The error you are seeing (

NullPointerException

on

UserIdentity.isEphemeral()

) typically occurs when StarRocks attempts to evaluate a function or expression that requires a user session context in a background process or metadata validation step where that context is missing. In StarRocks, Custom SQL UDFs (including Global UDFs) are currently not supported in partition expressions. Why this is happening: 1. Limited Function Support: StarRocks' "Expression Partitioning" and "Range Partitioning" only support a specific subset of built-in functions (primarily time functions like

date_trunc

,

time_slice

,

from_unixtime

, etc.). 2. Context Missing: When you use a custom SQL UDF in a

PARTITION BY

clause, the StarRocks Frontend (FE) tries to resolve and evaluate that function during the table creation or data loading phase. Because custom UDFs require permission checks and ownership validation (which uses

UserIdentity

), and the partitioning engine often runs these checks in a simplified context, it fails with a

NullPointerException

because it can't find the "user" who is supposed to be running the function. Recommended Solutions: 1. Use Supported Built-in Functions If your UDF performs common time or string manipulations, check if a built-in StarRocks function can replace it. Supported functions for expression partitioning include: *

date_trunc

,

time_slice

*

date_add

,

date_sub

,

datediff

*

cast

,

from_unixtime

,

str2date

2. Transformation at Load Time (ETL) If you need complex logic that only your UDF provides, the standard practice is to: * Add a dedicated column to your table schema (e.g.,

partition_key_col

). * Apply the UDF logic during the data ingestion process (e.g., in the

COLUMNS

clause of a Routine Load, Stream Load, or an

INSERT INTO ... SELECT

statement). * Partition the table by this new column directly. Example: Instead of:

sql
-- This will fail
CREATE TABLE my_table (
    id INT,
    raw_time STRING
) PARTITION BY my_custom_udf(raw_time) ...

Use:

sql
CREATE TABLE my_table (
    id INT,
    raw_time STRING,
    processed_date DATE  -- Column to hold UDF result
) PARTITION BY (processed_date) ...

-- Then during load:
INSERT INTO my_table
SELECT id, raw_time, my_custom_udf(raw_time) FROM source_data;

Summary The

NullPointerException

you encountered is essentially a bug in how StarRocks reports the lack of support for UDFs in partitioning. You should avoid using UDFs in the

PARTITION BY

clause and instead use built-in functions or pre-process the data into a dedicated partitioning column. References • table_design/data_distribution/expression_partitioning.md • table_design/data_distribution/feature-support-data-distribution.md