Open fulmicoton opened 5 months ago
Right now we have a readiness and a liveness probe. We also do not have any startup probe.
The readiness probe relies on what is advertise on the cluster.
The liveness probe checks the indexing service and the janitor if any, by ask
them an adhoc message called Healthz
.
A node readiness reporting tasks polls a bunch of conditions and updates the READINESS key of the self cluster state.
As identified on airmail...
(It could be the disk cache loading. We probably need to have two level startup like we do for ingesters.)