Release Notes

0.1.6

2026-06-10

Targets

DuckDB

New target. Replication to DuckDB over the native Quack protocol, with DuckLake support. Read the announcement.

MotherDuck

New target. Replication to MotherDuck over its Postgres wire protocol endpoint, with DuckLake support. Read the announcement.

0.1.5

2026-06-08

Targets

Apache Doris

New target. Replication to Apache Doris and VeloDB.

ClickHouse

Defaults to the faster Parquet v3 reader (input_format_parquet_use_native_reader_v3) on ClickHouse >= 25.8.

Kafka

Failed producer delivery acks are now retried indefinitely.

Apache Iceberg

System columns like _sm_synced_at are now available as partition fields.

Sources

MongoDB

Schema-mode timestamp parsing now handles extended IANA timezone formats like 2025-11-03T14:22:47-05:00[America/New_York].
Validation now accepts sharded clusters.

0.1.4

2026-06-02

Core

Identifier naming transforms. A new opt-in identifier_naming connector setting normalizes schema, table, and column names before they reach the target. Set it to FivetranNaming to match Fivetran's naming conventions, so you don't have to rewrite SQL queries or dbt models after migration.
Connector health metric. A connector.health gauge enables threshold alerts on sustained connector failures.
Force delete. Removes a connector and skips source cleanup, for cases where the source database is no longer reachable.
Connector phase tracking. Connector phase now tracks correctly across mid-CDC re-snapshots (SQL Server).

Sources

MongoDB

When infer_typed_strings is enabled, a single unparseable value no longer widens the entire column to text. The majority type wins and outliers coerce to NULL.

Targets

Snowflake

Improves schema reflection and S3 stage handling for virtual-hosted-style URLs.

0.1.3

2026-05-25

Core

Tables without a primary key. Tables with no primary key can now replicate with deduplication. DedupByRowHash synthesizes an _sm_id surrogate key by hashing row content.

System columns. Adds _sm_synced_at, an opt-in timestamp recording when Supermetal synced each row.

Sources

MongoDB

Fixes a type inference bug where BSON arrays were replicated as strings. Arrays now land as JSON on targets with Variant or native JSON column support.

Extends CDC retry coverage to additional transient error codes.

UI

New Included and Excluded filter tabs in the Select Data step show only the tables you have selected or excluded, so you can audit your configuration without scanning the full catalog.

Search now matches column names in addition to table and schema names.

The Select Data step handles databases with thousands of tables.

2026-05-26

Targets

ClickHouse

Adds a disable_schema_prefix option. When enabled, a source table public.users lands as users rather than public_users. Only safe when table names are unique across all source schemas.

Retries when ClickHouse encounters a transient S3 read error.

SSH tunnel support. Connect Supermetal to your database through a bastion host when direct port access is not available or when you want an additional layer of security. The bastion forwards traffic to your database, so the database can stay inside a private network. Initial coverage is MySQL source, PostgreSQL source, and PostgreSQL target.

2026-05-20

Targets

Apache Iceberg

Caps S3 multipart part size at 64 MiB by default, fixing an edge case where large snapshot writes failed on oversized parts. Also fixes a Variant column bug that surfaced on CDC equality deletes.

0.1.1

Targets

Apache Iceberg

Adds a positional deletes mode for Merge on Read, backed by a per connector local primary key index. Emits V2 positional deletes or V3 deletion vectors, making Merge on Read tables readable from Snowflake and Databricks.

Sources

MongoDB

Snapshot document transform now respects row boundaries to avoid arrow Utf8 i32 offset overflow on collections with large string fields.

0.1.0

2026-05-09

Core

Adds an opt in Prometheus /metrics endpoint alongside existing OTel support. Enable it by setting the PROMETHEUS env variable.

2026-05-08

Targets

Kafka

Batches delivery acks across source transactions to fix replication stalls.

2026-05-06

Sources

MongoDB

BSON strings now infer to numeric and temporal columns. A majority vote across documents picks the dominant type per field.

2026-04-29

Sources

SQL Server

CDC sources recover from stale capture instances mid stream without operator intervention.

Core

The local metrics store that powers the dashboard now uses tiered compaction.

2026-04-22

Core

Redesigned Supermetal UI.

2026-04-06

Targets

Apache Iceberg

Writes Parquet directly to Iceberg tables via REST, AWS Glue, or S3 Tables catalogs over S3 or GCS storage. Defaults to Iceberg V3 with variant, with V2 fallback. Two write modes are available. Merge on Read uses equality or positional deletes. Append works with any Iceberg engine. Read the launch post.

2026-04-05

Sources

MySQL

Adds binlog file and position CDC mode alongside the existing GTID mode.

2026-04-02

Sources

SQL Server

Detects and switches to new capture instances as they become available, picking up schema changes without operator intervention.

2026-03-18

Sources

File

Ingests files from S3, GCS, Azure Blob, R2, and local filesystems. Detects format and compression from the file extension. Infers CSV dialect and column types. Streams every file without local disk and without a file size cap. Incremental syncs process only changed files. Maps to tables by single table, filename, or path regex.

MongoDB

Adds AWS DocumentDB support.

2026-02-10

Targets

Webhook

Pushes change events to HTTP endpoints in Debezium or Supermetal format. Rows batch into JSON arrays. Bodies compress with gzip, Zstd, or Brotli. Rate limits apply by request count and byte throughput. Deterministic idempotency keys guarantee exactly once delivery.

On this page