< Rocky> Can we run alter table with new path for external t StarRocks #questions-and-troubleshooting

<@U0A71G31CDV> Can we run alter table with new pat...

Akhilesh Tripathi

02/28/2026, 8:59 PM

@Rocky Can we run alter table with new path for external tables? Eg.

Copy code

CREATE EXTERNAL TABLE IF NOT EXISTS external_table_name (

) 
ENGINE = file
PROPERTIES (
    "path" = "{GCS_PATH}",
    "format" = "parquet"
);

Rocky

02/28/2026, 8:59 PM

Hello! I'm Rocky. For external tables using ENGINE = file
, StarRocks does not currently support modifying the

"path"

property (or most other properties) via an

ALTER TABLE

statement. Since it is an external table, the data resides in your cloud storage (GCS) and StarRocks only stores the metadata (the schema and the pointer to the path). Dropping the table in StarRocks will not delete your actual data files in GCS. Recommended Approach To "update" the path, you should drop the existing table and recreate it with the new path: 1. Drop the table metadata:

Copy code

sql
   DROP TABLE IF EXISTS external_table_name;

2. Recreate the table with the new path:

Copy code

sql
   CREATE EXTERNAL TABLE external_table_name (
       -- column definitions
   )
   ENGINE = file
   PROPERTIES (
       "path" = "<gs://new_gcs_path/data/>",
       "format" = "parquet"
   );

A Few Tips: * Wildcards: If you want to include multiple files or subdirectories, you can often use wildcards in the path (e.g.,

"path" = "<gs://bucket/folder/*>"

). * External Catalogs: If you are managing a large number of tables or the paths change frequently, consider using External Catalogs (like Hive or Iceberg catalogs) which automatically synchronize table metadata, or use the newer

FILES()

table function for ad-hoc queries without creating a table at all. Let me know if you have any other questions!

Akhilesh Tripathi

02/28/2026, 9:06 PM

@Rocky what about List[Paths] in definition instead of wildcards?

Rocky

02/28/2026, 9:06 PM

Hi there! I'm Rocky. While I'm having a little trouble reaching my documentation index right now, I can certainly help answer this based on StarRocks' standard behavior for loading data and external tables. In StarRocks, if you want to specify a specific set of files instead of using wildcards (like

), the approach depends on which method you are using: 1. Using the
FILES()
Table Function (Recommended) When using the

FILES()

table function (often used for

INSERT INTO ... SELECT

or ad-hoc queries from S3, GCS, or Azure Storage), you can provide a comma-separated list of paths within the

path

parameter.

Copy code

sql
SELECT * FROM FILES(
    "path" = "<s3://my-bucket/data/file1.parquet>, <s3://my-bucket/data/file2.parquet>",
    "format" = "parquet"
);

Note: Ensure there are no spaces between the comma and the next path unless they are part of the file name. 2. Using Broker Load If you are using a Broker Load (the standard asynchronous load method), you have two options for specifying multiple specific files: * Multiple paths in one `DATA INFILE`:

Copy code

sql
    LOAD LABEL my_db.label_1
    (
        DATA INFILE("<s3://bucket/file1.csv>", "<s3://bucket/file2.csv>")
        INTO TABLE my_table
    )
    WITH BROKER my_broker (...);

* Multiple
DATA INFILE
clauses:

Copy code

sql
    LOAD LABEL my_db.label_2
    (
        DATA INFILE("<s3://bucket/file1.csv>") INTO TABLE my_table,
        DATA INFILE("<s3://bucket/file2.csv>") INTO TABLE my_table
    )
    WITH BROKER my_broker (...);

3. External Tables For most External Tables (like those pointing to HDFS or S3), the definition usually points to a directory rather than a list of specific files. StarRocks then scans that directory. If you need to limit the query to specific files, it is usually handled via: * Partitioning (to limit which directories are scanned). * Using the

FILES()

function instead of a formal External Table if the file list changes frequently. Summary: While StarRocks doesn't use a literal

List[Paths]

object (like a JSON array), you can achieve the same result using a comma-separated string in the

FILES()

function or by listing multiple strings in a

DATA INFILE

clause for Broker Loads. Does one of these methods fit the specific workflow you're building?

Akhilesh Tripathi

03/01/2026, 3:31 PM

@Rocky getting following error -

Copy code

SELECT 
eventinfo
FROM FILES(
    "path" = "<gs://gcs-one-dsp-prod-eus-commons-adrr/prod/2026/03/01/01/adserv_servingfills_00-*.parquet>",
    "format" = "parquet"
)
LIMIT 10;

SQL Error [1064] [42000]: Apache Arrow type (nested) list does not match the type STRUCT in StarRocks: BE:10024

Any idea what's going one?

Rocky

03/01/2026, 3:31 PM

Hi there! Rocky here. It looks like you're hitting a type mismatch between how the Parquet file stores the

eventinfo

column and how StarRocks is inferring its schema. The error

Apache Arrow type (nested) list does not match the type STRUCT

typically happens when the

FILES

table function infers that a column should be a

STRUCT

, but the actual data in the Parquet file is structured as a

LIST

(Array), or vice versa. This often happens with complex nested data where Parquet uses the "3-level list" encoding. Here are a few steps to diagnose and fix this: 1. Check the Inferred Schema First, let's see what StarRocks thinks the schema is. Run this command:

Copy code

sql
DESCRIBE SELECT * FROM FILES(
    "path" = "<gs://gcs-one-dsp-prod-eus-commons-adrr/prod/2026/03/01/01/adserv_servingfills_00-*.parquet>",
    "format" = "parquet"
);

Look for the

eventinfo

column. If it shows as

STRUCT

but your data is actually an array of items, that's the conflict. 2. Verify the Parquet File Schema If you have access to

parquet-tools

or a tool like DuckDB, check the physical schema of the file. You want to see if

eventinfo

is defined as a

repeated

group (which is a list/array) or a

required/optional

group (which is a struct). 3. Explicitly Define the Schema (Recommended Fix) If the automatic inference is getting it wrong, you can manually define the columns and their types in the

FILES

function to match the Parquet file's actual structure. For example, if

eventinfo

is actually an ARRAY of STRUCTs, you should define it like this:

Copy code

sql
SELECT eventinfo
FROM FILES(
    "path" = "<gs://gcs-one-dsp-prod-eus-commons-adrr/prod/2026/03/01/01/adserv_servingfills_00-*.parquet>",
    "format" = "parquet",
    "columns" = "eventinfo ARRAY<STRUCT<field1 INT, field2 STRING>>" -- Adjust fields to match your actual data
)
LIMIT 10;

4. Check for Schema Evolution If you are querying a glob pattern (

adserv_servingfills_00-*.parquet

), it's possible that some files have

eventinfo

as a Struct and others as a List. StarRocks'

FILES

function infers the schema from a sample of files. If the files are inconsistent, you'll need to ensure the schema matches across all files or define a schema that is compatible. Quick Tip: If you just want to get the data moving and don't need to parse the nested fields immediately, you can try loading it as a

JSON

type if you are using an older version, though defining the correct

ARRAY

STRUCT

is the preferred way in newer StarRocks versions. Let me know if the

DESCRIBE

output shows anything unexpected!

Open in Slack

Previous Next