a.c.OverseerCollectionMessageHandler Executing Collection
Cmd=action=RELOAD, asyncId=null
...
2021-01-25 19:45:10.941 INFO (qtp1448061896-108) [ ] o.a.s.s.HttpSolrCall
[admin] webapp=null path=/admin/cores
params={core=coll-1_shard1_replica_n1&qt=/admin/cores&action=RELOAD&wt=javabin&version=2}
status=0 QTime=133
2021-01-25 19:45:10.944 INFO (qtp1448061896-27) [ ] o.a.s.s.HttpSolrCall
[admin] webapp=null path=/admin/collections params={name=coll-1&action=RELOAD}
status=0 QTime=143
{noformat}
{noformat:title=Java 11 RELOAD}
2021-01-25 19:43:47.073 INFO
(OverseerThreadFactory-18-thread-3-processing-n:10.0.0.160:8983_solr) [ ]
o.a.s.c.a.c.OverseerCollectionMessageHandler Executing Collection
Cmd=action=RELOAD, asyncId=null
...
2021-01-25 19:43:47.221 INFO (qtp1275028674-99) [ ] o.a.s.s.HttpSolrCall
[admin] webapp=null path=/admin/cores
params={core=coll-1_shard1_replica_n1&qt=/admin/cores&action=RELOAD&wt=javabin&version=2}
status=0 QTime=144
2021-01-25 19:43:47.297 INFO (qtp1275028674-28) [ ] o.a.s.s.HttpSolrCall
[admin] webapp=null path=/admin/collections params={name=coll-1&action=RELOAD}
status=0 QTime=274
{noformat}
The time of the *core* reload is pretty close - 133 v 144. I'm willing to call
that within margin of error based on whatever else the OS was doing at the time.
The time of the *collection* reload is more suspicious. With Java 8, we log
that we are executing the cmd 3ms after the timer starts, while with Java 11
the time has already been running for *50ms* by the time we get to the same
point. Perhaps there's something different about hash map lookup or call site
resolution since we're technically using a method reference here. Then, there's
an additional *76ms* pause after the core reload has completed before we
acknowledge the collection reload. Together, these account for almost all of
the performance difference that we observe.
> [REGRESSION] Collection Delete Performance significantly degraded in Java 11
> v 8
> --------------------------------------------------------------------------------
>
> Key: SOLR-15096
> URL: https://issues.apache.org/jira/browse/SOLR-15096
> Project: Solr
> Issue Type: Bug
> Affects Versions: master (9.0)
> Reporter: Mike Drob
> Priority: Blocker
> Fix For: master (9.0)
>
> Attachments: Screen Shot 2021-01-21 at 5.44.25 PM.png
>
>
> While doing some other performance testing I noticed that collection deletion
> in 8.8 (RC1) would take approximately 200ms, while the same operation would
> take 800ms using a recent snapshot ({{a233ed2fd1b}}) from master branch.
> I have not done further investigation at this time.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
