Choose your data mode¶
This guide shows how to choose your data mode in a reliable, repeatable way.
Who this is for¶
- Developers and platform engineers deciding how to integrate and operate RecSys.
What you will get¶
- A simple decision rule for choosing DB-only vs artifact/manifest mode
- The operational tradeoffs (rollback, freshness, complexity)
The two modes¶
DB-only mode¶
Use DB-only mode when you want:
- Fastest time-to-first-success (few moving parts)
- A simple integration pilot (API + tenancy + exposure logging)
- Manual or external control of data updates
You provide signals by writing directly into the serving DB tables.
Start with: Quickstart (10 minutes)
Artifact/manifest mode¶
Use artifact/manifest mode when you want:
- Atomic publish + rollback of signals
- Clear separation of “build” vs “serve”
- A production-like operating model (pipelines produce artifacts, the manifest pointer drives serving)
Start with: production-like run (pipelines → object store → ship/rollback)
Decision rule (one screen)¶
Choose DB-only if:
- You’re piloting integration and do not need automated daily refresh yet
- You can tolerate manual/semi-manual data updates during the pilot
Choose artifact/manifest if:
- You need safe ship/rollback and repeatable daily refresh from day 1
- You want evaluation + releases to move through a controlled publish pipeline
Read next¶
- Data modes in depth: Data modes: DB-only vs artifact/manifest
- Operational reliability: Operational reliability and rollback
- Pilot plan: Pilot plan (2–6 weeks)