0.1.12

0.1.12

2026-08-09

New

• Amazon Redshift: Replication to Amazon Redshift.
• File sink: Replication to files on object stores (S3, GCS, Azure Blob), SFTP, Google Drive, and Dropbox. Writes Parquet, CSV, JSON, or Avro with optional Hive style partitioning.
• PostgreSQL unchanged TOAST: New placeholder mode preserves unchanged TOAST values on the target without requiring REPLICA IDENTITY FULL. Currently supported on the Snowflake target.

Core

• Snapshot retries: Failed snapshot loads on buffered connectors are retried up to max attempts. Snapshots for connectors without a buffer (Kafka, Webhook, File sink) are not retried yet. Iceberg support is coming soon.
• Metrics: New connector state and info gauges.
• Identifier naming: FivetranNaming now has configurable conflict handling when normalized names collide, either keeping the first column or stopping the sync.
• UI:
  • Export connector configs as JSON and import them to create new connectors, a quick way to bootstrap a new instance.
  • Tabbed connector sidebar with a task list for table resyncs.
  • Fixes catalog selection, save, and reload.
  • The connector state formerly labeled Shutdown now shows as Stopped.

ClickHouse

• Cluster support for self managed ClickHouse clusters.
• Opt-in type based compression codecs.
• History tables preserve the source primary key in ORDER BY.
• Behavior change: source timezone unaware temporal types are now declared as UTC on newly created tables.

Apache Iceberg

• Table truncation is now committed as an overwrite snapshot.
• Fixes a positional delete mode issue where an external compaction running between write batches could leave duplicate rows in the table.

SQL Server

• Tables with physical partitions now chunk within each individual partition.
• Fixes the latency.total metric by accounting for the server's local timezone when reading CDC commit timestamps.
• Additional retry coverage, including Always On failover and database recovery errors.

MySQL

• Snapshot chunking now handles arbitrary string primary keys (VARCHAR, CHAR, TEXT), in addition to UUIDs.
• Tables with physical partitions now chunk within each individual partition.
• GTID checkpoints now advance on databases with no activity. System schemas are excluded from CDC.

MongoDB

• Fixes snapshot failures on very large collections that exceeded the S3 multipart limit.
• Invalid UTF-8 in BSON strings is now tolerated.

File Source

Remote files (S3, GCS, Azure Blob) now load into ClickHouse zero-copy. ClickHouse reads the files directly instead of through Supermetal.