Deployment

DBReplay ships as one combined capture/replay container image, with the role and privileges selected at deployment. Capture and replay are never deployed to the same location — the captured artefact moves between them.

Prerequisites

Item	Detail
Capture point	Wherever the source traffic actually is — see Capture Topology. Confirmed by a trial capture before anything else is scheduled.
Capture-side privileges	Raw socket / packet-capture capability on the capture host. No target credentials — the capture role is never handed anything the replay role needs.
Replay placement	Same VPC and availability zone as the target instance. Client-to-server latency across a WAN link contaminates every timing the run records.
Source catalog extract	Required alongside the capture — trigger bodies, view definitions and check constraints contain SQL that never crosses the wire. Oninit supplies the extraction script; taken before any rebind or DDL that postdates the capture.
Target credentials	From a secret store, provisioned to the replay role only — never derived from the capture artefact, even where the capture method can see an authentication exchange on the wire.
Target environment	A disposable instance, restorable between runs. Comparison report validity depends on every run starting from a known, consistent state.

Engagement steps

Trial capture. A plain packet capture, no DBReplay software involved, confirming the capture point can keep up at real volume. No unusual security approval needed.
Scoping call. Confirms topology, engine pair, catalog extract logistics, and any environmental factors (data sharing, encryption, client platform) that shape the capture strategy.
Deploy capture at the agreed point, and confirm keep-up rate and coverage against the trial capture's own numbers.
Catalog extraction, run by the customer's team using the supplied script, timed before any rebind or DDL that would postdate the capture.
Pre-process runs against the capture and the catalog extract, producing the first coverage report — reviewed with the customer before any replay is attempted.
Deploy replay next to the provisioned target, and confirm every captured user can connect — a pre-check run before the real replay, since a connection failure discovered mid-run costs a full restore.
Run, watch, and report. The read-only subset of the workload runs first where possible, then the full transactional replay, with live status visible throughout.

Engine coverage notes

Role	What's needed
Source engine	A wire decoder — the specialised, expensive half of engine coverage. See Engines & Platform Support.
Target engine	A driver and an emitter — comparatively inexpensive, since the dialect changer's neutral intermediate form means a new target doesn't require new source-side work.