Open major opened 4 years ago
If I delete one manually from the /var/lib/osbuild-composer/artifacts
and kill osbuild-composer, I can delete composes again.
I seriously doubt we can do anything once the disk is already full. We could possibly avoid that from happening hough...?
I wonder if we can avoid making the temp file for the atomic write in this situation? Or perhaps we can stop accepting data when there's less than 50MB left or something like that?
I wonder if we can avoid making the temp file for the atomic write in this situation?
I don't think this would be a good solution. The way it is now, we at least keep the state consistent and tell the client that deleting the compose did not work. Inconsistent state would (potentially) break all future access to composer, even after disk space is freed.
Or perhaps we can stop accepting data when there's less than 50MB left or something like that?
I like this. I think @dvdhrm and @teg were talking about this in the context of the osbuild store before.
Note that the koji API doesn't write state when deleting a compose.
I think if we run out of space, all bets are off anyway, so not sure it makes sense to put too much effort into mitigating this... Other services will likely also randomly fail.
I would expect the user to notice the low disk situation via the cockpit dashboard. But it might also be good to require a minimum of free space available when starting a compose since we know, approximately, how much space will be needed. We should encourage users (other than us running tests
I was doing some testing in AWS and the disk filled up from composes being uploaded from the worker to the composer instance. I tried to delete some composes to save space, but all calls for deletions failed. The following appeared in the system journal: