Egor Kuts created IGNITE-28205:
----------------------------------
Summary: Rocksdb flush is taking > 1min on node shutdown
Key: IGNITE-28205
URL: https://issues.apache.org/jira/browse/IGNITE-28205
Project: Ignite
Issue Type: Bug
Components: storage engines ai3
Affects Versions: 3.2
Reporter: Egor Kuts
On some TC run cluster failed to shutdown. The stack trace shown a shutdown
hang in RocksDB's native {{closeDatabase()}} method, occurring in the
MetaStorage component during cluster shutdown. Later it printed:
{code:java}
[LoggingRocksDbFlushListener] Finishing rocksdb flush process [name='rocksdb
metastorage kv storage', nodeName='ihaprbfut_tnwflnfctcboams_3348', durat
1051260 ion=61708ms]{code}
Unlike other storages, MetaStorage's DBOptions doesn't use
{code:java}
setAvoidFlushDuringShutdown(true){code}
. Solution - disable flush on shutdown in all storages.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
