[
https://issues.apache.org/jira/browse/FLINK-39892?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated FLINK-39892:
-----------------------------------
Labels: pull-request-available (was: )
> Exactly-once kafka sink reports a negative numBytesOut delta after every
> checkpoint
> -----------------------------------------------------------------------------------
>
> Key: FLINK-39892
> URL: https://issues.apache.org/jira/browse/FLINK-39892
> Project: Flink
> Issue Type: Bug
> Components: Connectors / Kafka
> Affects Versions: 2.2.1
> Reporter: Efrat Levitan
> Priority: Major
> Labels: pull-request-available
>
> The following counters
> {{taskmanager.job.task.numBytesOutPerSecond}}
> {{taskmanager.job.task.numBytesOut}}
> {{taskmanager.job.task.operator.numBytesOutPerSecond}}
> {{taskmanager.job.task.operator.numBytesSend}}
> frequently report a value decrement for exactly once kafka sink.
> This is because {{KafkaWriter#initKafkaMetrics}} is called per producer and
> resets {{byteOutMetric}} but not {{latestOutgoingByteTotal}}
> {{KafkaWriter#registerMetricSync}} calculates the delta of
> [outgoingBytesUntilNow -
> latestOutgoingByteTotal|https://github.com/apache/flink-connector-kafka/blob/main/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/sink/KafkaWriter.java#L276-L278]
> so every time a new producer is spawned, {{numBytesOutCounter.inc}} is
> called with a negative delta, (0 - {{latestOutgoingByteTotal}} of the former
> producer)
> Different exporters handle this differently - otel exporter
> [drops|https://github.com/apache/flink/blob/master/flink-metrics/flink-metrics-otel/src/main/java/org/apache/flink/metrics/otel/OpenTelemetryMetricAdapter.java#L66-L73]
> the datapoint with a warn
> While prometheus exporter treats flink counters as
> [gauges|https://github.com/apache/flink/blob/master/flink-metrics/flink-metrics-prometheus/src/main/java/org/apache/flink/metrics/prometheus/AbstractPrometheusReporter.java#L210-L212]
> so the incorrect datapoint is still reported.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
