Is jobs = n ignored on local stores?

I have a few TB of data in a local store.
When I dvc pull, dvc creates a crap load1 of jobs and saturates the zfs pool, making the copies take a long time.

I added jobs 10 in .dvc/config, but it seems to have been ignored.

Am I missing something?

  1. A crap load is a common unit of measure equivalent to a whole bunch.
1 Like

The screenshot seems to show checkout part of dvc pull, which is not affected by remote’s jobs, since it is not really dealing with remote there, but with local cache. We don’t have an option to regulate checkout parallelism right now. A possible workaround is to use dvc fetch instead and dvc checkout files/directories as needed.

Adding support for core.checkout_jobs (or cache.checkout_jobs or smth like that) is possible and quite straightforward for dvc pull, so feel free to create an issue and/or even consider contributing.

1 Like