Open olifre opened 1 month ago
@olifre, this sounds very interesting. Thanks for the pointer.
Some interesting links with further information:
condor_rooster
which can run an unhibernate
command if offline machines are matched: https://htcondor.readthedocs.io/en/latest/admin-manual/configuration-macros.html#condor-rooster-configuration-file-macrosThis sounds interesting indeed and could be reasonably straightforward. We'll need a separate "empty" mode but most tooling should already be there.
As described in this talk: https://indico.cern.ch/event/1386170/contributions/6142517/attachments/2934270/5153418/Kubernetes%20and%20HTCSS%20-%20HTC-EU-24.pdf (slide 9 and 10), it is possible to place "offline ads" and let the HTCondor negotiator annotate these ads if a job would match the "offline drones", which could in turn feed back into COBalD to trigger starting of drones if zero drones are running for the site.
This would allow to re-use the full negotiator functionality with ClassAd evaluation without further effort.