I'm having trouble getting this to work. Looks like handle_upload() on the hub remains volume unaware, likewise the checkUpload() call

Forgot debug changes, new commit + rebase.

volume not passed to the checkUpload in fastUpload()

I adjusted that and added a unit test for it here:
https://github.com/mikem23/koji-playground/commits/upload-filers

The change in downloadTaskOutput doesn't seem right. That call should not be accepting absolute pathnames, even with the sanity check.

I think get_upload_path should pass volume through to pathinfo.work() instead of setting topdir.

I've added your commits + get_upload_path fix.

About downloadTaskOutput - not sure if I understand. I've not made any changes to that function, it was already possible to use it that way. I can add volume to its signature which could be tested in absolute pathnames codepath. So, test in assert there could be changed from:

if not fileName.startswith('/'):
    filePath = '%s/%s/%s' % (koji.pathinfo.work(), koji.pathinfo.taskrelpath(taskID), fileName)
else:
    filePath = fileName
    assert(koji.pathinfo.taskrelpath(taskID) in filePath)

to

if not fileName.startswith('/'):
    filePath = '%s/%s/%s' % (koji.pathinfo.work(volume=volume), koji.pathinfo.taskrelpath(taskID), fileName)
else:
    filePath = fileName
    assert(filePath.startswith(koji.pathinfo.task(taskID, volume)

With default volume=None it would be backward-compatible. Other option is to remove else part completely, but it means changing API.

I've removed absolute path support. Also added some more volume-aware calls in various places. Questionable stays:

• BuildNotificationTask - listTaskOutput(all_volumes=True) is not used there, not sure if we want to pollute notification e-mail with full paths on hub - probably not
• WrapperRpmTask - Does it make sense to use it there? I expect that also not, as it tries to put all artifacts to one directory, which simply wouldn't work with multiple files with same name.

WrapperRpmTask - Does it make sense to use it there? I expect that also not, as it tries to put all artifacts to one directory, which simply wouldn't work with multiple files with same name.

WrapperRPMTask explicitly handles that case:

https://pagure.io/koji/blob/master/f/builder/kojid#_1720

You also need to make BaseTaskHandler.localPath() volume-aware:

https://pagure.io/koji/blob/master/f/koji/tasks.py#_296

Is it really needed to change localPath?

Question for wrapperRPM is also if it is semantically correct to have two artifacts with same name (differing only in volume). Should it be allowed (which means spec template logic needs to be changed, as all_artifacts will contain multiple records) or forbidden (raise an exception) or ignored?

Is it really needed to change localPath?

I don't think so. The way it is written it just takes paths relative to topdir.

Now, some of the places this call is used will need to be made volume aware.

Now, some of the places this call is used will need to be made volume aware.

Looks like you hit the important one. For the other two, the cli is not upload a file to a non-default volume, so I think those can wait until that changes.

Small patch to ensure the specified volume exists
https://github.com/mikem23/koji-playground/commits/upload-filers

Apart from that, I think this might be ready to merge

Rebase + your patch added.

Commit f2390dad fixes this pull-request