Open
Description
Short description
Due to the way the status plugin receives updates within OPA (an unbuffered channel), a slow Status Service API can delay bundle activation to the point that scaling out an OPA deployment is not possible (instances do not come up ready in time).
Steps To Reproduce
- Implement a Status API that blocks infinitely
- Configure OPA to report status to that API
- Start OPA with a bundle
Expected behavior
Bundle activation (data plane) should be independent of intermittent control plane failures
Additional context
I think switching the existing channel for listening to bundle status updates to a buffered channel with just 5-10 capacity and dropping old status updates would go a long way.
Metadata
Metadata
Assignees
Type
Projects
Status
Planning - v1.4