CASE STUDY 07 · Reliability & Operations
Datastream CDC Reliability & Data Consistency Investigation
A CDC reliability investigation comparing stream health, backfill signals, source access, and destination data to reason about end-to-end completeness.
Incident investigatedFull historical recovery unverified
DatastreamBigQueryPostgreSQLCDC
Engineering challenge
Operational status and destination checks did not provide the same evidence of synchronization completeness, while source access depended on consistent identity and grant configuration. The investigation needed to distinguish healthy control-plane signals from verified data outcomes.
Architecture & design
- Evaluate the source grant and connection-profile identity as one permission path.
- Compare stream state and backfill progress with data observed in the destination.
- Treat destination completeness as a separate validation signal from stream health.
Implementation
- Compared stream state and backfill progress with data observed at the destination.
- Reviewed the source grants alongside the identity configured for the connection profile.
- Investigated a source-access configuration mismatch across environments and separated it from destination completeness checks.
- Documented why stream health and backfill status need independent data-level verification.
Validation & impact
- The investigation distinguished control-plane status from destination data checks and identified source identity configuration as a separate validation point.
- The findings show that stream state and backfill status alone are insufficient evidence of end-to-end data completeness.
Limitations & next steps
- A repeatable automated audit is not yet evidenced.
- Row-count comparison, freshness checks, lag monitoring, principal consistency checks, and historical recovery remain follow-up validation work.