Table Identifier

• The <table_identifier> identifies the Iceberg table. The <table_identifier> is catalog-specific but is typically in the form: database_name.table_name.

CONFIG

CREDENTIALS (<credentials_json>)

• The <credentials_json> specifies S3 credentials in JSON format. With this specification, credentials are directly included in the pipeline. While including credentials in the pipeline is simpler, it is less secure than using vended credentials or EKS IRSA. For information about the supported credential options, refer to CREATE PIPELINE.

• When using Vended Credentials or EKS IRSA, the CREDENTIALS clause is optional; however, the clause may be included to add additional authentication parameters.

REPLACE INTO

REPLACE INTO is required for pipelines that use insert_mode: upsert. SingleStore recommends that REPLACE INTO not be used with other types of pipelines as doing so causes unnecessary overhead.

When you execute a REPLACE INTO statement on a table with a PRIMARY KEY or UNIQUE index, the engine checks for matching values in these indexes. If a match is found, it deletes the existing row and replaces it with the new row. If no match exists or if there is no PRIMARY KEY or UNIQUE index, REPLACE INTO operates like INSERT.

MERGE INTO

The MERGE INTO clause creates a pipeline, called a MERGE pipeline, that continuously ingests data from an Iceberg table into a SingleStore table. A MERGE pipeline continuously detects inserts, updates, and deletes to and from the Iceberg table and applies those updates to the SingleStore table.

SingleStore recommends MERGE pipelines as the preferred method for ingesting Iceberg data. Refer to Types of Ingest for details on MERGE and a comparison of MERGE and ingest_mode.

Subvalue Mappings

The iceberg_subvalue_mapping assigns fields from the Iceberg table to columns in the SingleStore table or to temporary variables. A ::-separated list of field names is used in iceberg_subvalue_path to look up fields in nested schemas. When the files in the Iceberg table are Parquet files, the ::-separated list of field names is used to look up fields in nested Parquet schemas. The following rules apply:

• The last field in iceberg_subvalue_path must be a primitive type.

• All iceberg_subvalue_path components containing whitespace or punctuation must be surrounded by backticks (`).

• The iceberg_subvalue_path may not contain Parquet nested types (list or map types). Refer to Parquet - Nested Types for more information.

Catalog Types

SingleStore supports the following Iceberg Catalogs: Glue, Snowflake, REST, JDBC, Hive, and Hadoop. Polaris and Unity are REST catalogs and are supported.

Vended Credentials

SingleStore supports "vended credentials" for Iceberg REST catalogs with S3 storage. Vended credentials allow clients to access Iceberg tables without the Iceberg catalog credentials. The Iceberg catalog generates temporary limited credentials for a client to access the storage, improving the security and ease of catalog administration.

Catalogs have catalog-specific settings for vended credentials. All such settings must be added as catalog properties using "catalog.<property>" in the CONFIG section. In addition, specify "iceberg_vended_credentials": true in the CONFIG.

AWS Elastic Kubernetes Service (EKS) IAM Roles for Service Accounts (IRSA)

AWS EKS IRSA can be used to provide credentials for a S3 pipeline. To use EKS IRSA:

• The SingleStore server must be deployed on an AWS EKS cluster using Helios BYOC.

• EKS IRSA must be enabled by setting the enable_eks_irsa engine variable to ON.

  SQL

  Copy

  SET GLOBAL enable_eks_irsa = ON;

• Include the "creds_mode": "eks_irsa" option in the pipeline CONFIG.

SingleStore supports two scenarios for access with IAM roles and EKS IRSA:

1. The role assigned to a pod has direct access to S3 resources.

   1. Include the "creds_mode": "eks_irsa" option in the pipeline CONFIG.

      SQL

      Copy

      CREATE PIPELINE <pipeline_name> AS
      LOAD DATA S3 's3://bucket_name/directory'
      CONFIG '{"region":"us-west-2", "creds_mode": "eks_irsa"}'
      CREDENTIALS '{}'
      INTO TABLE <destination_table>
      <iceberg_subvalue_mapping>
      FORMAT ICEBERG;

2. The role assigned to a pod does not have direct access to the S3 resources, but it has permissions to assume a role that has the desired access.

   1. Provide the role to be assumed with role_arn in the pipeline CREDENTIALS.

      SQL

      Copy

      CREATE PIPELINE <pipeline_name> AS
      LOAD DATA S3 's3://bucket_name/directory'
      CONFIG '{"region":"us-west-2", "creds_mode": "eks_irsa"}'
      CREDENTIALS '{"role_arn": "<role_arn_to_assume>"}'
      INTO TABLE <destination_table>
      <iceberg_subvalue_mapping>
      FORMAT ICEBERG;

Refer to the Minimum Required S3 Pipeline Syntax and AWS Elastic Kubernetes Service (EKS) IAM Roles for Service Accounts (IRSA) Authentication examples in CREATE PIPELINE.

AWS Secrets Manager Secrets

Secrets in the AWS Secrets Manager can be used in the pipeline CONFIG or CREDENTIALS.

To place a secret part of a pipeline CONFIG or CREDENTIALS into the AWS Secrets Manager:

1. Identify the part of the configuration that you want to hide. Use the secrets feature only for credentials; do not store the entire configuration into the Secrets Manager, as this may cause problems with the pipeline.

2. Add the secret part as a secret in AWS Secrets Manager. As AWS Secrets Manager stores secrets as JSON, a secret may contain multiple key-value pairs. For example, you might create a secret that contains values for "catalog.credential" and "catalog.another_credential".

3. Ensure the pipeline's AWS credentials allow accessing the AWS Secrets Manager. Those credentials could be either static credentials, or a service role provided to the pod by EKS IRSA.

4. Remove the secret part or parts from the pipeline CONFIG and replace them with "ref": "arn_of_your_secret".

   JSON

   Copy

   {
   "catalog_type":"REST",
   "region":"us-east-1",
   "catalog.uri":"https://whatever.snowflakecomputing.com/polaris/api/catalog",
   "ref": "arn:aws:secretsmanager:us-east-1:188440000000:secret:test-secret-VkYPHc",
   ...
   }

Every "ref" property in the pipeline CONFIG or CREDENTIALS is replaced by the values contained in the secret mentioned in the property's value. These values are placed on the same level as the "ref" they replace.

The AWS Secrets Manager can be accessed with a role that must be assumed before accessing the secret. Set this role in CREDENTIALS using secretsmanager_role_arn, or in ref by using a pair of role ARN and secret ARN separated by a pipe; for example: "ref": "arn:aws:iam::188440000000:role/role_to_assume|arn:aws:secretsmanager:us-east-1:188440000000:secret:my-secret".

When AWS credentials are used only to access the Secrets Manager, and not to access data on S3 (for example, when vended credentials are used to access S3) those credentials must be prefixed with the string "secretsmanager_", such as, "secretsmanager_role_arn".

Compare Types of Continuous Ingest

SingleStore supports several types of continuous ingest: MERGE pipelines, Append-Only, and Upsert. With continuous ingest, a running pipeline automatically detects updates to the Iceberg table and ingests them into the SingleStore table.

These types of ingest have different characteristics, SingleStore recommends using MERGE pipelines as the preferred method for ingesting Iceberg data. Append-Only and Upsert are provided as performance optimizations.

• MERGE can process Append, Overwrite, Replace, and Delete (positional and equality) Iceberg snapshots; Append-Only and Upsert using ingest_mode have limitations on the types of Iceberg snapshots processed.

  • The MERGE pipeline has the broadest coverage of what can be ingested as incremental updates, merge pipelines are therefore inherently more complex.

• MERGE requires adding three columns ($_file, $_row, $_delete) in the destination (SingleStore) table. These three extra columns have the following impacts:

  • MERGE is slightly slower on initial ingest compared to Append-Only mode.

  • MERGE takes slightly more space in storage.

  • The MERGE pipeline statement is more complex, but that can be mitigated with INFER.

• When using MERGE, users may not directly update the destination (SingleStore) table. If the destination (SingleStore) table is updated, updates made to the source Iceberg table may not propagate properly.

• When using MERGE, the Replace type of snapshot is not a no-op. Data files will be re-ingested if the source table went through compaction.

• MERGE implements merge-on-read (MOR).

• MERGE requires an extra condition for SELECT statements to manage MOR-style deletes.

  • An extra entity (VIEW) is created by INFER by default.

  • Impacts scan performance.

• ingest_mode updates the SingleStore table directly and does not have a read penalty.

Merge Pipelines (MERGE)

The MERGE clause creates a merge pipeline, that continuously ingests data from an Iceberg table into a SingleStore table. Append, Overwrite, Replace, and Delete (positional and equality) Iceberg V2 snapshots are supported.

For a merge pipeline, the SingleStore (destination) table:

• Must have the following three columns:

  • `$_file` VARCHAR(2048) COMMENT 'ICEBERG_FILE_PATH'

  • `$_row` BIGINT COMMENT 'ICEBERG_FILE_POS'

  • `$_delete` JSON DEFAULT '{}' COMMENT 'ICEBERG_DELETED_BY'

• You may use different column names for the `$_file`, `$_row`, and `$_delete` columns as long as the columns are marked with the COMMENT clauses shown above.

• Must have shard key defined as:

  • SHARD KEY(`$_file`, `$_row`)

• Must not be updated or modified. If the destination table is modified, updates may be applied incorrectly.

The pipeline declaration must set the `$_file` and `$_row` columns as follows:

SET `$_file` = pipeline_source_file(), `$_row` = pipeline_source_file_row()

To ingest deletes, a merge pipeline must be created using a VIEW over the destination table. This view must have the following WHERE clause:

WHERE JSON_LENGTH(`$_delete`) = 0

Merge pipelines perform merge-on-read on-demand during table scan when a view is used.

MERGE Pipeline provides a merge pipeline example.

Append-Only (ingest_mode)

Append-only mode is used for scenarios in which new rows are added to the Iceberg table, but no rows are deleted or modified.

In append-only mode ("ingest_mode":"append") the pipeline will process APPEND Iceberg snapshots. If a DELETE or OVERWRITE snapshot is encountered, an error is raised. Users may override those errors by specifying "ignore_unsupported_modifications":true in the pipeline configuration. SingleStore does not recommend setting "ignore_unsupported_modifications":true as doing so may lead to data inconsistency.

Upsert (ingest_mode)

Upsert mode is used for scenarios where updates modify non-key columns based on a specified key column(s). An upsert is an insert and update, an upsert updates a row in the table if there is a primary key match or inserts the row if there is no primary key match.

In upsert mode ("ingest_mode":"upsert"), the pipeline will process updates to the Iceberg table as upserts to the SingleStore table. That is, the pipeline will process Iceberg APPEND and OVERWRITE snapshots of the Iceberg table as upserts to the SingleStore table.

Requirements:

• The SingleStore table must have a primary key or a unique index. In the <iceberg_subvalue_mapping>, a column(s) in the Iceberg table must be mapped to the column(s) in SingleStore on which there is a key or unique index. Typically, the column(s) in the Iceberg table will also be declared as a key.

• The pipeline must be created using REPLACE.  Refer to Additional CREATE PIPELINE Examples for more information.

Limitations:

• Pipelines will fail on delete snapshots. Users may override these errors by specifying "ignore_unsupported_modifications":true in the pipeline configuration. SingleStore does not recommend setting "ignore_unsupported_modifications":true as doing so may lead to data inconsistency.

• Limitations are expected to be addressed in future releases.

One-Time (ingest_mode)

A one-time ingest ("ingest_mode":"one_time") is supported. In this mode, SingleStore requests Iceberg table metadata, and loads data from the latest snapshot available at that moment. Subsequent updates to the Iceberg table are not automatically ingested.

CREATE OR REPLACE

When a pipeline with a specified pipeline_name already exists, the CREATE OR REPLACE command functions similarly to CREATE PIPELINE, with the added benefit of preserving existing pipeline metadata, such as loaded offsets and data files. Running CREATE OR REPLACE on an existing pipeline initiates the Iceberg pipeline to retrieve a new snapshot, schema, and data files, and inserts data from these new files into the destination table in SingleStore.

Executing CREATE OR REPLACE on an existing Iceberg pipeline may cause some data files to be ingested twice. To avoid this, use CREATE OR REPLACE only with REPLACE statements or in an upsert configuration.

CREATE PIPELINE books_create_pipe AS
LOAD DATA S3 'db.books'
CONFIG '{"region":"us-west-2",
        "catalog_type": "GLUE",
        "catalog_name": "s3_catalog",
        "ingest_mode": "one_time"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
REPLACE INTO TABLE books
  (Id <- id,
  Name <- name,
  NumPages <- numPages,
  Rating <- rating)
FORMAT ICEBERG;

Refer to CREATE PIPELINE for syntax for CREATE OR REPLACE PIPELINE, REPLACE INTO TABLE, and ON DUPLICATE KEY UPDATE. Refer to Performing Upserts for more information on upserts.

Data Type Translation

The following table lists Iceberg Types, the SingleStore data type that can be used to store those types, and the recommended conversion to be applied with a SET clause.

Data File Filtering

When ingesting from Iceberg tables, SingleStore applies file-level filtering to reduce the number of data files downloaded.

File-level filtering is determined by the WHERE clause in the pipeline definition.

• Only a subset of expressions can be used for file filtering.

• Expressions that are not eligible for file-level filtering are applied later as row-level filtering after the file is downloaded.

To enable file filtering before download, write the WHERE clause using the following rules:

• Use any combination of the boolean operators AND and OR.

• Use the binary operators =, <, <=, >, >= with a column name on one side and a literal value on the other. For example, WHERE column_name_1 < 9999

• Function calls are not supported in file filtering.

Pipelines always evaluate the full WHERE clause during ingestion, after file-level filtering.

• Row-level filtering may include more complex expressions, such as function calls, that are not eligible for file-level filtering.

• When the WHERE clause uses an AND operator, unsupported expressions can be discarded for file-level filtering while still being applied during row-level filtering.

For example, WHERE col1 > 1000 AND col2 = my_udf(col3)

• Per-file filtering: col1 > 1000

• Per-row filtering: col1 > 1000 AND col2 = my_udf(col3)

If none of the expressions in the WHERE clause can be applied at the file level, the CREATE PIPELINE statement issues the following warning:

Pipeline's where clause is not suitable for file-level filtering

File-level filtering relies on column-level metadata that Iceberg generates for each data file. Refer to Manifest Entry Fields for more information.

Monitor Ingest

The per-file status of the Iceberg table ingestion can be queried through the PIPELINES_ICEBERG_FILES information schema view. This view lists all files currently tracked by the pipeline and their corresponding status.

Enabling OFFSETS_METADATA_GC on the Iceberg pipeline reduces the number of entries by tracking files only referenced by the pipeline’s current Iceberg table snapshot. The option OFFSETS_METADATA_GC can be set when you CREATE the pipeline or modify it using ALTER PIPELINE.

Configure Iceberg Ingest

The global engine variable enable_iceberg_ingest must be set to ON to use Iceberg ingest. This variable is set to OFF by default.

SET GLOBAL enable_iceberg_ingest = ON;

Engine variables can be used to control parameters including memory usage and timeouts for Iceberg pipelines. The default values of these engine variables are defined to work well for common Iceberg pipelines uses.

Pipeline parameters including memory usage and timeouts can be controlled using global engine variables with the pipelines_* prefix as specified in List of Engine Variables.

Memory usage can also be controlled using the java_pipelines_heap_size and pipelines_iceberg_data_workers_heap_size engine variables. Separate variables are used to control the memory used by leaf and aggregator nodes. Adjust these variables if there are out of memory (OOM) errors in Java extractors.

Pipeline timeouts can be controlled using pipelines_extractor_get_offsets_timeout_ms and pipelines_extractor_idle_timeout_ms. For Iceberg pipelines, the values of these variables are adjusted to be a minimum of 5 minutes. Increase the values of these variables if there are timeouts in pipeline creation or while running pipeline batches.

Refer to List of Engine Variables for more information.

Glue Catalog on Amazon S3

An Iceberg table with data files in Parquet format that is stored in an AWS S3 bucket using AWS Glue can be loaded into a SingleStore table using a pipeline (CREATE PIPELINE. Refer to Apache Iceberg - Glue Catalog for information on using a GLUE catalog with Iceberg.

In this example, a table named books is created and data from an Iceberg table that meets this schema is loaded into the books table.

Create the table.

CREATE TABLE books(
  Id INT,
  Name TEXT,
  NumPages INT,
  Rating DOUBLE,
  PRIMARY KEY(Id));

The following data is used for this example.

(1, 'Happy Place', 400, 4.9)
(2, 'Legends & Lattes', 304, 4.9)
(3, 'The Vanishing Half', 352, 4.9)
(4, 'The Wind Knows My Name', 304, 4.9)

The PIPELINE statement below will load data from an Iceberg table containing the data above into the books table. The column names on the left side of the <- are the column names from the SingleStore table into which the data will be loaded. The column names on the right side of the <- are the column names from the Iceberg table which is to be loaded into SingleStore.

CREATE PIPELINE books_pipe AS
LOAD DATA S3 'db.books'
CONFIG '{"region":"us-west-2",
         "catalog_type": "GLUE"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
              "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE books
   (Id <- id,
   Name <- name,
   NumPages <- numPages,
   Rating <- rating)
FORMAT ICEBERG;

Test the pipeline.

TEST PIPELINE books_pipe;

+------+------------------------+----------+--------+
| Id   | Name                   | NumPages | Rating |
+------+------------------------+----------+--------+
|    4 | The Wind Knows My Name |      304 |    4.9 |
|    1 | Happy Place            |      400 |    4.9 |
|    2 | Legends & Lattes       |      304 |    4.9 |
|    3 | The Vanishing Half     |      352 |    4.9 |
+------+------------------------+----------+--------+

Refer to START PIPELINE for more information on starting pipelines.

Use Subvalue Mappings

This example shows the use of subvalue mappings to load nested elements from an Iceberg schema into a SingleStore table.

Create a table.

CREATE TABLE addresses(
Id INT,
Name TEXT,
Street TEXT,
City TEXT,
Country TEXT,
PRIMARY KEY(Id));

The following data is used for this example.

(1, 'Mylo', struct('123 Main St', 'New York', 'USA'))
(2, 'Naya', struct('456 Elm St', 'San Francisco', 'USA'))

The PIPELINE statement below will load data from an Iceberg table containing the data above into the addresses table. The column names on the left side of the <- are the column names from the SingleStore table into which the data will be loaded. The column names on the right side of the <- are the column names from the Iceberg table which is to be loaded into SingleStore.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db2.addresses'
CONFIG '{"region":"us-west-2",
         "catalog_type": "GLUE",
         "catalog_name": "s3_catalog"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
              "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
   (Id <- id,
   Name <- name,
   Street <- address::street,
   City <- address::city,
   Country <- address::country)
FORMAT ICEBERG;

Test the pipeline.

TEST PIPELINE addresses_pipe;

+------+------+-------------+---------------+---------+
| Id   | Name | Street      | City          | Country |
+------+------+-------------+---------------+---------+
|    1 | Mylo | 123 Main St | New York      | USA     |
|    2 | Naya | 456 Elm St  | San Francisco | USA     |
+------+------+-------------+---------------+---------+

Refer to START PIPELINE for more information on starting pipelines.

Snowflake Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 with a Snowflake catalog.

Iceberg tables to be ingested in SingleStore must be created on an external volume. Refer to Tutorial: Create your first Iceberg table, Create an external volume, and Snowflake Iceberg Catalog SDK for more information.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.schema_name.table_name'
CONFIG '{"region": "us-west-2",
        "catalog_type": "SNOWFLAKE",
        "catalog.uri": "jdbc:snowflake://tpq12345.snowflakecomputing.com",
        "catalog.jdbc.user":"<user_name>",
        "catalog.jdbc.password":"<password>",
        "catalog.jdbc.role":"<user role>"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
  (Id <- id,
  Name <- name,
  Street <- address::street,
  City <- address::city,
  Country <- address::country)
FORMAT ICEBERG;

For the Snowflake catalog, the <table_identifier> must consist of three parts - the database name, the schema name, and the table name, db_name.schema_name.table_name in the example above.

The catalog.uri can be obtained from running SELECT SYSTEM$ALLOWLIST(); in the Snowflake system.

In addition, the catalog.uri, catalog.jdbc.user, catalog.jdbc.password, and catalog.jdbc.role are required when using the Snowflake catalog.

REST Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 with REST catalog.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.table_name'
CONFIG '{"region": "us-west-2",
        "catalog_type": "REST",
        "catalog.uri": "http://host.addresss:8181"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
  (Id <- id,
  Name <- name,
  Street <- address::street,
  City <- address::city,
  Country <- address::country)
FORMAT ICEBERG;

File Filtering with REST Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 with REST catalog using file filtering.

Assume there exists an Iceberg table that stores weather sensor data and is partitioned with a day transform on the timestamp as follows.

CREATE TABLE sensors.weather (
  sensorid int,
  temperature double,
  humidity double,
  pressure double,
  ts timestamp)
PARTITIONED BY day(ts);

This data is to be loaded into a SingleStore table defined as follows:

CREATE TABLE weather_data(
  SensorId INT,
  Temperature DOUBLE,
  Humidity DOUBLE,
  Pressure DOUBLE,
  TS DATETIME,
  PRIMARY KEY(SensorId));

The following pipeline will load all weather data for January 1, 2026 and later into the weather_data table. As the Iceberg table is partitioned by day(ts), SingleStore will read and load only the Iceberg files containing data for 2026 (and future years).

CREATE PIPELINE weather_pipe AS
LOAD DATA S3 'sensors.weather'
CONFIG '{"region": "us-west-2",
       "catalog_type": "REST",
       "catalog.uri": "http://host.addresss:8181"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
            "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE weather_data
 (SensorId <- sensorid,
  Temperature <- temperature,
  Humidity <- humidity,
  Pressure <- pressure,
  TS <- ts)
FORMAT ICEBERG
WHERE timestamp >= '2026-01-01';

JDBC Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 with JDBC catalog. Refer to Iceberg JDBC Integration for more information on using Iceberg with JDBC catalog.

SingleStore supports Postgres, MySQL, and SQLite JDBC drivers by default.

The following example uses JDBC with SQLite.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.table_name'
CONFIG '{"region": "us-west-2",
        "catalog_type": "JDBC",
        "catalog_name": "catalog_name",
        "catalog.warehouse": "s3://path_to_warehouse",
        "catalog.uri":"jdbc:sqlite_:file:/path_jdbc"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
  (Id <- id,
  Name <- name,
  Street <- address::street,
  City <- address::city,
  Country <- address::country)
FORMAT ICEBERG;

The following example uses JDBC with MySQL.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.table_name'
CONFIG '{"region": "us-west-2",
        "catalog_type": "JDBC",
        "catalog_name": "catalog_name",
        "catalog.warehouse": "s3://path_to_warehouse",
        "catalog.uri": "jdbc:mysql://host.address:3306/default",
        "catalog.jdbc.user": "<user_name>",
        "catalog.jdbc.password": "<password>"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
  (Id <- id,
   Name <- name,
   Country <- address::country)
FORMAT ICEBERG;

The catalog.warehouse and catalog_name are required for JDBC catalogs.

Hive Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 using Hive Catalog. Iceberg tables to be ingested in the Hive catalog must use Hive Metastore service. Refer to Apache Hive and Hive - Apache Iceberg for more information.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.table_name'
CONFIG '{"catalog_type": "HIVE",
         "catalog.uri": "thrift://<service_endpoint>:46590",
         "region": "us-east-1",
         "catalog.hive.metastore.client.auth.mode": "PLAIN",
         "catalog.hive.metastore.client.plain.username": "<username>",
         "catalog.hive.metastore.client.plain.password": "<password>",
         "catalog.metastore.use.SSL": "true",
         "catalog.hive.metastore.truststore.type": "PKCS12",
         "catalog.hive.metastore.truststore.path": "/path/to/your/project/hive/truststore.12"
         "catalog.hive.metastore.truststore.password": <truststore_password>
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
              "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
 (Id <- id,
 Name <- name,
 Street <- address::street,
 City <- address::city,
 Country <- address::country)
FORMAT ICEBERG;

The catalog.uri is the base URL for accessing the Hive catalog's API or service endpoint. 

The catalog.hive.metastore.client.auth.mode is the authentication mode for connecting to the Hive Metastore. 

The catalog.hive.metastore.client.plain.username is the username used to authenticate with the Hive Metastore. 

The catalog.hive.metastore.client.plain.password is the password for the authenticated user. 

The catalog.metastore.use.SSL is a boolean flag that secures communication with the Hive Metastore. 

The catalog.hive.metastore.truststore.type is the truststore format used to validate the SSL certificate. 

The catalog.hive.metastore.truststore.path is the file path that contains the SSL certificate. 

The catalog.hive.metastore.truststore.password is the password needed to access the truststore.

Refer to GitHub for additional Hive configurations.

Hadoop Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 with Hadoop catalog. Hadoop catalog supports S3 compatible backend storages. Refer to Apache Hadoop and Hadoop - Apache Iceberg for more information.

CREATE PIPELINE addresses_pipe AS
LOAD DATA s3 'db_name.table_name'
CONFIG '{"catalog_type": "HADOOP",
         "catalog.warehouse": "s3://path_to_warehouse",
         "endpoint_url": "<s3_compatible_storage_endpoint>",
         "region": "us-east-1"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key":"<your_secret_access_key>"}'

INTO TABLE addresses
(Id <- id,
Name <- name,
Street <- address::street,
City <- address::city,
Country <- address::country)
FORMAT ICEBERG;

Polaris Catalog on Amazon S3

Ingest an Iceberg table stored in Amazon S3 using Polaris Catalog. Polaris catalog uses the Apache Iceberg REST API. Refer to Getting Started with Snowflake Open Catalog for more information.

CREATE PIPELINE addresses_pipe AS
LOAD DATA S3 'db_name.table_name'
CONFIG '{"catalog_type": "REST",
        "catalog.warehouse": "<polaris_catalog_name>",
        "table_id": "db_name.table_name",
        "region":"us-east-1",
        "catalog.uri":"https://tpq12345.snowflakecomputing.com/polaris/api/catalog",
        "region":"us-east-1",
        "catalog.credential":"catalog.credential",
        "catalog.scope": "PRINCIPAL_ROLE:ALL"}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
           "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE addresses
(Id <- id,
Name <- name,
Street <- address::street,
City <- address::city,
Country <- address::country)
FORMAT ICEBERG;

The catalog.uri is the base URL for accessing the Polaris catalog's API or service endpoint.

The catalog.scope defines the access permissions for the Polaris catalog.

The catalog.connection is a secret key from Polaris catalog connection, formatted as <ClientID>:<Secret>. Get this when you configure the service connection in Snowflake while creating a connection for a Polaris catalog.

Upsert (ingest_mode)

The example below shows a pipeline using ingest_mode of upsert. As required when ingest_mode is upsert, the books table has a primary key, Id, in this example.

When started, this pipeline will ingest data from the latest snapshot of the Iceberg table into the SingleStore books table. Then, when the Iceberg table is updated, those updates will automatically be applied to the SingleStore table as upserts.

CREATE PIPELINE books_upsert_pipe AS
LOAD DATA S3 'db.books'
CONFIG '{"region":"us-west-2",
        "catalog_type": "GLUE",
        "catalog_name": "s3_catalog"
        "ingest_mode": "upsert"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
REPLACE INTO TABLE books
  (Id <- id,
  Name <- name,
  NumPages <- numPages,
  Rating <- rating)
FORMAT ICEBERG;

Append (ingest_mode)

The example below shows a pipeline using ingest_mode of append.

When started, this pipeline will ingest data from the latest snapshot of the Iceberg table into the SingleStore books table. Then, when the Iceberg table is updated, any appends to that table will automatically be appended to the SingleStore table. If non-append updates are encountered, an error will be thrown.

CREATE PIPELINE books_append_pipe AS
LOAD DATA S3 'db.books'
CONFIG '{"region":"us-west-2",
        "catalog_type": "GLUE",
        "catalog_name": "s3_catalog",
        "ingest_mode": "append"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
INTO TABLE books
  (Id <- id,
  Name <- name,
  NumPages <- numPages,
  Rating <- rating)
FORMAT ICEBERG;

MERGE Pipeline

The example below shows an example of a creating a merge pipeline.

In this example, a destination table in SingleStore , named _books , is created with the necessary $_file, $_row, and $_delete columns. Then a view, books , is created on top of the _books table. Finally, a merge pipeline is defined.

When started, this pipeline will ingest data from the latest snapshot of the source Iceberg table into the SingleStore _books table. When the source Iceberg table is updated, those updates (append, replace, delete) will be propagated to the _books table.

Create the destination _books table.

CREATE TABLE _books(
  Id INT,
  Name TEXT,
  NumPages INT,
  Rating DOUBLE,
  `$_file` VARCHAR(2048),
  `$_row` BIGINT,
  `$_delete` JSON DEFAULT '{}',
  KEY(Id),
  PRIMARY KEY(`$_file`,`$_row`));

Create the books view.

CREATE VIEW books
AS
SELECT Id, Name, NumPages, Rating from _books
WHERE JSON_LENGTH(`$_delete`) = 0;

Define the pipeline.

CREATE PIPELINE books_pipe AS
LOAD DATA S3 'db.books'
CONFIG '{"region":"us-west-2",
        "catalog_type": "GLUE",
        "catalog_name": "s3_catalog"
}'
CREDENTIALS '{"aws_access_key_id": "<your_access_key_id>",
             "aws_secret_access_key": "<your_secret_access_key>"}'
MERGE INTO TABLE _books
  (Id <- id,
  Name <- name,
  NumPages <- numPages,
  Rating <- rating)
FORMAT ICEBERG
SET `$_file` = pipeline_source_file(),
    `$_row` = pipeline_source_file_row();

Schema Change Detection

Consider an Iceberg table stored external to SingleStore that is to be ingested into SingleStore with a pipeline. In this example, we refer to the Iceberg table as the source table and the SingleStore table as the destination table.

Let the source Iceberg table have the following schema:

(
id int,
bigid long,
fl float,
doub double,
occurrence_da date,
data string,
st struct<a int, b int, c struct<d string, e int>>
)

Assume a pipeline has been created to ingest the source Iceberg table into a destination SingleStore table.

Assume the following alter table queries are performed on the source Iceberg table. These queries demonstrate the possible types of schema evolution: promote a type, rename a column, add a column, and drop a column.

ALTER TABLE <table_name> ALTER COLUMN id TYPE bigint
ALTER TABLE <table_name> RENAME COLUMN id to identity
ALTER TABLE <table_name> ADD COLUMN name string
ALTER TABLE <table_name> DROP COLUMN occurrence_da

If "schema.change.stop":true is set, when the engine detects these changes to the source Iceberg table, the pipeline is stopped, and the following error message is generated.

Iceberg Table Schema at Source has changed. Schema Diff :
{
  "schema_id": {
    "before": 0,
    "after": 8
  },
  "diff": [
    {
      "op": "column_type_change",
      "column": "id",
      "before": "int",
      "after": "long",
      "column_id": 1
    },
    {
      "op": "column_rename",
      "before": "id",
      "after": "identity",
      "column_id": 1
    },
    {
      "op": "column_add",
      "before": null,
      "after": "name",
      "column_id": 13
    },
    {
      "op": "column_delete",
      "before": "occurrence_da",
      "after": null,
      "column_id": 5
    }
  ]
}

In this error message, the Iceberg schema version is included in schema_id, and the diff includes one entry for each modification to the table schema.

Debug Datetime Conversions

Datetime value conversion can be debugged by loading raw datetime from an Iceberg file into a BLOB. The example below creates a books_debug table with a PublishTimestamp attribute of type DATETIME(6) and an extra attribute, RawTimestamp, into which the raw timestamp data will be loaded. If the datetimes are stored as strings, try loading the raw values into a TEXT attribute in addition to, or instead of, using a BLOB attribute.

CREATE TABLE books_debug(
 Id INT,
 Name TEXT,
 NumPages INT,
 Rating DOUBLE,
 PublishTimestamp DATETIME(6),
 RawTimestamp BLOB);

The PIPELINE statement below loads the publishTimestamp from the Iceberg file into both the PublishTimestamp column and the RawTimestamp column. A SET statement is used to convert the publishTimestamp from Iceberg to the PublishTimestamp in SingleStore.

The column names on the left side of the <- are the column names from the SingleStore table into which the data will be loaded. The column names on the right side of the <- are the column names from the Iceberg file which is to be loaded into SingleStore.

CREATE PIPELINE books_debug_pipe AS
LOAD DATA S3 ''
CONFIG '{
  "region":"us-west-2",
  "catalog_type": "GLUE",
  "catalog_name": "s3_catalog",
  "table_id": "dbtest.books"
}'
CREDENTIALS '{
  "aws_access_key_id": "",
  "aws_secret_access_key": "",
  "aws_session_token": ""}'
INTO TABLE books_debug
 (Id <- IdIce,
  Name <- Name,
  NumPages <- NumPages,
  Rating <- Rating,
  @ts<-PublishTimestamp,
  RawTimestamp<-PublishTimestamp
)
FORMAT ICEBERG
SET PublishTimestamp = DATE_ADD('1970-01-01', INTERVAL @ts MICROSECOND);

Use TEST PIPELINE to compare the values in the two columns. The PublishTimestamp column shows the converted timestamp, the RawTimestamp column contains the timestamp from the Iceberg file, in microseconds in this example. The combination of these two columns can be used to debug the conversion used in the SET statement.

TEST PIPELINE books_debug_pipe;

+------+--------------------+----------+--------+------------------+-------+---------------------+
| Id   | Name               | NumPages | Rating | RawTimestamp     | PublishTimestamp           |
+------+--------------------+----------+--------+------------------+-------+---------------------+
|    1 | Happy Place        |      400 |    4.9 | 1680721200000000 | 2023-04-05 12:00:00.000000 |
|    2 | Legends & Lattes   |      304 |    4.9 | 1669665600000000 | 2022-11-28 12:00:00.000000 |
|    3 | The Vanishing Half |      352 |    4.9 | 1591124400000000 | 2020-06-02 12:00:00.000000 |
+------+--------------------+----------+--------+------------------+-------+---------------------+