Estuary has shipped several updates to its PostgreSQL change data capture (CDC) connector, aimed at backfill recovery, discovery in large schemas, and schema-change detection. The release is relevant to teams running CDC pipelines where re-reading entire tables is costly or where connectors must handle very large catalogues of tables.
Targeted Backfills are the headline change. PostgreSQL capture bindings now support an Additional Backfill Filter, a SQL expression applied to backfill queries for an individual table. The concrete changes:
- The filter lets teams limit rows included in a recovery backfill, for example using an indexed identifier or a timestamp column.
- It applies to backfills only; ongoing CDC continues to capture new changes independently of the filter.
- Because the expression is injected directly into backfill queries, Estuary flags it as an advanced option that should be configured carefully.
- Changing the filter requires the affected binding to be backfilled again.
Discovery controls address large PostgreSQL schemas. Captures can now control which tables are surfaced during connector discovery, including including only selected schemas and excluding selected schemas. This keeps discovery output manageable when a database holds far more tables than a given capture needs.
The release also covers schema-change detection, though the source summary does not detail the specific mechanics beyond grouping it with the backfill and discovery work.
For developers, the practical value is narrower, cheaper recovery. When a single large table needs re-backfilling, the Additional Backfill Filter can sharply reduce the rows re-read, cutting load on the source database and time to recovery — provided the expression targets an indexed column and is maintained deliberately, since filter changes trigger another backfill. The schema-based discovery filters make onboarding large, multi-tenant or shared PostgreSQL instances more predictable, letting teams scope a capture to exactly the schemas they intend to replicate.