[
https://issues.apache.org/jira/browse/OAK-4069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15735409#comment-15735409
]
Tomek Rękawek commented on OAK-4069:
------------------------------------
First, benchmark results:
{{noformat}}
# ConcurrentWriteReadTest C min 10% 50% 90% max
N
Oak-Mongo (readconcern=majority) 1 78 108 184 492 1430
1061
Oak-Mongo (readconcern=local) 1 87 110 193 491 1244
1029
{{noformat}}
I used the [local replica|https://github.com/trekawek/global-mongo] set with
latency=0. Benchmark invocation:
{{noformat}}
$ java -Druntime=300 -jar oak-run/target/oak-run-1.6-SNAPSHOT.jar benchmark
ConcurrentWriteReadTest Oak-Mongo \
--mongouri
'mongodb://mongo1:27017,mongo2:27018,mongo3:27019/oak?w=majority&readPreference=nearest&readconcernlevel=majority'
{{noformat}}
> Use read concern majority when connected to a replica set
> ---------------------------------------------------------
>
> Key: OAK-4069
> URL: https://issues.apache.org/jira/browse/OAK-4069
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: mongomk
> Reporter: Tomek Rękawek
> Assignee: Tomek Rękawek
> Labels: resilience
> Fix For: 1.6, 1.5.16
>
> Attachments: OAK-4069-2.patch, OAK-4069.patch
>
>
> Mongo 3.2 introduces new query option: {{readConcern}}. It allows to read
> only these changes that have been already committed to the majority of
> secondary instances.
> It prevents stale reads - a situation in which a change has been committed on
> the primary (and read from it), but due to the network partition a new
> primary is elected and the change is rolled back.
> We should use this new option (together with {{w:majority}} implemented in
> OAK-3554) when running Oak on MongoDB replica set.
> References:
> * [Jepsen: MongoDB stale
> reads|https://aphyr.com/posts/322-jepsen-mongodb-stale-reads]
> * [MongoDB documentation: Read Concern
> in|https://docs.mongodb.org/manual/reference/read-concern/]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
