Hi Team, We are using flink kubernetes operator 1....
# troubleshooting
s
Hi Team, We are using flink kubernetes operator 1.10 on EKS . I have flink job v1.18 reading kafka topic with 50 partitions with initial task slot 1 . After getting 2-3k records/sec in the kafka source, it scales up to 50 task slots/parallelism. However after getting very few records/no records after 30-90 minutes, there is no scale down at all. Operator logs shows below ''`All vertex processing rates are within target.`
All job vertices are currently running at their target parallelism'.
After I do job restart, job scales up to 50 and does scale down to 1. Message in logs after restart :
pendingRecords metric for ***** could not be found. Either a legacy source or an idle source. Assuming no pending records.
Copy code
job.autoscaler.enabled: "true"
    job.autoscaler.scaling.enabled: "true"
    job.autoscaler.stabilization.interval: 1m
    job.autoscaler.metrics.window: 5m
    job.autoscaler.utilization.target: "0.6"
    job.autoscaler.target.utilization.boundary: "0.2"
    job.autoscaler.restart.time: 2m
    job.autoscaler.catch-up.duration: 2m
    pipeline.max-parallelism: "720"
    #enable in-place scaling
    jobmanager.scheduler: adaptive
    job.autoscaler.scale-up.grace-period: 2m
    job.autoscaler.scale-down.grace-period: 1m
    job.autoscaler.scale-down.interval: "2m"
a
Hi @Sushant, Did you figure out the root cause and solution? I am facing the same issue, no scale down in my flink session jobs even with 0 busy rate. I have same issue with both flink autoscaler 1.9.0 and 1.10.0.
➕ 1