For installation instructions check out the getting started guide.
Fixed
- aws_dynamodb_cdc: Settle acks by handle and isolate per-shard backpressure (@squiidz, #4739)
- prometheus: Fixed prometheus exporter accumulating frozen metric series for deleted streams by purging their labels when streams are deleted. (@squiidz, #4742)
- aws_dynamodb_cdc: Fixed shard discovery aborting on transient failures by allowing partial success and retrying failed shards on the next cycle, enabling convergence with large shard backlogs. (@squiidz, #4737)
- oracledb_cdc: Fix handling of flashback / redo log resets (@josephwoodward, #4760)
- websocket input/output: Connection attempts (dial and upgrade handshake) and input reads now honor context
cancellation, preventing graceful shutdown from hanging on unresponsive servers or idle connections. (@Leward, #490) - websocket output: A failed write now closes the connection instead of leaking it before the reconnect. (@Leward, #492)
Changed
- websocket input: The default value of
max_message_sizehas changed from0(unlimited) to33554432(32 MiB). An unlimited read allows the websocket server to make the process allocate an unbounded amount of memory with a single streamed message. Pipelines that receive larger messages must now setmax_message_sizeexplicitly; a value of0restores the previous unlimited behaviour. (@Leward, #491) - avro: Added avro.input_encoding config to explicitly declare whether messages should be read as Avro JSON or native Go values, replacing automatic inference that could misinterpret bytes fields. (@twmb, #4704)
The full change log can be found here.