Stephen Yeager (Feb 28 2024 at 19:21): Stephen Yeager (Feb 28 2024 at 19:21):I've noticed that my dask dashboard often shows fewer dask workers than when I do qstat -u yeager on casper, even after waiting for the dashboard to catch up. I've verified that the workers showing up in qstat are not holdovers from previous calls to Client(). Right now my dashboard shows 3 workers while qstat shows 13. Why the discrepancy?
Negin Sobhani (Feb 28 2024 at 20:08):Hello @Stephen Yeager , Do you specify the number of workers to 13 or 3 workers that the dashboard shows?
Stephen Yeager (Feb 28 2024 at 20:16):I do a cluster.scale(30) and usually get 30 workers. Right now, dashboard shows 3 while qstat shows 13.
Negin Sobhani (Feb 28 2024 at 20:32):I only see one job running for you now (not Dask) but it seems like you must have killed your running Dask cluster.
Anyhow, I would suggest adding something like this when you want to scale up workers to ensure all workers are ready:
# number of workers:
n = 30
# Scale the cluster to n workers
cluster.scale(n)
client.wait_for_workers(n)
Let me know if the issue persists.
Last updated: May 16 2025 at 17:14 UTC