Does Scylla include an option to automatically remove inaccessible nodes after some time? Otherwise do you think periodically checking the cluster status and executing
nodetool removenode a bad idea?
I don’t want unhandled issues that couldn’t be mitigated automatically and require manual intervention on some nodes to interfere with new nodes joining the cluster…
Thanks in advance
It is recommended to call removenode
But after 72h of a node not available it should be deleted
from the gossip information on its own.
It might interfere if you gonna try and reuse that ip address
I would be careful doing automation based on the gossipinfo, to remove nodes
I would also cross check with the cloud provider, to make sure you aren’t removing a node that was just out of contact for a few minutes.
I see… Is there any way to slightly shorten the expire time?
After briefly looking at the gossiper implementation, it seems like a hard coded value; But I would love to be proven wrong :')
I’m not sure why it’s not configurable
You can suggest it in an issue or even in a Pull request
Ron Harel reacted to your message: