I know we already had a talk about that but I want to make it an issue to keep track of it 😊
It would be very good for our trino clusters to be listed as "ready" / "healthy for lb" when 9 of 10 workers are ready, but because of the statefulset not matching its desired replica count lb starts to queue everything although there is an almost ready cluster waiting for queries.
Background: our k8s cluster nodes get patched very often node by node and that leads to very long query queue-times, even though they would not be required because 90 percent of the cluster is up and running
is it possible to define a treshold of pods available or percent availability to make lb decide wether to route queries to the cluster or not? I know that you already get the trinocluster resource and look into the status, maybe thats something the trino operator should manage? If I remember correctly that's the reason why the lb currently doesn't support my case, isn't it?
I know we already had a talk about that but I want to make it an issue to keep track of it 😊
It would be very good for our trino clusters to be listed as "ready" / "healthy for lb" when 9 of 10 workers are ready, but because of the statefulset not matching its desired replica count lb starts to queue everything although there is an almost ready cluster waiting for queries.
Background: our k8s cluster nodes get patched very often node by node and that leads to very long query queue-times, even though they would not be required because 90 percent of the cluster is up and running
is it possible to define a treshold of pods available or percent availability to make lb decide wether to route queries to the cluster or not? I know that you already get the trinocluster resource and look into the status, maybe thats something the trino operator should manage? If I remember correctly that's the reason why the lb currently doesn't support my case, isn't it?