[
https://issues.apache.org/jira/browse/CAMEL-8421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14345652#comment-14345652
]
Jyrki Ruuskanen commented on CAMEL-8421:
----------------------------------------
Well, I created a new strategy FileOlderThanExclusiveReadLockStrategy which
extends MarkerFileExclusiveReadLockStrategy. I didn't get far until it started
to look wrong, the code is just too close to
FileChangedExclusiveReadLockStrategy.
Then extending FileChangedExclusiveReadLockStrategy came to mind, but having
basically all code in one method and the fields private doesn't really give the
subclass any advantage.
So, I came back to the conclusion that it's best to just add the readLockMinAge
option to FileChangedExclusiveReadLockStrategy. I updated the PR with some
polish and more tolerance in the unit tests. Unit tests for File run fine, but
FTP test support still won't work for me (mock endpoints just won't receive
anything).
> Add minimum age option to readLock=changed
> ------------------------------------------
>
> Key: CAMEL-8421
> URL: https://issues.apache.org/jira/browse/CAMEL-8421
> Project: Camel
> Issue Type: Improvement
> Components: camel-core, camel-ftp
> Reporter: Jyrki Ruuskanen
> Assignee: Claus Ibsen
> Priority: Minor
> Fix For: Future
>
>
> I'm a fan of noop=true in file consumers since it means I don't have to worry
> about how many readers I have and where. But eventually I came across a
> scenario where current features are not sufficient.
> Let's say we have a source system which writes files with name
> <timestamp>_something.xml, and it won't use temp files or .done marker files
> or anything like that. We want to get the latest file as soon as it's
> created. Consider the following route:
> {code}
> from("file:////somewhere/data?noop=true&include=.*_something[.]xml&readLock=changed&sortBy=file:name")
> .aggregate(constant(true), new
> UseLatestAggregationStrategy()).completionFromBatchConsumer()
> .to("amq:topic:something");
> {code}
> When this route is started it will go through the files in order and get the
> last one. Then it will wait for new files. This works fine as long as the
> writer is not "slow".
> Now, we had cases of incomplete files being read and I was requested to not
> to read the file before it is 10 minutes old, just in case. If I increase
> readLockCheckInterval to 10 minutes getting to the latest file at route
> startup will take close to forever. The current readLock=changed
> implementation always sleeps for at least one readLockCheckInterval per file.
> If we had readLockMinAge option to define the minimum age for the target file
> the consumer could acquire readLock on the first poll and breeze through the
> files until too young a file is reached.
> The route below would poll a file every 500ms (default poll delay), while the
> current readLock=changed would take 1500ms (default poll delay + default
> readLockCheckInterval) per file. Consumer goes through the files until it
> hits the end and gets the last one as soon as it becomes old enough.
> {code}
> from("file:////somewhere/data?noop=true&include=.*_something[.]xml&readLock=changed&readLockMinAge=600000&sortBy=file:name")
> .aggregate(constant(true), new
> UseLatestAggregationStrategy()).completionFromBatchConsumer()
> .to("amq:topic:something");
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
