Hi chris, I think there is no existing method that allows you to customize placing the operator on the specified node.
By the way, I think "predicting the optimal parallelism" of flink jobs is interesting. Currently flink supports autoscaling mechanism, you can find detailed information in this FLIP [1]. [1] https://cwiki.apache.org/confluence/display/FLINK/FLIP-271%3A+Autoscaling Best, Shammon FY On Fri, May 5, 2023 at 11:48 PM John Gerassimou <john.gerassi...@unity3d.com> wrote: > Sorry for the mix-up. I read your message wrong. Please ignore my last > reply. > > On Fri, May 5, 2023 at 11:42 AM John Gerassimou < > john.gerassi...@unity3d.com> wrote: > >> Hi Chris, >> >> You should be able to do this using nodeSelector, or taints and >> tolerations. >> >> >> https://github.com/apache/flink-kubernetes-operator/blob/main/helm/flink-kubernetes-operator/templates/flink-operator.yaml#L55:L61 >> >> Thanks >> John >> >> On Fri, May 5, 2023 at 8:38 AM <chris-weil...@web.de> wrote: >> >>> Hi, >>> >>> is there a way to manually define which node an operator should be >>> placed in, using Kubernetes? >>> >>> To give a bit more context, for my master's thesis, I'm looking into >>> predicting the optimal parallelism degree for a node. To do so, we use a >>> Zero Shot Model, which predicts the latency and throughput for a given >>> query. To increase performance, we need to manually place operators on >>> different nodes in the network and incorporate other learning methods to >>> see the best configuration. >>> >>> Regards, >>> Chris >>> >>
