[
https://issues.apache.org/jira/browse/NIFI-7557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17190182#comment-17190182
]
ASF subversion and git services commented on NIFI-7557:
-------------------------------------------------------
Commit fd068fe978da5a4a3e5df14b9a8e43f51e88c606 in nifi's branch
refs/heads/main from Mark Payne
[ https://gitbox.apache.org/repos/asf?p=nifi.git;h=fd068fe ]
NIFI-7557: uses a canonical representation of strings when recovering data from
FlowFile Repository in order to avoid using huge amounts of heap when not
necessary
- Fixed some problems with unit/integration tests
This closes #4507.
Signed-off-by: Bryan Bende <[email protected]>
> Cache large/common FlowFile attributes when restoring FlowFile Repository
> -------------------------------------------------------------------------
>
> Key: NIFI-7557
> URL: https://issues.apache.org/jira/browse/NIFI-7557
> Project: Apache NiFi
> Issue Type: Improvement
> Components: Core Framework
> Reporter: Mark Payne
> Assignee: Mark Payne
> Priority: Major
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> When NiFi is restarted, it restores FlowFiles from the repository. Each
> attribute on a FlowFile is read from disk and put into a HashMap. There are
> times when a Processor will add a large attribute to every FlowFile that it
> sees, and this results in using much more heap upon NiFi restart to store
> FlowFiles than it does while NiFi is running. This is because the Processor
> holds the value of that FlowFile as a single String object and adds that
> String to the HashMap of attributes on every FlowFile.
> However, on restart, NiFi deserializes a byte stream to come up with the
> attribute value. As a result, each FlowFile that has that attribute value
> ends up with its own String object, even though the same value is repeated
> many times.
> As a result, a huge amount of heap may be used on restart, causing NiFi to
> encounter OOME when attempting to restore the FlowFile Repository.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
