Hello, community; it has been a long time since I have had an issue with applying deployments updates. When I update a resource, even if the strategy is the one as default, or even if I specify RollingUpdate like the following snippet

strategy: {
            rollingUpdate: { maxUnavailable: '75%', maxSurge: '75%' },
          },

Pulumi deletes the resources, in this case, all the 3 replicas, and then recreates the 3 replicas. I also tried to use the parameter

deleteBeforeReplace: false,

but nothing changed. The problem is annoying in production since the alerting system always notifies of an issue because the liveness probe results in a HTTP 503 error. Any advice? Many thanks for considering my request.

when you run an

up

what happens? does it recreate the task definition?

I’m running the

up

in cd pipeline, in the up output the changes are always generating a replacement

-- kubernetes:apps/v1:Deployment app1_appDeployment deleting original (0s) 
...
   -- kubernetes:apps/v1:Deployment app1_appDeployment deleted original (0.96s)

The strange behavior is that below in the log it output that is trying a replacement

++ kubernetes:apps/v1:Deployment app1_appDeployment creating replacement (0s) 
   ++ kubernetes:apps/v1:Deployment app1_appDeployment creating replacement (0s) [1/2] Waiting for app ReplicaSet be marked available
  
   ++ kubernetes:apps/v1:Deployment app1_appDeployment creating replacement (0s) warning: [MinimumReplicasUnavailable] Deployment does not have minimum availability.
  
   ++ kubernetes:apps/v1:Deployment app1_appDeployment creating replacement (0s) [1/2] Waiting for app ReplicaSet be marked available (0/3 Pods available)

Then after there is a warning on the initContainer: containers with implemete status

oh, it’s Kubernetes

There is also another resource that is affected by changes but is replaced, not destroyed and recreated

it’s really hard to say I’m afraid, a lot of complexity there

ok thanks, so there isn’t anything I can do to understand why pulumi up is always destroying the resource? I had a look at the replicaset logs, I can say also the replicaset is always deleted and recreated

This sort of K8s error means the deployment is failing - the K8s replicaset controller (I think) is the piece that manages this, and stops deploying new replicas until a new deployment is pushed