You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@nifi.apache.org by "ASF subversion and git services (Jira)" <ji...@apache.org> on 2021/11/04 20:49:00 UTC

[jira] [Commented] (NIFI-9364) ExecuteStateless performance suffers if using "Content Storage Strategy" of "Store Content on Disk"

    [ https://issues.apache.org/jira/browse/NIFI-9364?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17438929#comment-17438929 ] 

ASF subversion and git services commented on NIFI-9364:
-------------------------------------------------------

Commit 373498445fe589e2d4855a0730fbb9127f0b4452 in nifi's branch refs/heads/main from Mark Payne
[ https://gitbox.apache.org/repos/asf?p=nifi.git;h=3734984 ]

NIFI-9364: Ensure that we delegate calls to write(byte[]) and write(byte[], int, int) to the underlying OutputStream when writing to the file-based content repository for stateless

Signed-off-by: Joe Gresock <jg...@gmail.com>

This closes #5508.


> ExecuteStateless performance suffers if using "Content Storage Strategy" of "Store Content on Disk"
> ---------------------------------------------------------------------------------------------------
>
>                 Key: NIFI-9364
>                 URL: https://issues.apache.org/jira/browse/NIFI-9364
>             Project: Apache NiFi
>          Issue Type: Bug
>          Components: Extensions
>    Affects Versions: 1.15.0
>            Reporter: Mark Payne
>            Assignee: Mark Payne
>            Priority: Major
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> When I tested performance of ExecuteStateless storing content on heap, performance was great. But I changed to store content on disk, expecting performance to drop some. But I saw a HUGE performance drop - about an order of magnitude. It appears to be because when copying the FlowFile content into the processor's Content Repository, the {{ContentOutputStream}} that is used to write content to disk does not override the parent {{FilterOutputStream's write(byte[], int offset, int len)}} method. Instead, it just inherit's the parent's implementation. The parent implementation performs horribly, calling {{void write(int b)}} for each byte in the given buffer. which means we write each byte to the underlying FileOutputStream as an individual operation.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)