FlowServer for WarehousePG 1.1.0 release notes

Release Date: July 13, 2026

Packages included

The release of FlowServer for WarehousePG 1.1.0 comprises the following packages and their dependencies:

New features

• Multiple target tables per job: A job can now load data into more than one target table by listing more than one entry under target.database.tables. FlowServer stages each batch once and applies each target table's write mode against that staging table independently. See Constructing the target section and the multiple target tables example.

• Kafka partition subset consumption: The partitions setting lets a job consume an explicit subset of a topic's partitions instead of the whole topic, so you can split a high-volume topic across multiple concurrent jobs. See Constructing the source section for a Kafka source.

• Parallel Kafka reader: Kafka polling and record formatting now run in parallel across partitions. The read_workers and read_buffer settings let you tune the degree of parallelism and the internal buffer depth. See Constructing the source section for a Kafka source.

• TLS connections to Kafka brokers: A new tls configuration block enables one-way TLS and mutual TLS (mTLS) connections to Kafka brokers. See Constructing the source section for a Kafka source.

• Kafka record metadata columns: An optional metadata block projects each Kafka record's topic, partition, offset, and timestamp into the target table as a JSON column. This release also adds a group.id property to override the consumer group name. See Constructing the source section for a Kafka source.

• Avro without the Confluent wire format: FlowServer can now read raw Avro binary data produced without the Confluent 5-byte wire format header, using a local schema file instead of a schema registry. See Avro format.

• Avro bytes type support: The new bytes_to_base64 option encodes Avro bytes and fixed fields as base64, making them safe to load into text or JSON target columns. See Avro format.

Improvements

• Removing a job now deletes its Kafka consumer group, so removed jobs no longer leave stale consumer groups behind.
• Avro codec and schema registry lookups are now cached across messages instead of being rebuilt per record, reducing Avro decoding overhead.
• Starting a job no longer blocks other job-control operations while the job connects to its source.

Bug fixes

• Fixed an issue where a job restarted from a stored offset incorrectly detected log truncation, reset to the start of the Kafka log, and silently finished with a Done status instead of resuming.
• Fixed an issue where the stop timer configured by running_duration didn't fire on a job's final scheduled run when used together with max_restart_times, leaving the job running indefinitely.