[ https://issues.apache.org/jira/browse/FLUME-3101?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16591062#comment-16591062
]
Jianqiao.He commented on FLUME-3101:
------------------------------------
hi [~hunhun], after reading TAILDIR source code, i found there is only a single thread accessing
log files whatever how many log files are. I'm considering if it's reasonable to use multiple
threads to do it.
Anything wrong please correct me?
> taildir source may endless loop when tail a file
> ------------------------------------------------
>
> Key: FLUME-3101
> URL: https://issues.apache.org/jira/browse/FLUME-3101
> Project: Flume
> Issue Type: Bug
> Components: Sinks+Sources
> Affects Versions: 1.7.0
> Reporter: hunshenshi
> Priority: Major
> Labels: patch, taildirsource
> Fix For: 1.9.0
>
> Attachments: FLUME-3101-0.patch, FLUME-3101-1.patch, FLUME-3101-2.patch, FLUME-3101-3.patch
>
>
> If there are many files in the path that need to be tail, and there is a file written
by *high frequency* (for example, there are file a, file b and file c in the path, file a
is written at high frequency), *taildir can read the batchSize size event from file a everytime*,
then taildir will only read data from file a, other files will not to be read, because in
TaildirSource.tailFileProcess will into an endless loop.
> code:
> {code:title=TaildirSource.java|borderStyle=solid}
> private void tailFileProcess(TailFile tf, boolean backoffWithoutNL)
> throws IOException, InterruptedException {
> while (true) {
> // if events.size >= batchSize will not break while,
> // then into endless loop to only read tf
> if (events.size() < batchSize) {
> break;
> }
> }
> }
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@flume.apache.org
For additional commands, e-mail: issues-help@flume.apache.org
|