> I was thinking you can stop the supervisor process. Then kill the running > topologies. Change the storm.yaml to reduce the number of ports (workers). > Then restart the supervisor.
Looks OK to me. The workers will be rescheduled elsewhere by nimbus. When the supervisor returns and its heartbeat is read by nimbus, it will be available again for scheduling with the number of slots you configure. Some schedulers may not anticipate a differing number of slots per node, but I think with the given scheduler it should be fine. You should not need to restart nimbus or any other daemon, and you should not need to change any running topologies. -- Derek ________________________________ From: Brian Fleming <[email protected]> To: [email protected] Sent: Wednesday, September 23, 2015 8:36 PM Subject: How to move jobs off a node What is the correct way to reduce the number of workers on a supervisor without impacting the system? If it must impact the system, then fine, but then still how is it done? My node is running out of resources. I was thinking you can stop the supervisor process. Then kill the running topologies. Change the storm.yaml to reduce the number of ports (workers). Then restart the supervisor. Is there major interruption to topologies? Thank you for any ideas.
