-
Notifications
You must be signed in to change notification settings - Fork 136
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Request: bulk: give priority to release
requests over stage
requests
#7632
Comments
onnozweers
changed the title
Request: bulk: give priority to release requests over stage requests
Request: bulk: give priority to Aug 6, 2024
release
requests over stage
requests
I see now that other schedulers have already been anticipated: https://github.com/dCache/dcache/blob/b772cd54a1ae45453da86dd87649add89f6e4745/skel/share/defaults/bulk.properties#L50C1-L58C92
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Dear dCache devs,
My impression is, that currently, the bulk service processes requests in the order they come in. The number of threads (concurrently active requests) is limited, so it can happen that a few stage requests are keeping all threads busy, and release (=unpin) requests are not getting through because they remain queued.
So, you might have a situation where, for instance, Atlas is staging a lot, and the stage pools fills up, and the unpin requests are waiting for the stage operations to complete, so they are not getting through, although they would be the solution for the space congestion.
I've had to do some tuning to make this work better:
With 1000 threads, I saw that unpin requests could get through so they could free up space so that staging could continue.
So I'd like to suggest that release requests have priority over stage requests. Release requests take very little time and they free up the necessary space for staging to succeed. If that is too difficult or undesired, it might help a lot to increase the default number of threads. And perhaps while we're at it,
bulk.limits.max-requests-per-user
could have a bigger default too (like 50000 instead of 5000). Atlas would be very happy.An alternative idea might be to reserve some threads for only
release
requests, so they never stall when stage requests are stuck.Cheers,
Onno
The text was updated successfully, but these errors were encountered: