[
https://issues.apache.org/jira/browse/CXF-6228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14298489#comment-14298489
]
Akitoshi Yoshida commented on CXF-6228:
---------------------------------------
If a file doesn't get deleted, that means someone is holding the stream (which
you have also found out in observing the non-empty stramList). Can you
reproduce this problem is a small example or you get it only sporadically? If
you can reproduce this problem by setting the threshhold to a very small value,
that will help in finding the cause.
> Using XSLTFeature with large messages creates unremovable temporary files
> -------------------------------------------------------------------------
>
> Key: CXF-6228
> URL: https://issues.apache.org/jira/browse/CXF-6228
> Project: CXF
> Issue Type: Bug
> Affects Versions: 3.0.3
> Reporter: Papp István Péter
> Priority: Minor
>
> Using the XSLT transformation feature causes the outgoing message to use a
> CachedOutputStream object. If the size of the message is above the configured
> threshold, a temporary file will be created to store the message on the disk;
> however, this file is not deleted after the message is sent.
> Based on info available on similar problems encountered with
> CachedOutputStream and attachments, I tried using an interceptor in the
> SETUP_ENDING phase to close the stream manually - this, however, did not work
> because even after closing the stream, the "streamList" variable was not
> empty, which resulted in the closing logic skipping file deletion in the
> "maybeDeleteTempFile" method. Attempts to delete the file directly also fail
> for obvious reasons.
> Some (sometimes all, not sure what makes the difference) of the temporary
> files are removed during a clean server shutdown, but since this is a rare
> event in production environments, it does not really solve the problem.
> (Actually, the issue was discovered when the temp files filled the disk
> completely on a production server.)
> The only viable workaround I found so far was to set the CachedOutputStream
> threshold high enough to avoid caching the data to disk altogether.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
