Qless keeps a list of active workers, but until now, heartbeating a running job was not sufficient to keep a worker listed as in the active pool.
We keep the max-worker-age configured to be relatively low (10 minutes) so that we can be made aware quickly when workers disappear. However, if a worker is saturated with jobs that take longer than that interval but are being heartbeated regularly, that presents a false impression of the number of active workers.
Qless keeps a list of active workers, but until now,
heartbeating
a running job was not sufficient to keep a worker listed as in the active pool.We keep the
max-worker-age
configured to be relatively low (10 minutes) so that we can be made aware quickly when workers disappear. However, if a worker is saturated with jobs that take longer than that interval but are being heartbeated regularly, that presents a false impression of the number of active workers.@b4hand @evanbattaglia @benkirzhner @kq2