1.23.0 - 2026-11-16
Added
- WASM-based plugin system for Bento @gregfurman
tlsfield added tosocketoutput enabling TLS connections whennetworkistcp@weird-oecophyllaelasticsearch_v2output fieldsretry_on_conflict(let Elasticsearch retryupdate/upsertactions on version conflicts) andnum_workers(number of BulkIndexer workers per batch;0keeps the library default) @JoranDoxaws_bedrock_chatprocessor enables sending messages to LLMs via the model-agnosticConverseAPI @zcmacklogger.filefieldsrotate_max_size_mb,rotate_max_backups&rotate_compress, which were previously fixed at 10 MB, one backup & gzip @piratecarrot- CLI flag
--remote-configenables fetching Bento stream config from HTTP servers on start @Omar1H1 & @jem-davies
Changed
- Shutdown timeouts that may cause duplicate delivery are now logged at
WARNinstead ofINFO@piratecarrot - The Hugging Face processors use hugot v0.8.3, which replaces the ONNX Runtime backend, and
enable_downloadnow creates thepathdirectory when it is missing @slachiewicz gcp_pubsubcomponents migrated to v2 - deprecatessyncfield as a no-op @jem-davies
Fixed
sftpoutput now drops its client after a lost connection, so the nextConnectre-establishes it instead of reusing the dead client and never delivering messages again @pcbeingused333elasticsearch_v2output no longer acks a batch whose whole_bulkrequest failed (unreachable cluster, non-2xx response):github.com/elastic/go-elasticsearch/v9is bumped to v9.3.6, which reports such failures per document, so they reachfallback/nacks like any other write error @jaascoservice.AutoRetryNacksandservice.AutoRetryNacksBatchedno longer drop a pending retry when a read is cancelled during backoff @ReguiguiMohamedelasticsearch_v2output now aborts the in-flight_bulkrequest when the write context is cancelled, instead of leaving it running in the background untiltimeout@pcbeingused333aws_kinesis_firehoseoutput includes a check to ensure request payload size is <4MiB @alecmerdlernsqinput no longer keeps consumed messages in memory until shutdown which was causing unbounded memory usage @mwchattenhttpprocessor now sends its requests with the processing context, so a forced stop of the stream ends their rate limit waits instead of leaving them to retry forever @weird-oecophylla
New Contributors
- @zcmack made their first contribution in #1071
- @piratecarrot made their first contribution in #1097
- @mwchatten made their first contribution in #1095
- @JoranDox made their first contribution in #1120
- @weird-oecophylla made their first contribution in #1128
- @jaasco made their first contribution in #1134
- @pcbeingused333 made their first contribution in #1147
Full Changelog: v1.22.0...v1.23.0