Context
Tracking issue for follow-up work on the Postgres CDC input connector (landed in #5988). Collects the smaller fixes and improvements from PR review that were deferred.
Items that need discussion have their own separate issues.
Checklist
Small fixes (clear, ~minutes each)
Real bug
Config/validation
Testing
Documentation
Context
Tracking issue for follow-up work on the Postgres CDC input connector (landed in #5988). Collects the smaller fixes and improvements from PR review that were deferred.
Items that need discussion have their own separate issues.
Checklist
Small fixes (clear, ~minutes each)
PgNumericNaN/Infinity →Value::Nullincell_to_jsonandarray_cell_to_jsonFelderaDestination::is_target_table; default topublic.when unqualifiedtracing::warn!whenwrite_table_rowsfalls back tocol_0, col_1, ...due to missing relation info%.fsuppresses trailing zeros)// NOTE:comment nearResume::Seekconstruction explaining identity-vs-position semanticsPostgresStorecreates anetl.*schema on the source DB inPostgresCdcReaderConfigdocstringReal bug
PostgresCdcReaderConfig::tlsthrough to etl'sTlsConfig. Currentlyparse_pg_urihardcodesTlsConfig::disabled(), so thetlsconfig field is silently ignored.Config/validation
uri,publication,source_tableat config parse time (fail at deploy, not pipeline start)(uri, publication, source_table)would compute the samepipeline_idand collide on the replication slotTesting
source_tablefiltering)TRUNCATEevents (currently logged and ignored — verify no crash)REPLICA IDENTITYwithoutFULL(updates/deletes without old row)Documentation
docs.feldera.com/docs/connectors/sources/covering setup, privileges, config, limitations, known issues