After creating GKE cluster, calling Pulumi Refresh causes a recreate on the cluster/node pools without other changes. about pulumi-gcp HOT 10 CLOSED

Note that in this case, I overwrote the version using pulumi config set to test (set it to a 1.12 version)

from pulumi-gcp.

"Console Upgrade of cluster" in this case is upgrading the cluster from 1.12 to 1.13 in the compute console.

from pulumi-gcp.

I would guess this is related to #88 - which I suspect is also the root cause of #116. If so, we will pick up the fix as part of adopting the (not yet released) 2.4.0 version of the Google Terraform Provider from upstream. See https://github.com/terraform-providers/terraform-provider-google/blob/master/CHANGELOG.md#240-unreleased.

We could better confirm that this is truly the root cause if you could run pulumi preview --logtostderr -v-7 2> out.txt and then find lines that mention replaces= in that output and share them here. This should make clear what properties are actually triggering the replacement in your particular case.

from pulumi-gcp.

pulumi preview --logtostderr -v-7 2> out.txt

I0403 12:07:49.445090 21056 log.go:56] Error closing 'nodejs' language plugin during shutdown; ignoring: 1 error occurred:

• operation not permitted

Not quite the output that I think you were hoping for.

from pulumi-gcp.

These are the resources in red on the output:
gke-cluster replace [diff: -additionalZones,clusterAutoscaling,clusterIpv4Cidr,defaultMaxPodsPerNode,location,masterAuth,nodeConfig,nodeLocations,project,zone

from pulumi-gcp.

Sorry - looks like I had a typo - I meant to say: pulumi preview --logtostderr -v=7 2> out.txt. Could you try that?

Unfortunately, due to pulumi/pulumi#2453, the diff presented in the user output is not reliable in these cases, and so to debug this we need to see what the provider really reported was requiring replacement.

from pulumi-gcp.

Output provided out of band, will let @lukehoban add any relevant output as needed.

from pulumi-gcp.

Confirmed from looking at logs that this is the same issue discussed in #88 (comment).

In particular, in @mfacenet's case - he is seeing:

#replaces=[nodePools]

So only the perceived nodePools change is triggering this as in #88 (comment) (and also related to use of removeDefaultNodePool: true.

from pulumi-gcp.

So interestingly enough, this seems to be related to the empty default-pool being specified, I was able to rectify this specific happening of the issue, however, after running a cluster node upgrade and refresh, pulumi tries to update what is labeled as (minVersion)
~ └─ gcp:container:Cluster gke-cluster update [diff: ~nodeVersion] ~ └─ gcp:container:NodePool corePool update [diff: ~version]

and then fails in the update run itself:
`Updating (repro):

 Type                            Name         Status                  Info
 pulumi:pulumi:Stack             repro-repro  **failed**              1 error
 └─ gcp:compute:Network          repro                                
    └─ gcp:compute:Subnetwork    repro                                

~ └─ gcp:container:Cluster gke-cluster updating failed [diff: ~nodeVersion]; 1 error

Diagnostics:
pulumi:pulumi:Stack (repro-repro):
error: update failed

gcp:container:Cluster (gke-cluster):
error: Plan apply failed: 1 error occurred:

* updating urn:pulumi:repro::repro::gcp:compute/network:Network$gcp:compute/subnetwork:Subnetwork$gcp:container/cluster:Cluster::gke-cluster: node_version was updated but default-pool was not found. To update the version for a non-default pool, use the version attribute on that pool.`

from pulumi-gcp.

Meh, looks like only the masters reflect a minimum version, this is fixed by remove nodeVersion from the nodePools and the master.

from pulumi-gcp.