Repository navigation
Upgrade Strategy for Argo Workflow Controller: v2.12 → v4.0.8GoalWe run Argo Workflows as our CI/CD backbone with 10k+ pipeline runs daily. We're upgrading server, client, and CRD from v2.12 to v4.0.8 and have a few questions before proceeding. Setup
Questions1. What is the recommended upgrade strategy for a jump this large (v2.12 → v4.0.8)? 2. Can two workflow controller versions coexist in the same cluster?
3. Is v4 CRD + v2.12 controller viable as a partial rollback? 4. What is the recommended recovery plan once the CRD is upgraded cluster-wide?
This is very customer-impacting. Is there a lighter recovery path the community recommends at this scale? similar open discussion:#15625 |
Replies: 2 comments 3 replies
|
@Ajitesh72 Two CRDs will not isolate the controllers. v2.12 and v4 both register For 10k/day, skip per-run label edits. v2.12 already supports The CRD is still one cluster object. While v2.12 is writing workflows, keep the minimal CRD ( Give each install its own
Keep PNS and the artifact patches in the v2 namespace only. v4 will not run that executor. Port the patches onto emissary before a pipeline moves. |
@Ajitesh72 Two CRDs will not isolate the controllers. v2.12 and v4 both register
argoproj.io/v1alpha1(pkg/apis/workflow/register.go). A CRD under another group is invisible to both binaries unless you rebuild them. @tk1475 already covered that for the rename idea. instanceID does not change it. It is only a label selector on the same type.For 10k/day, skip per-run label edits. v2.12 already supports
--namespaced(cmd/workflow-controller/main.go). Put the patched v2.12 controller in one namespace and v4 in another. Each controller only watches its namespace (one install per namespace). New submits go to the v4 namespace. Rollback is "submit to the v2 namespace again". In-flight workflows…