[
https://issues.apache.org/jira/browse/IGNITE-23180?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Philipp Shergalis updated IGNITE-23180:
---------------------------------------
Labels: ignite-3 (was: )
> Fix fsync for rockDB shared log storage
> ---------------------------------------
>
> Key: IGNITE-23180
> URL: https://issues.apache.org/jira/browse/IGNITE-23180
> Project: Ignite
> Issue Type: Improvement
> Components: persistence
> Reporter: Philipp Shergalis
> Assignee: Philipp Shergalis
> Priority: Major
> Labels: ignite-3
>
> If fsync is not set in writeOptions explicitly, rocksdb uses fsync=false,
> ignoring value in DBOptions. We should restore a line setting fsync option in
> write options for *RocksDbSharedLogStorage*
> Without fix:
> {code:java}
> Benchmark (clusterSize) (fsync) (partitionCount)
> (replicaCount) Mode Cnt Score Error Units
> InsertBenchmark.sqlInsertMulti 1 false 4
> 2 avgt 20 373.648 ± 35.654 us/op
> InsertBenchmark.sqlInsertMulti 1 true 4
> 2 avgt 20 395.280 ± 43.948 us/op
> InsertBenchmark.kvInsert 1 false 4
> 2 avgt 20 70.850 ± 9.786 us/op
> InsertBenchmark.kvInsert 1 true 4
> 2 avgt 20 66.360 ± 4.070 us/op{code}
> (results with and without fsync are the similar, because the setting is
> ignored)
> With fix:
> {code:java}
> Benchmark (clusterSize) (fsync) (partitionCount)
> (replicaCount) Mode Cnt Score Error Units
> InsertBenchmark.sqlInsertMulti 1 false 4
> 2 avgt 20 347.610 ± 39.477 us/op
> InsertBenchmark.sqlInsertMulti 1 true 4
> 2 avgt 20 3507.467 ± 2418.805 us/op
> InsertBenchmark.kvInsert 1 false 4
> 2 avgt 20 71.843 ± 19.696 us/op
> InsertBenchmark.kvInsert 1 true 4
> 2 avgt 20 482.806 ± 166.221 us/op{code}
> Results differ
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
