Openflow Connector for MySQL: Iceberg table destinations¶
The Openflow Connector for MySQL supports writing to Snowflake-managed Apache Apache Iceberg™ tables
as an opt-in destination format. Iceberg v2 and v3 are both supported. Setting Table Storage Format = ICEBERG
in the connector’s parameter context is the only change required. All Iceberg storage settings
(external volume, catalog, version, and serialization policy) are inherited from the Snowflake
destination database defaults.
Storage can be either Snowflake-managed (EXTERNAL_VOLUME = 'SNOWFLAKE_MANAGED') or a
customer-configured external volume. With Snowflake-managed storage, no external cloud storage
or IAM grants are required. For more information, see
Snowflake storage for Iceberg tables.
Existing connectors using standard tables aren’t affected.
Prerequisites¶
- Openflow runtime: An existing runtime to host the connector.
- MySQL source configured for CDC: Binary logging enabled (
log_bin = ON,binlog_format = ROW,binlog_row_image = FULL), a user withREPLICATION SLAVEandREPLICATION CLIENTprivileges, and a sufficiently longbinlog_expire_logs_secondsfor snapshot reconciliation. For details, see Set up the Openflow Connector for MySQL. - Snowflake external volume (customer-managed storage only): An external volume configured
for Iceberg storage, with USAGE granted to the connector’s Snowflake role. See
CREATE EXTERNAL VOLUME. Not required when using
Snowflake-managed storage (
EXTERNAL_VOLUME = 'SNOWFLAKE_MANAGED'). - Snowflake destination database: An existing database configured with Iceberg parameters (next section).
Step 1: Configure the Snowflake destination database¶
Set the Iceberg defaults on the destination database. The connector reads these defaults at runtime, so no per-connector Iceberg configuration is needed beyond Table Storage Format.
Option A: Snowflake-managed storage¶
With Snowflake-managed storage, Snowflake stores and manages the Iceberg table files for you. No external cloud storage or IAM grants are required.
To configure an existing database:
Option B: Customer-configured external volume¶
If you need to keep table files in your own cloud storage, configure the database with your external volume:
To configure an existing database:
| Parameter | Required | Notes |
|---|---|---|
| EXTERNAL_VOLUME | Yes | The external volume for Iceberg file storage. |
| ICEBERG_VERSION_DEFAULT | Yes |
|
| STORAGE_SERIALIZATION_POLICY | Yes |
|
Note
CATALOG = 'SNOWFLAKE' is set automatically by the connector on each CREATE ICEBERG TABLE
statement. Don’t set it at the database level.
The base location for each table is auto-derived using the
flat layout:
STORAGE_BASE_URL/database/schema/table_name.randomId/[data | metadata]/.
No user configuration is needed.
If using a customer-configured external volume (Option B), grant the connector’s Snowflake role USAGE on the external volume:
This step is not required for Snowflake-managed storage.
Step 2: Set Table Storage Format in the connector’s parameter context¶
Set the Table Storage Format parameter to ICEBERG in the connector’s destination parameter context.
The default is STANDARD.
For the full connector creation and configuration workflow, see Set up the Openflow Connector for MySQL.
Step 3: Start and verify¶
Start the connector as usual. After the initial snapshot completes, verify the destination tables are Iceberg:
Known limitations¶
- Tri-Secret Secure accounts and Snowflake-managed storage: Accounts with Tri-Secret Secure (TSS) enabled may be unable to create new Snowflake-managed Iceberg tables on Snowflake storage. For details, see Encryption.
- Incompatible type changes on the source cause table failure: Only same-family type widening
is allowed on Iceberg tables (for example,
inttolong,floattodouble, ordecimal(P,S)todecimal(P',S)with wider precision). Cross-family changes such asinttodecimal,longtostring, orinttobooleanare not allowed. If a source column’s type changes to an incompatible Iceberg type, the table is marked as failed and requires a rebootstrap. - BIGINT to BIGINT UNSIGNED schema evolution not allowed: Signed
BIGINTmaps tolongwhileBIGINT UNSIGNEDmaps todecimal(20,0). Iceberg does not allow promotion fromlongtodecimal, so this schema change on the source will fail replication. - Do not change Table Storage Format or Iceberg version after the connector starts: The connector’s Table Storage Format and the destination database’s Iceberg version should not be modified while the connector is running or stopped. Mixing settings across destination tables is not supported. To switch, follow the steps in Switching table storage format or Iceberg version.
Type mapping reference¶
The following table shows how MySQL types map to Snowflake standard and Iceberg destination types:
| MySQL type | Snowflake (Standard) | Iceberg v3 | Iceberg v2 |
|---|---|---|---|
| TINYINT / SMALLINT / MEDIUMINT (signed or unsigned) | INT | int | int |
| INT (signed) | INT | int | int |
| INT UNSIGNED | INT | long | long |
| BIGINT (signed) | INT | long | long |
| BIGINT UNSIGNED | INT | decimal(20,0) | decimal(20,0) |
| YEAR | INT | int | int |
| FLOAT (signed/unsigned) | FLOAT | double | double |
| DOUBLE (signed/unsigned) | FLOAT | double | double |
| DECIMAL(P,S) (P ≤ 38) | NUMBER(P,S) | decimal(P,S) | decimal(P,S) |
| DECIMAL(P,S) (P > 38) | TEXT | string | string |
| BOOLEAN / BOOL | INT | int | int |
| DATE | DATE | date | date |
| TIME | TIME | time | time |
| DATETIME | TIMESTAMP_NTZ | timestamp | timestamp |
| TIMESTAMP | TIMESTAMP_TZ | timestamptz | timestamptz |
| CHAR / VARCHAR / TEXT / TINYTEXT / MEDIUMTEXT / LONGTEXT | TEXT | string | string |
| ENUM / SET | TEXT | string | string |
| BIT | TEXT | string | string |
| BINARY / VARBINARY / BLOB / TINYBLOB / MEDIUMBLOB / LONGBLOB | BINARY | binary | binary |
| JSON | VARIANT | variant | string |
| GEOMETRY family | TEXT | string | string |
Source types not listed in the table are mapped to TEXT on standard tables and string on Iceberg
tables.
Switching table storage format or Iceberg version¶
Switching between Standard and Iceberg, or between Iceberg v2 and v3, requires recreating the connector. Follow these steps:
- Stop the connector.
- Delete the process group in Openflow.
- Manually clean up the destination database (drop the replicated schemas/tables, or use a new database).
- Reimport the connector with the new Table Storage Format or Iceberg version configured on the destination database.
This ensures all connector state is correctly cleaned up within Openflow. The new connector performs a fresh snapshot into the destination.