[
https://issues.apache.org/jira/browse/FLINK-15585?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Timo Walther updated FLINK-15585:
---------------------------------
Parent: FLINK-25217
Issue Type: Sub-task (was: Improvement)
> Improve function identifier string in plan digest
> -------------------------------------------------
>
> Key: FLINK-15585
> URL: https://issues.apache.org/jira/browse/FLINK-15585
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / Planner
> Reporter: Jark Wu
> Assignee: Timo Walther
> Priority: Major
> Labels: auto-deprioritized-major, auto-unassigned,
> pull-request-available
> Fix For: 1.15.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Currently, we are using {{UserDefinedFunction#functionIdentifier}} as the
> identifier string of UDFs in plan digest, for example:
> {code:java}
> LogicalTableFunctionScan(invocation=[org$apache$flink$table$planner$utils$TableFunc1$8050927803993624f40152a838c98018($2)],
> rowType=...)
> {code}
> However, the result of {{UserDefinedFunction#functionIdentifier}} will change
> if we just add a method in UserDefinedFunction, because it uses Java
> serialization. Then we have to update 60 plan tests which is very annoying.
> In the other hand, displaying the function identifier string in operator name
> in Web UI is verbose to users.
> In order to improve this situation, there are something we can do:
> 1) If the UDF has a catalog function name, we can just use the catalog name
> as the digest. Otherwise, fallback to (2).
> 2) If the UDF doesn't contain fields, we just use the full calss name as the
> digest. Otherwise, fallback to (3).
> 3) Use identifier string which will do the full serialization.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
