Overview
CDC vocabulary can be confusing if you are new to database operations. This glossary explains the words Synapsor uses around external Postgres/MySQL CDC private preview.
CDC/mirrored subsets are private preview/hardening work. Do not enable CDC for customer production data until the production-readiness signoff is approved.
Use this page when reading the CDC runbook, daily evidence output, operator console status, or RDS smoke artifacts.
Core CDC terms
CDC means change data capture: reading database change logs after an initial setup so a downstream system can learn what changed.
A mirror is the selected copy Synapsor keeps for CDC preview. It is not your full database. It should contain only reviewed tables, columns, and tenant-scoped rows.
A snapshot is an initial copy of selected rows taken before CDC polling catches later changes.
A checkpoint records where a CDC worker stopped so it can continue from the same source position later.
A source position is the database-specific location in the change log. For Postgres this is usually an LSN in WAL. For MySQL this is usually binlog file/position or GTID.
Postgres and MySQL log terms
Postgres WAL means write-ahead log. It is the durable log Postgres writes before data pages change. Logical replication exposes selected changes from WAL.
Postgres LSN means log sequence number. It identifies a position in the WAL stream.
MySQL binlog means binary log. It records row or statement changes for replication and recovery.
MySQL GTID means global transaction identifier. It can identify a transaction position without relying only on a binlog filename and byte offset.
Operational health terms
Lag is how far behind the mirror is from the source database. High lag means Synapsor may not reflect recent source changes.
Capture age is how long it has been since the CDC path last captured a change or checkpoint. A stale capture age can mean the worker is stuck.
Retention is how long the source database keeps WAL/binlog history. If CDC falls behind retention, the worker may not be able to catch up.
Fail closed means Synapsor blocks mirrored reads instead of returning data when CDC state is unsafe, stale, disabled, or needs restore validation.
Restore validation required means an operator must validate the mirror against a known checkpoint after a restore-like event before reads resume.
Resume required means the mirror was paused/blocked and needs an explicit acknowledged resume before reads are allowed again.
Cleanup proof is evidence that disabling CDC left mirrored reads blocked and recorded whether any Synapsor-owned replication artifacts needed cleanup.
Evidence and incident terms
Soak means a repeated operating-time proof window. For CDC, Synapsor currently requires daily evidence over the chosen review window before CDC support can be considered.
RDS smoke means a dev/test-only AWS RDS Postgres/MySQL test stack used to validate connector behavior against real remote databases. It is disabled by default and must be destroyed after tests.
A P0 incident is the highest-severity event, such as data loss, cross-tenant exposure, or serious security failure.
A P1 incident is a severe incident that may block production rollout, such as major restore failure, sustained CDC lag without fail-closed behavior, or billing/security regression.
Developer notes
- If checkpoint or source position is missing, do not call CDC healthy.
- If restore validation or resume is required, mirrored reads should fail closed.
- If lag or capture age is high, investigate before continuing soak.
- If RDS smoke leaves resources running, clean them up before doing anything else.
- If there is a P0/P1 incident, do not approve CDC until it is root-caused and closed.