Pulling a large image on a paused worker automatically resumes it at the end

[ruimarinho]

Hi!

I think I'm describing the correct trigger for this bug, but the association is definitely there - it seems like at the end of pulling a large OCI image like macos-runner (~218GB compressed), which can take ~20mins in my scenario, a paused worker will automatically resume its work.

I'm guessing some routine inside the pull command is blocking and causing some kind of orchard connection restart, leading to the establishment of a fresh connection which is then likely resetting the 'pause' status (which should be a stored state on the server?).

[edigaryev]

Was the worker paused shortly before, or while pulling the image?

[ruimarinho]

Not while pulling for sure. If I’m not wrong, shortly before pulling, but with list workers showing as paused before proceeding with the pull.