Weiqing Yang created FLINK-41447:
------------------------------------
Summary: An invalid table.exec.udf-metric.sample-interval is
accepted at planning and puts the job in a restart loop
Key: FLINK-41447
URL: https://issues.apache.org/jira/browse/FLINK-41447
Project: Flink
Issue Type: Bug
Components: Table SQL / Planner
Affects Versions: 2.4.0
Reporter: Weiqing Yang
Fix For: 2.4.0
*What happens*
With UDF metrics enabled, a {{table.exec.udf-metric.sample-interval}} of 0 or
less is accepted when the query is submitted. The job then fails every time its
operators open, and keeps restarting:
{noformat}
java.lang.IllegalArgumentException: UDF metric sample interval must be >= 1,
but was 0.
{noformat}
*How to reproduce* (SQL Client, any scalar UDF, here {{MyUpper}})
{code:sql}
SET 'table.exec.udf-metric-enabled' = 'true';
SET 'table.exec.udf-metric.sample-interval' = '0';
CREATE TEMPORARY TABLE src (s STRING) WITH ('connector' = 'datagen',
'rows-per-second' = '10');
CREATE TEMPORARY TABLE sink (a STRING) WITH ('connector' = 'blackhole');
INSERT INTO sink SELECT MyUpper(s) FROM src;
{code}
The INSERT is submitted without error. After 30 seconds the job is
{{RESTARTING}} with 8 restarts. With {{table.exec.udf-metric-enabled}} set to
false, the same job runs normally.
*Cause*
The planner reads the interval during code generation
({{BridgingFunctionGenUtil#udfMetricsTermIfEnabled}}) and passes it into the
generated operator without checking it. The check only happens in
{{UdfMetrics#register}}, which runs when the operator opens on the TaskManager.
*Proposed fix*
Validate the interval where the planner reads it, and throw a
{{ValidationException}} with the same message when metrics are enabled and the
value is below 1. The query is then rejected at submission instead of
restarting. Keep the runtime check as a safeguard.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
