qcl, awi-ciroh, leap, 2i2c: gke k8s clusters upgraded to 1.27 #3375
Conversation
consideRatio
changed the title
This comment was marked as resolved.
This comment was marked as resolved.
This comment was marked as outdated.
This comment was marked as outdated.
consideRatio
changed the title
consideRatio
force-pushed
the
pr/upgrade-qcl-awi-ciroh
branch
from
84fc8c9 to
e53a558
Compare
consideRatio
force-pushed
the
pr/upgrade-qcl-awi-ciroh
branch
5 times, most recently
from
1b76fac to
388eae8
Compare
consideRatio
changed the title
| }, | ||
| "n2-highmem-64" : { | ||
| min : 0, | ||
| max : 100, | ||
| machine_type : "n2-highmem-64" | ||
| } | ||
| # FIXME: Remove node pool specific node_version pin when given the chance and no such nodes are running |
There was a problem hiding this comment.
Removing the pin + terraform apply will delete the current node pool and recreate one with the newest k8s version? Or are extra steps needed to upgrade these machine types' k8s version?
There was a problem hiding this comment.
When node_version is removed, it will result in a rolling upgrade. If users pods are active on the node pool's nodes, then they will be terminated or the rolling upgrade will fail.
Due to this, we should wait for no users to be on the node before doing this.
GKE sais its okay that a node pool is 2 minor versions behind the control plane / k8s api-server though, so for now this is okay.
consideRatio
force-pushed
the
pr/upgrade-qcl-awi-ciroh
branch
from
a5c9cf5 to
1edc5b7
Compare
consideRatio
force-pushed
the
pr/upgrade-qcl-awi-ciroh
branch
from
1edc5b7 to
b8090f0
Compare
|
Thank you @GeorgianaElena!! I rebased this on two other PRs merged, but otherwise since your review I only removed a fixme note that you caught to be resolved (b8090f0). |
These clusters were opportunistically upgraded as part of me trying to gain experience to document an upgrade policy for GKE clusters (#3250).
Semi-related extras
388eae8 - I used
deployer generate resource-allocation daemonset-requests <cluster name>to update information about daemonsets' resource requests in clusters as welld1e193d - I made the deploy hubs workflow not redeploy all hubs because of changes to unrelated
generatecommands not used in the workflow