[Condor-devel] RFC: partitionable slots as "per-machine concurrency limits"

[Erik Erlandson <eje@xxxxxxxxxx>]

The Partitionable/Dynamic slot mechanism is a source of some complexity
(and bugs).  It triples the number of slot types we support at various
places in the code.  It complicates the accounting of available
resources for group quotas, and plays funny with weighted slots.  It's
interactions with the ded-schedd are famous.  It increases the
complexity of the claim state transitions, and we maintain machinery for
re-using the dynamic slots that get peeled off because they're expensive
to instantiate.

To my knowledge, the main use case for partitionable slots is to allow
pooling of memory and disk across cpus.  In that respect they function a
lot like a concurrency limit, scoped to a single machine.

Given that we are considering some extensions to the concurrency limit
features, it made me wonder if it would be a net win to dispense with
partitionable/dynamic slots in favor of mem/disk pooling via some kind
of per-machine concurrency limit accounting mechanism.