[argo-cd] Switch to bitnami/redis and bitnami/redis-cluster chart

[aslafy-z]

Is your feature request related to a problem?

I have some issues with the redis-ha chart. If some pods are destroyed, they don't synchronize back well and I have to delete all the pods and wait for all of them to become ready again.

Related helm chart

argo-cd

Describe the solution you'd like

I feel like this chart should use bitnami maintained charts which are now the "default" for a major part of the community.

See
https://artifacthub.io/packages/helm/bitnami/redis
https://artifacthub.io/packages/helm/bitnami/redis-cluster

Describe alternatives you've considered

No response

Additional context

No response

[gmoirod]

👍
Actually it does not work on Openshift out-of-the-box.
It needs to create RoleBindings and ServiceAccount specific for Redis before.

[mkilchhofer]

The kustomize manifests living in the upstream project over there uses the rendered YAMLs from @DandyDeveloper's chart https://github.com/argoproj/argo-cd/blob/v2.3.3/manifests/ha/base/redis-ha/chart/requirements.yaml

The intent of this helm-repository here is to use the same architecture as the upstream projects (Argo CD, Workflows, etc.). IMHO you should file an issue over there: https://github.com/argoproj/argo-cd/issues/new/choose

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[aslafy-z]

No-stale

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[DandyDeveloper]

@mkilchhofer What's the issue here? Redis clustering vs. Sentinel are very different so this change could impact Argo quite a lot.

Feel free to raise the issue in my chart for Redis, I'm pretty active and try my best to maintain. Right now, I'm the only real active maintainer.

[mkilchhofer]

Hi @DandyDeveloper,

Context: We at @swisspost tried switching on redis-ha in the Argo CD chart. We used it like 1-2 month or so our AWS EKS clusters. We use cluster autoscaling and also upgrade our clusters once a week (new AWS AMI for the workers).

Issue: One problem we saw is that one of the 3 redis pods is unhappy:

$ kubectl logs argocd-server-6499778d-2n56j
(..)
redis: 2022/03/11 10:35:37 pubsub.go:168: redis: discarding bad PubSub connection: EOF
redis: 2022/03/11 10:35:37 pubsub.go:168: redis: discarding bad PubSub connection: EOF
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: EOF
time="2022-03-11T10:35:38Z" level=warning msg="Failed to resync revoked tokens. retrying again in 1 minute: EOF"
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: write tcp 10.116.191.151:46704->172.20.36.205:6379: write: broken pipe
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: write tcp 10.116.191.151:46704->172.20.36.205:6379: write: broken pipe
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: EOF
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: EOF
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: EOF
redis: 2022/03/11 10:35:38 pubsub.go:168: redis: discarding bad PubSub connection: EOF

The redis logs of at least one replica was full of:

kubectl logs argocd-redis-ha-server-1 -c redis
(..)
1:S 11 Mar 2022 07:09:31.312 * Non blocking connect for SYNC fired the event.
1:S 11 Mar 2022 07:09:31.312 * Master replied to PING, replication can continue...
1:S 11 Mar 2022 07:09:31.313 * Partial resynchronization not possible (no cached master)
1:S 11 Mar 2022 07:09:31.313 * Master is currently unable to PSYNC but should be in the future: -NOMASTERLINK Can't SYNC while not connected with my master
1:S 11 Mar 2022 07:09:32.324 * Connecting to MASTER 172.20.145.11:6379
1:S 11 Mar 2022 07:09:32.324 * MASTER <-> REPLICA sync started
1:S 11 Mar 2022 07:09:32.324 * Non blocking connect for SYNC fired the event.
1:S 11 Mar 2022 07:09:32.325 * Master replied to PING, replication can continue...
1:S 11 Mar 2022 07:09:32.325 * Partial resynchronization not possible (no cached master)
1:S 11 Mar 2022 07:09:32.326 * Master is currently unable to PSYNC but should be in the future: -NOMASTERLINK Can't SYNC while not connected with my master
1:S 11 Mar 2022 07:09:33.328 * Connecting to MASTER 172.20.145.11:6379
1:S 11 Mar 2022 07:09:33.328 * MASTER <-> REPLICA sync started
1:S 11 Mar 2022 07:09:33.329 * Non blocking connect for SYNC fired the event.
(..)

Resolution: We then always fixed it like this:

$ kubectl -n argocd delete po -l app=redis-ha
pod "argocd-redis-ha-server-0" deleted
pod "argocd-redis-ha-server-1" deleted
pod "argocd-redis-ha-server-2" deleted

And after 2 monthes of annoying redis issues we switched back to single-replica redis. After that we never faced a redis related issue again.

[DandyDeveloper]

@mkilchhofer How long ago was this?

We had a split brain scenario that was the result of bad Sentinel election. Its been resolved permanently a while back by introducing a pod for checking on this and explicitly solving split brain issues like above.

This is surface level assumption, I'd need more logs from the elected master / cluster state to provide more context.

The latest Argo should include the latest Redis chart, so, I would highly recommend trying this again.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[pierluigilenoci]

🎛️

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[pierluigilenoci]

Why closed???

[pierluigilenoci]

@mkilchhofer ?

[pdeva]

we are seeing this issue too. not sure why it’s been closed.

[DandyDeveloper]

I maintain the redis being used, the problem in question should be resolved long ago.

If people are experiencing problems, throw me a link to the issue or describe the issue so I can investigate.

I believe they closed this because my reply indicated things are fixed and we had no follow up.