Server does not correctly close and delete partially uploaded files

[bain3]

If the user aborts the upload in some way, for example by clicking the "Cancel" button during upload, the server does not correctly close the upload pipeline.

Steps to reproduce

1. start the server in dev mode (following instructions in readme)
2. start uploading a large file
3. interrupt upload
4. see that there is a file in the uploads folder (which should be normally deleted on this line)
5. run lsof -p <pid> and see that the file is still open (this blocks automatic deletion by an external script! File does not get actually deleted from fs until the descriptor is closed...)
6. be sad :(

[bain3]

Ok I managed to create a hotfix for our instance. .pipe() does not properly propagate errors, and on top of that, the ws stream is not properly closed.

https://git.nolog.cz/NoLog.cz/send/commit/f7e889b875538d57ec95f0a5be7a5f7ca25e16db

[bain3]

If you've been running send for a long time with a bit of traffic, then I recommend running fsck on the filesystem where you store files. This reclaimed multiple hundreds of gigs of disk space for us. I assume ext4 did not like all the open but deleted files.

[timvisee]

The general recommendation is to set up an S3 lifecycle configuration. It can be used to clean orphaned files.

Example: https://github.com/timvisee/send/blob/master/s3-lifecycle-example.xml

[bain3]

We did recently switch to S3, which works fine, but this still is a bug when using local fs as the backend.

Send does not properly close the file when an upload fails. Even if something is automatically deleting them externally,
the underlying fs cannot actually delete the file, since it is still opened by the send's node process.

I would also expect send to delete incomplete uploads, but that is easily solvable if you have a cronjob running every night. The main issue is that the files are kept open.