[
https://issues.apache.org/jira/browse/KAFKA-18610?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Jacot updated KAFKA-18610:
--------------------------------
Parent: (was: KAFKA-14048)
Issue Type: Improvement (was: Sub-task)
> Add flush metrics to runtime
> ----------------------------
>
> Key: KAFKA-18610
> URL: https://issues.apache.org/jira/browse/KAFKA-18610
> Project: Kafka
> Issue Type: Improvement
> Reporter: Jeff Kim
> Assignee: Jeff Kim
> Priority: Major
> Labels: kip-848
>
> Add 3 metrics
> # flush-interval-ms-\{max, p99, p999}: track how long it takes from when a
> batch is first created, to right before writing to disk. The bulk of the time
> will represent the time spent in the timer, waiting to execute the task.
> # records-size-bytes-\{max, p99, p999}: Write events generate records in the
> runtime. Most of the time, the generated records have to be appended
> atomically into the same batch. If any atomic records exceed
> __consumer_offset's `max.message.bytes`, the runtime throws
> RecordTooLargeException.
> # flush-batch-size-\{max, p99, p999}: Track the size of the batch that is
> flushed. Note that multiple atomic records can be batched together. This
> helps identify how large batches become before they are written, or whether
> they hit the max bytes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
