zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Brian Nixon (JIRA)" <j...@apache.org>
Subject [jira] [Created] (ZOOKEEPER-3311) Allow a delay to the transaction log flush
Date Thu, 14 Mar 2019 00:05:00 GMT
Brian Nixon created ZOOKEEPER-3311:

             Summary: Allow a delay to the transaction log flush 
                 Key: ZOOKEEPER-3311
                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-3311
             Project: ZooKeeper
          Issue Type: New Feature
          Components: server
    Affects Versions: 3.6.0
            Reporter: Brian Nixon

The SyncRequestProcessor flushes writes to disk either when 1000 writes are pending to be
flushed or when the processor fails to retrieve another write from its incoming queue. The
"flush when queue empty" condition operates poorly under many workloads as it can quickly
degrade into flushing after every write -- losing all benefits of batching and leading to
a continuous stream of flushes + fsyncs which overwhelm the underlying disk.
A configurable flush delay would ensure flushes do not happen more frequently than once every
X milliseconds. This can be used in-place of or jointly with batch size triggered flushes.

This message was sent by Atlassian JIRA

View raw message