Import Apache Parquet Files from Amazon S3 or GCS into TiDB Cloud

You can import both uncompressed and Snappy compressed Apache Parquet format data files to TiDB Cloud. This document describes how to import Parquet files from Amazon Simple Storage Service (Amazon S3) or Google Cloud Storage (GCS) into TiDB Cloud.

Step 1. Prepare the Parquet files

1. If a Parquet file is larger than 256 MB, consider splitting it into smaller files, each with a size around 256 MB.

   TiDB Cloud supports importing very large Parquet files but performs best with multiple input files around 256 MB in size. This is because TiDB Cloud can process multiple files in parallel, which can greatly improve the import speed.

2. Name the Parquet files as follows:

   • If a Parquet file contains all data of an entire table, name the file in the ${db_name}.${table_name}.parquet format, which maps to the ${db_name}.${table_name} table when you import the data.
   • If the data of one table is separated into multiple Parquet files, append a numeric suffix to these Parquet files. For example, ${db_name}.${table_name}.000001.parquet and ${db_name}.${table_name}.000002.parquet. The numeric suffixes can be inconsecutive but must be in ascending order. You also need to add extra zeros before the number to ensure all the suffixes are in the same length.
   Note

   If you cannot update the Parquet filenames according to the preceding rules in some cases (for example, the Parquet file links are also used by your other programs), you can keep the filenames unchanged and use the Mapping Settings in Step 4 to import your source data to a single target table.

Step 2. Create the target table schemas

Because Parquet files do not contain schema information, before importing data from Parquet files into TiDB Cloud, you need to create the table schemas using either of the following methods:

• Method 1: In TiDB Cloud, create the target databases and tables for your source data.

• Method 2: In the Amazon S3 or GCS directory where the Parquet files are located, create the target table schema files for your source data as follows:

  1. Create database schema files for your source data.

     If your Parquet files follow the naming rules in Step 1, the database schema files are optional for the data import. Otherwise, the database schema files are mandatory.

     Each database schema file must be in the ${db_name}-schema-create.sql format and contain a CREATE DATABASE DDL statement. With this file, TiDB Cloud will create the ${db_name} database to store your data when you import the data.

     For example, if you create a mydb-scehma-create.sql file that contains the following statement, TiDB Cloud will create the mydb database when you import the data.

     CREATE DATABASE mydb;
     

  2. Create table schema files for your source data.

     If you do not include the table schema files in the Amazon S3 or GCS directory where the Parquet files are located, TiDB Cloud will not create the corresponding tables for you when you import the data.

     Each table schema file must be in the ${db_name}.${table_name}-schema.sql format and contain a CREATE TABLE DDL statement. With this file, TiDB Cloud will create the ${db_table} table in the ${db_name} database when you import the data.

     For example, if you create a mydb.mytable-schema.sql file that contains the following statement, TiDB Cloud will create the mytable table in the mydb database when you import the data.

     CREATE TABLE mytable (
     ID INT,
     REGION VARCHAR(20),
     COUNT INT );
     

     Note

     Each ${db_name}.${table_name}-schema.sql file should only contain a single DDL statement. If the file contains multiple DDL statements, only the first one takes effect.

Step 3. Configure cross-account access

To allow TiDB Cloud to access the Parquet files in the Amazon S3 or GCS bucket, do one of the following:

• If your Parquet files are located in Amazon S3, configure Amazon S3 access.

  You can use either an AWS access key or a Role ARN to access your bucket. Once finished, make a note of the access key (including the access key ID and secret access key) or the Role ARN value as you will need it in Step 4.

• If your Parquet files are located in GCS, configure GCS access.

Step 4. Import Parquet files to TiDB Cloud

To import the Parquet files to TiDB Cloud, take the following steps:

• Amazon S3
• Google Cloud

When you run an import task, if any unsupported or invalid conversions are detected, TiDB Cloud terminates the import job automatically and reports an importing error.

If you get an importing error, do the following:

1. Drop the partially imported table.

2. Check the table schema file. If there are any errors, correct the table schema file.

3. Check the data types in the Parquet files.

   If the Parquet files contain any unsupported data types (for example, NEST STRUCT, ARRAY, or MAP), you need to regenerate the Parquet files using supported data types (for example, STRING).

4. Try the import task again.

Supported data types

The following table lists the supported Parquet data types that can be imported to TiDB Cloud.

Troubleshooting

Resolve warnings during data import

After clicking Start Import, if you see a warning message such as can't find the corresponding source files, resolve this by providing the correct source file, renaming the existing one according to Naming Conventions for Data Import, or using Advanced Settings to make changes.

After resolving these issues, you need to import the data again.

Zero rows in the imported tables

After the import progress shows Completed, check the imported tables. If the number of rows is zero, it means no data files matched the Bucket URI that you entered. In this case, resolve this issue by providing the correct source file, renaming the existing one according to Naming Conventions for Data Import, or using Advanced Settings to make changes. After that, import those tables again.