Weiqing Yang created FLINK-40295:
------------------------------------
Summary: Document UDF metrics
Key: FLINK-40295
URL: https://issues.apache.org/jira/browse/FLINK-40295
Project: Flink
Issue Type: Sub-task
Components: Documentation
Reporter: Weiqing Yang
This is the fourth of four sub-tasks implementing FLIP-485.
FLIP:
https://cwiki.apache.org/confluence/spaces/FLINK/pages/373885706/FLIP-485+Add+UDF+Metrics
Vote thread: https://lists.apache.org/thread/symqpswsohl2s5wmtkcw0jjp1w5dot0n
Document the UDF metrics feature in docs/content/docs/ops/metrics.md and its
Chinese
counterpart.
Two additions:
* A UDF section in the metrics reference listing {{udfProcessingTime}} and
{{udfExceptionCount}}, their types, and their operator-scoped naming, where
the full
identifier is {{<operator_name>.udf.<udf_name>.<metric>}}.
* A "UDF metrics" section alongside the existing state-access-latency section,
covering
how to enable the feature with {{table.exec.udf-metric-enabled}}, how
{{table.exec.udf-metric.sample-interval}} trades accuracy against overhead,
the fact
that exceptions are counted on every invocation rather than sampled, which
function
kinds are covered, and the recommendation to use the metrics for debugging.
The overhead wording notes that because only the sampled invocations are timed,
the
remaining invocations cost a single counter increment, so the default interval
keeps the
overhead of the enabled path small. This answers a question raised on the
reference
implementation about the cost of the feature when it is switched on.
Not covered, and stated as such in the docs: process table functions
(PROCESS_TABLE),
aggregate functions, and functions registered through the deprecated
registerFunction
API.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
