Weiqing Yang created FLINK-41448:
------------------------------------
Summary: UDF metric group name follows how the function is spelled
in the query, so one function can get several groups
Key: FLINK-41448
URL: https://issues.apache.org/jira/browse/FLINK-41448
Project: Flink
Issue Type: Bug
Components: Table SQL / Planner
Affects Versions: 2.4.0
Reporter: Weiqing Yang
Assignee: Weiqing Yang
Fix For: 2.4.0
*What happens*
Function names are case-insensitive in SQL, but the UDF metric group is named
after the spelling used at each call site. The same function written in two
spellings in one query gets two metric groups, each with half of the data.
*How to reproduce* (SQL Client, any scalar UDF)
{code:sql}
SET 'table.exec.udf-metric-enabled' = 'true';
CREATE TEMPORARY SYSTEM FUNCTION MyUpper AS 'repro.Upper';
CREATE TEMPORARY TABLE src (s STRING) WITH ('connector' = 'datagen',
'rows-per-second' = '10');
CREATE TEMPORARY TABLE sink (a STRING, b STRING) WITH ('connector' =
'blackhole');
INSERT INTO sink SELECT myupper(s), MYUPPER(s) FROM src;
{code}
The Calc operator reports two groups, {{udf.myupper}} and {{udf.MYUPPER}}. The
docs describe one group per function, and a dashboard built for one spelling
misses calls made with another.
*Cause*
The metric group name is {{BridgingSqlFunction#getName}}. For a temporary
system function, {{FunctionCatalog}} looks the function up by its normalized
name but builds the identifier from the name as written in the query
({{FunctionIdentifier.of(funcName)}}), so the operator name, and with it the
metric group, follows the call-site spelling.
*Proposed fix*
Name the metric group after the normalized function name
({{FunctionIdentifier#normalizeName}}), so every spelling of one function maps
to one group, and state in the docs that group names are lower-case. This only
changes the metric name, not how the planner names the function anywhere else.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
