[
https://issues.apache.org/jira/browse/NIFI-3630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15949783#comment-15949783
]
Joseph Witt commented on NIFI-3630:
-----------------------------------
+1 will merge.
Ran some pretty legit before and after testing on this. For a fairly
pathological/taxing flow I am seeing consistently 50% better io utilization
while doing the same amount of work, seeing slightly improved cpu idle, seeing
about 20-30% faster checkpointing of the ff wal. These are all consistent with
what such a change would do. My testing is on an SSD and on a flow not doing
much else other than annoying the ff wal. So on HDDs, etc.. , we could expect
even better gains.
> FlowFileRepository / WriteAhead Log should use a BufferedOutputStream when
> checkpointing
> ----------------------------------------------------------------------------------------
>
> Key: NIFI-3630
> URL: https://issues.apache.org/jira/browse/NIFI-3630
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Reporter: Mark Payne
> Assignee: Joseph Witt
> Fix For: 1.2.0
>
>
> Currently, calls to MinimalLockingWirteAheadLog's checkpoint() method create
> a FileOutputStream, then wrap that in a DataOutputStream and start writing to
> the DataOutputStream, passing that along the SerDe as well. The
> FileOutputStream that it writes to should be wrapped in a
> BufferedOutputStream, as each call to checkpoint() is very taxing on the file
> system currently.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
