Hello community, am I missing something or is it n...
# troubleshooting
l
Hello community, am I missing something or is it not possible to get a metric with state size per state store and task manager/slot? Currently we are using checkpoint size as proxy but it’s not ideal given there can be data skews on some partitions? Using in memory state
d
With the in-memory, hashmap state backend, I would expect the checkpoint sizes to be an accurate measure of state size. What makes you think it's less than ideal? The checkpoint size metric is reported on a per-subtask level, so any skew will be apparent.
l
Thanks for the response, that's exactly how I expected this metric to be published but I only see checkpoint size metric coming from JM as a total. I'll revisit my metric configs
hey David, I looked into this a bit more; it seems that what I'm after is available in flink 2+, but not in 1.19/1.20 https://nightlies.apache.org/flink/flink-docs-master/docs/ops/metrics/#state-size could you confirm my understanding?
d
The state size metric is new, and more useful. But you can get per-subtask checkpoint size information in older versions of Flink: https://nightlies.apache.org/flink/flink-docs-release-1.20/docs/ops/monitoring/checkpoint_monitoring/#all-subtask-statistics
l
👍 this info is only available in the flink UI, though, right? it is not forwarded to metric reporters?
d
Good question. It's part of the REST API -- https://nightlies.apache.org/flink/flink-docs-release-1.20/docs/ops/rest_api/#jobs-jobid-checkpoints -- but maybe it's not a metric.
gratitude thank you 1