I’ve playing around on the GCP 2i2c hub and had my kernel die.
Unfortunately the dask-gateway clusters are still live. Can someone kill them please?
This is surprising to me. Usually when my kernel dies (on HPC, dask_jobqueue), I find that the dask cluster is killed, and all jobs get cancelled.
Most likely that Dask Gateway is deployed with an idle timeout and so the cluster will close on its own if no tasks are active.
If necessary, you can use the gateway client and call stop_cluster.
This is surprising to me. Usually when my kernel dies (on HPC, dask_jobqueue), I find that the dask cluster is killed, and all jobs get cancelled.
dask-gateway does register a handler to close clusters when the Python process is existing (source). But if the process closely abruptly, from a segfault, say, the atexit handlers might not be run.
Thanks! Turns out they did shutdown, soo the timeout worked.