This message was deleted.
# generala
adamant-kite-43734
06/14/2023, 5:39 PMThis message was deleted.
c
creamy-pencil-82913
06/14/2023, 5:49 PMfor controller deployments it doesn’t usually make sense to have more than 2, as they are leader-elected so only one is active at a time. Running more doesn’t get you anything.
creamy-pencil-82913
06/14/2023, 5:49 PMCoredns is a bit different in that it actually provides a service to the cluster, and it may make sense to scale out the pods if you find yourself needing to do so to keep up with client DNS load.
creamy-pencil-82913
06/14/2023, 5:50 PMthe short answer is the same as it is to everything in Kubernetes - it depends.
a
ancient-car-21816
06/14/2023, 5:51 PMis it safe to have 2 running? We previously had only 1, but this is on a three-node cluster, and during monthly maintenance windows bringing one instance down (for patching or replacing) resulted in issues
c
creamy-pencil-82913
06/14/2023, 5:52 PM2 of what
creamy-pencil-82913
06/14/2023, 5:52 PMthey all do different things
a
ancient-car-21816
06/14/2023, 5:53 PM2 pods of system deployments, in general? I suppose that's not the right question to ask...
c
creamy-pencil-82913
06/14/2023, 5:53 PMyes its fine to have 2. It is recommended for HA. I wouldn’t go more than that for anything but coredns, and for coredns only if you find that you need to due to load.
creamy-pencil-82913
06/14/2023, 5:55 PMas I said, controllers are leader elected so one of them will sit there doing nothing until the current leader goes away.
creamy-pencil-82913
06/14/2023, 5:55 PMcoredns is not a controller, it hosts a service (cluster dns) so you can actually see a benefit from scaling it out
a
ancient-car-21816
06/14/2023, 5:56 PMthat makes sense - are there docs anywhere for this btw?
ancient-car-21816
06/14/2023, 5:57 PMwe have a 6-node cluster (3 workers, 3 controllers) built in Packer and deployed via terraform rke, and the default number of Pods for system deployments is 1 in some cases:
ancient-car-21816
06/14/2023, 5:58 PMso when we patch monthly (by rebuilding instances in Packer with the latest AMIs and redeploying them one at a time) we frequently encounter errors as nodes are cordoned and drained then replaced, and I suspect part of that is related to there only being 1 pod for some system deployments
ancient-car-21816
06/14/2023, 5:59 PM(thanks very much for your help btw)
c
creamy-pencil-82913
06/14/2023, 6:30 PMI am not aware of any docs that address scaling the individual components of both the cluster and rancher
creamy-pencil-82913
06/14/2023, 6:31 PMyou kind of have to know Kubernetes well enough to understand what role they each play, and why or when you would want to scale them.
a
ancient-car-21816
06/14/2023, 6:32 PMsounds good - we'll change our default for these HA deployments to 2 since 1 seems problematic
2 Views