Hey there, folks. I changed my flink k8s operator to version 1.6. I was checking the diff and I’ve found this commit. Today I’m defining the pipeline like

flinkStream
        .environment() // StreamExecutionEnvironment
        .fromSource(
            createKafkaSource(),
            WatermarkStrategy.forMonotonousTimestamps(),
            flinkStream.pipelineName() + "-source")
        .uid(flinkStream.pipelineName());

Is setting the

uid

from source enough to be sure the Job will always end up with the same jobId? Sometimes I’m seeing different jobIds and I’m struggling to automatically restore state.

Jobs shouldn't end up with the same jobId; upgrades should be done via a snapshot

Without the operator and using a regular flink deployment in the application mode, deployment would always end up with the same JobId. Why cant they have the same id?

IIRC it's not an idempotent operation

Gotcha. Thank you for that. One question then. When you’re upgrading do you usually stop to upgrade from the. snapshot or do you deploy with a

initialSavepointPath

without stopping the previous one and allowing rolling upgrades from k8s?