Productize Load Shedding for KC22

Ideas:

Confirm with Quarkus team
Confirm with CND team
Try to sort out non-blocking probes first
Keep defaults of unlimited queue in KC22, add a sensible default (1000?) in KC23
Add a CLI options and documentation for Keycloak to configure load shedding (queue length)
Backport changes to KC22
- Provide an HTTP 503 response when the thread pool rejected the request keycloak#21963
- Avoid direct access to executor to record worker pool metrics keycloak#22743

Load shedding will fundamentally change once Keycloak adopts RESTEasy reactive, see keycloak/keycloak#15728

Do we still see the need to backport it to KC22?

Probably yes, as we've seen failures in situations where Keycloak blocks, for example when JGroup communications is down / a node is suspected.

Created a PR keycloak/keycloak#23343 to discuss with the CND team. This would work in a similar way also for KC22.

With PR keycloak/keycloak#24684 being merged, this is now implemented for KC23.

Navigation Menu