Gang Xie created HDFS-13594:
-------------------------------
Summary: the lock of ShortCircuitCache is hold while close the
ShortCircuitReplica
Key: HDFS-13594
URL: https://issues.apache.org/jira/browse/HDFS-13594
Project: Hadoop HDFS
Issue Type: Improvement
Components: hdfs-client
Affects Versions: 3.0.2
Reporter: Gang Xie
Attachments: no_hdfs.svg
When we are profiling SC read, we find that ShortCircuitCache's lock is a hot
spot. After look into the code, we find that when close BlockReaderLocal, it
tries to trimEvictionMaps, and several ShortCircuitReplicas are closed while
the lock being hold. This slows down the close of the BlockReaderLocal, and
the worse is that it blocks the other allocating of the new
ShortCircuitReplicas.
An idea to avoid this is to close the replica in an async way. I will do a
prototype and get the performance.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org
