Reapit — Target Architecture

What is changing, and why

The redesign puts an Azure Data Lake Storage backup database between Snowflake and Fabric. It exists for two reasons, both about disaster recovery and compliance:

Raw availability. A complete copy of all the tables, reportable in their raw form should it ever be required — exposed through the Raw Schema.
Point-in-time recovery. The lake is backed up by Databarracks, so the entire reporting platform can be rebuilt from a point-in-time backup should the need arise.

The flow

Snowflake

Snowflake (Reapit RPS)

Source system — REAPIT_RAW, SAU / SAV / PP schemas

↓Orchestration — driven by orchestration table / config file

Azure — Data Lake Storage

Staging — incremental extract

Only the delta can be pulled from Snowflake — changed rows since the last watermark land here as-is

40 tables hourlyRest nightlyDelta only

↓Merge — changed rows upserted into the full tables

Azure Data Lake Storage — backup database

Durable copy of the full Reapit estate.

Full data14 yearsAll tablesMerged

⇆

Databarracks

Back up / restore to a point in time — full recovery of the Data Lake and everything built from it

↓Shortcut — no copy

Microsoft Fabric

Reapit Bronze workspace

Bronze — shortcut to Data Lake

Raw data exposed via OneLake shortcut — no data movement into Fabric.

90 tablesFull data14 yearsShortcut

↓ ↓

Reapit SilverGold workspace

Raw Schema

Silver (Raw)

Raw · shortcut through from Bronze

↓

Gold (Raw)

Raw · shortcut — exposes all 90 tables in raw form to consumers

Reporting Schema

Silver (Reporting)

Merged full data · merges only changed data incrementally from Bronze · 40 tables · once per hour

↓

Gold (Reporting)

ODS views over Silver, plus the dimensional model (facts and dimensions)

The two schemas side by side

	Raw Schema	Reporting Schema
Purpose	Expose the raw Reapit estate untouched	Business-ready reporting — ODS views and the dimensional model
Tables	90 — everything	40 — reporting set
Mechanism	Shortcuts all the way — no data movement	Incremental merge of changed data from Bronze
History	Full data, 14 years	Merged full data, maintained incrementally
Cadence	Follows the lake — 40 tables refresh hourly, the rest nightly	Once per hour, in step with the hourly lake refresh of its 40 tables
Compute cost	Near zero (shortcuts)	Merge notebooks, as today