ingestr v1.1.8 - CDC Schema Evolution and Partitioned Replace

ingestr v1.1.8, published on July 22, 2026, fixes CDC schema evolution across destinations. The release also adds operational cleanup for PostgreSQL replication slots and broadens ingestion support for partitioned replace jobs and changing Fabric column types.

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

The main correctness change addresses CDC schema evolution across destinations. Schema changes are a difficult boundary for replication jobs because the source event stream and destination table can move through different schema states. The changelog does not identify the affected destination combinations or the exact failure mode, so it does not support a more specific compatibility claim.

For operators, the practical test is direct. Run an expected source schema change in a staging replication job, then confirm that the destination schema and subsequent rows remain consistent. This release is relevant to that test, but the notes do not publish a destination matrix or a migration procedure.

The command line interface also gains a delete command for CDC replication slots. That gives operators an explicit cleanup path when a PostgreSQL CDC job is retired or rebuilt. Stale slots can retain write ahead log data on the source, so slot lifecycle belongs in the same runbook as job removal. The release notes do not provide the command syntax. Check the installed command help before automating cleanup, and verify the slot target before deletion.

Partitioned replace ingestion is now supported. This matters for jobs that refresh bounded slices of a table, such as a date partition, instead of treating every replace load as one broad operation. The notes do not define partition selection rules, overwrite boundaries, or destination coverage. Test those semantics against a disposable target before changing a production schedule.

The other data model change adds support for Fabric column type changes. Pipelines writing to Fabric can now account for a source column changing type rather than assuming a fixed shape for the lifetime of the load. The terse changelog leaves conversion rules and incompatible type pairs undocumented. A representative widening and narrowing test is still necessary for any schema that changes under load.

Together, these changes extend mutable ingestion paths in two different directions. Partitioned replace changes which rows a job may replace, while Fabric type support changes how the destination schema may evolve. Neither entry claims a performance improvement, and neither should be read as one.

The v1.1.8 metrics handler excludes the standard expvar built ins from /debug/vars. This reduces unrelated runtime output in an endpoint used for ingestr diagnostics. It also changes what existing scrapers may receive. Any dashboard or alert that reads fields from that endpoint should be checked after the upgrade rather than assuming its current field set remains intact.

Capture the endpoint keys before the upgrade, repeat the request after it, and confirm that collectors tolerate missing runtime fields. This is an observability compatibility check, not a data validation exercise. The release notes make no broader claim about the metrics contract.