axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Rodrigo Ruiz (JIRA)" <>
Subject [jira] Updated: (AXIS-2574) Reading an attachment (slowly) can cause resource deleted error
Date Fri, 20 Apr 2007 07:13:15 GMT


Rodrigo Ruiz updated AXIS-2574:

    Attachment: Call.diff

The problem is that the Call object finalization triggers a chain of finalisations that ends
in the disposal of all the AttachmentPart instances in the response message (among other things).
When finalised, AttachmentPart closes the internal datahandler, and deletes the file associated.
This last step is the source for the error. The file is deleted from the filesystem while
it is being read.

This patch uses a WeakHashMap to keep a hard reference to the Call instance from where the
attachments were obtained. This way, the Call instance will not be garbage collected while
the attachments are being processed.

I am not sure this is an all-cases solution, but it does work for me.

> Reading an attachment (slowly) can cause resource deleted error
> ---------------------------------------------------------------
>                 Key: AXIS-2574
>                 URL:
>             Project: Axis
>          Issue Type: Bug
>          Components: Basic Architecture
>    Affects Versions: 1.2.1
>            Reporter: Steve Sowerby
>         Attachments: Call.diff
> When reading the data from an attachment we periodically get the following error:
> Resource has been deleted.
>  at org.apache.axis.attachments.ManagedMemoryDataSource$
> Having run this throught a debugger and had a brief look at the code it seems to me there
is a race condition of sorts.
> The MemoryManagedDataSource that provides the InputStream has been marked as deleted
by the finalize method of AttachmentPart.
> So basically if the client doesn't read off the attachment fast enough then the writer
will have finished and the AttachmentPart gets finalized and blam, you've got a deleted MemoryManagedDataSource.
> I'm not sure what the best fix is.  Perhaps the deletion should actually be some reference
count rather than a simple boolean.  That way the MemoryManagedDataSource gets deleted once
all the writers and readers are done?  Although perhaps then there would be an issue if the
client was very slow to even open the attachment?

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message