Redpanda Connect v4.100.0 - CDC And Iceberg Fixes

Redpanda Connect v4.100.0 was published on July 9, 2026, with a useful mix of source connector work and correctness fixes for data movement. The main operator story is not volume of changes. It is that Jira ingestion, Iceberg writes, Oracle CDC, Postgres CDC, and timezone handling all received fixes that reduce quiet failure modes in pipelines.

The full release notes and downloads are on the GitHub release page.

The new Jira input component is the only new component in this release. It streams issues, comments, and changelog events into Connect pipelines, which makes it relevant for teams that treat ticket systems as operational data sources.

The important detail is state handling. The input uses cursor based incremental polling and deduplicates boundary issues. That matters for scheduled ingestion, because ticket updates do not arrive as a clean append only log. Poll windows overlap, clocks are involved, and the same issue can sit on the edge of two reads.

For ETL operators, this is the difference between a source that is useful for back office reporting and one that can feed a repeatable pipeline without a cleanup query downstream. The pull request for the component is #4484, and the release notes are on the v4.100.0 release page.

The Postgres CDC fix is small in wording and large in consequence. Redpanda Connect now adds an acknowledgement barrier so snapshot rows are fully acknowledged before promoting the replication slot. Without that ordering, a crash at the wrong point can turn a snapshot transition into data loss.

This is the kind of failure that does not show up as a noisy pipeline error. It shows up later as a missing row that looks like application behavior until someone compares source and sink state. The fix is tracked in #4584.

Oracle CDC also gets multiple hardening changes around LogMiner SQL parsing. Redpanda Connect now supports bare rowid based predicates in SQL_REDO, with better debug logging for query failures in #4585. The parser and LOB assembly path also received fixes for hex decoding, UNISTR national character support, LOB offset validation, and empty primary key sets in #4586.

Those Oracle changes are not cosmetic. Parser panics and corrupt LOB reconstruction are pipeline trust problems. If an operator cannot tell whether a failed row is a source oddity, a parser limitation, or a data corruption bug, the connector becomes hard to keep in an automated CDC path.

The Iceberg sink has two changes worth separating. First, commit retries are now idempotent. The release notes call out the case where a commit lands on the server side, but the client observes a transient failure. Retrying that operation without idempotency can create duplicate data files, which then pushes the repair work onto table maintenance and downstream consumers. The fix is in #4591.

Second, Redpanda Connect now caches Iceberg schema field metadata to reduce per record allocations. This is a throughput oriented change rather than a semantic one, but it matters for high volume sinks where allocation pressure turns into garbage collection time. The release also adds tuning documentation with recommended configuration for throughput optimization, covered by #4590.

Taken together, the Iceberg work is about making writes less fragile under retry and less expensive under load. That is the right failure surface to watch for lakehouse sinks, especially when transient object store, catalog, or network failures are part of normal operation.

The general runtime fix embeds the IANA timezone database in the CGO enabled build. The notes state that this makes timezone aware features work correctly in minimal container runtimes.

The unreleased note gives the concrete symptom. With the timetzdata build tag, time.LoadLocation works without system tzdata instead of silently falling back to UTC. That fallback can shift JQL date predicates in the Jira input, which is exactly the sort of time boundary bug that creates missing or duplicated records around polling windows.

This is a practical container packaging fix. Minimal images are common for data services, and they often remove packages that application code later assumes are present. Embedding timezone data narrows that runtime dependency.