Feedback/beta release for TEP-0135: Coscheduling PipelineRun pods

[lbernick]

Feature looking for Promotion

Promote the "pipelineruns" and "isolate-pipelineruns" values for the "coschedule" feature flag to "beta". These features allow scheduling all of a pipelinerun's pods to the same node, allowing users to bind multiple PVC-based workspaces to the same taskrun. The "isolate-pipelineruns" value allows users to require 1-1 mappings between nodes and pipelineruns (useful for clusters with autoscaling enabled, and where pipelineruns are deleted after completion).

To try out this feature, set "disable-affinity-assistant" to "true" and "coschedule" to the desired value. Docs here.

Current Feature Version: alpha
Raise Feature Version: beta

Pipeline Release

v0.51

Issues Reported by Users

• Consider PVC access mode in affinity assistant #6989
• Mark PipelineRun completed after binded PVCs are deleted #6985
• Deprecate and remove disable-affinity-assistant feature flag #6988
• Consider renaming the feature in documentation to "coscheduling" rather than "affinity assistant" to focus more on the desired outcome and less on implementation details
• Cluster Autoscaler conflict with volumneClaim and or affinity-assistant #4699
• TEP-0135 isolating pipelineruns and multitenancy #7095

#5776 is related, but this issue only applies to the "workspaces" mode of the affinity assistant, since PVCs are auto-deleted in the "pipelineruns" and "isolate-pipelineruns" mode.

This feature may address the use cases described in #3440, so we may want to close out that issue if this feature is sufficient.

Usage in Dogfooding Cluster

None yet

Examples/Tests

• TestAffinityAssistantPerPipelineRun