[Bug]: Cluster removal fails when addons are installed

[mbevc1]

What were you trying to accomplish?

Deleting EKS cluster provisioned with addons.

What happened?

When deleting the cluster first step is cordoning the nodes. EBS addon is running controller replica and is blocking cordoning of the nodes. We should probably be removing the addons first to ensure only workloads are running before cordoning.

How to reproduce it?

Provision EKS cluster with a addon (EBS). Then delete is using eksctl delete cluster -f config.yaml

Logs

2023-02-13 14:36:56 [ℹ]  deleting EKS cluster "mb"
2023-02-13 14:36:57 [ℹ]  will drain 0 unmanaged nodegroup(s) in cluster "mb"
2023-02-13 14:36:57 [ℹ]  starting parallel draining, max in-flight of 1
2023-02-13 14:36:58 [ℹ]  cordon node "ip-192-168-29-32.eu-west-1.compute.internal"
2023-02-13 14:36:58 [ℹ]  cordon node "ip-192-168-46-245.eu-west-1.compute.internal"
2023-02-13 14:36:58 [ℹ]  cordon node "ip-192-168-92-221.eu-west-1.compute.internal"
2023-02-13 14:38:12 [!]  1 pods are unevictable from node ip-192-168-29-32.eu-west-1.compute.internal
2023-02-13 14:39:14 [!]  1 pods are unevictable from node ip-192-168-29-32.eu-west-1.compute.internal
2023-02-13 14:40:16 [!]  1 pods are unevictable from node ip-192-168-29-32.eu-west-1.compute.internal
2023-02-13 14:41:18 [!]  1 pods are unevictable from node ip-192-168-29-32.eu-west-1.compute.internal
2023-02-13 14:42:20 [!]  1 pods are unevictable from node ip-192-168-29-32.eu-west-1.compute.internal

Anything else we need to know?

Versions

$ eksctl info

eksctl version: 0.129.0 kubectl version: v1.26.1 OS: linux

[TiberiuGC]

Hi @mbevc1 ! As per our documentation on how to delete clusters here, Pod Disruption Budget policies are preventing the EBS addon from being properly removed. You should run your command with --disable-nodegroup-eviction flag. i.e.

eksctl delete cluster -f cluster.yaml --disable-nodegroup-eviction

[mbevc1]

Thanks @TiberiuGC for pointing ou this doc note. I guess removing the addon first or in parallel could prevent this issue?

[TiberiuGC]

I think you're right, removing the addon first would probably work as-well.

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[mbevc1]

Any more decisions on this?

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[mbevc1]

We should keep this alive ;)

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[mbevc1]

keep-alive :)

[dabcoder]

I think you're right, removing the addon first would probably work as-well.

Just running into this issue as we speak, so I'd suggest adding a note about removing the add-on on this page: https://eksctl.io/usage/creating-and-managing-clusters/

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[mbevc1]

keep-alive again :)

[AmitBenAmi]

@mbevc1 have you tried to run with --disable-nodegroup-eviction?

[mbevc1]

I haven't tried that one, but looks like it could work. But I reckon we still want to have a graceful and build-in support to deprovision add-ons

[EmpeRoar]

Hi @mbevc1 ! As per our documentation on how to delete clusters here, Pod Disruption Budget policies are preventing the EBS addon from being properly removed. You should run your command with --disable-nodegroup-eviction flag. i.e.

eksctl delete cluster -f cluster.yaml --disable-nodegroup-eviction

this did not work for me. im still having the issue

2023-12-17 04:59:01 [ℹ] deleting EKS cluster "jx-angular-clusterx" 2023-12-17 04:59:04 [ℹ] will drain 2 unmanaged nodegroup(s) in cluster "jx-angular-clusterx" 2023-12-17 04:59:04 [ℹ] starting parallel draining, max in-flight of 1 2023-12-17 04:59:08 [✔] drained all nodes: [ip-172-31-13-6.ec2.internal ip-172-31-38-140.ec2.internal] 2023-12-17 05:00:08 [!] 1 pods are unevictable from node ip-172-31-47-9.ec2.internal 2023-12-17 05:01:09 [!] 1 pods are unevictable from node ip-172-31-47-9.ec2.internal

[github-actions[bot]]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[mbevc1]

keep-alive

[m0un10]

It's probably worth calling out that some addons get added by default for new clusters. So, I don't think there are actually any cases now where the delete will actually work. At least --disable-nodegroup-eviction unblocks though! :)

[yuxiang-zhang]

It's probably worth calling out that some addons get added by default for new clusters.

Indeed, the coredns addon now has a default Pod Disruption Budget (PDB), preventing the pod to be evicted when draining the nodegroups. As a workaround for now, we can use eksctl delete cluster --disable-nodegroup-eviction or eksctl delete nodegroup --disable-eviction.

[LoNeWoLf003]

Keep-alive

[dabcoder]

Just thinking out loud but would it make sense (would it be feasible?) to add a command flag to remove add-ons? e.g. eksctl delete cluster --remove-add-ons