CSA-in-a-Box vs Microsoft Fabric — feature-by-feature¶

Last updated: 2026-05-08

This page is a side-by-side capability comparison between CSA-in-a-Box (this repository's reference architecture, running on Azure PaaS) and Microsoft Fabric (the unified analytics SaaS).

If you're trying to decide which one to build on, start with the Fabric vs Databricks vs Synapse decision tree or ADR-0010 — Fabric strategic target. This page complements those by giving you the per-capability detail so you can validate the recommendation against your specific workload.

Positioning at a glance¶

The repo is built for forward migration into Fabric — every primary technology choice (ADR-0001 through ADR-0009) has a documented Fabric counterpart. See the migration paths below.

Capability matrix¶

Storage and table format¶

Fabric advantage: Single namespace (OneLake) for the whole tenant — no per-storage-account sprawl. CSA-in-a-Box mimics this with consistent naming + ADR-0006 Purview cataloging.

CSA-in-a-Box advantage: Direct ADLS access from any Azure service; zero workspace abstraction cost; Gov-cloud ready today.

Data ingestion / orchestration¶

Fabric advantage: Mirroring connects operational DBs (Cosmos, SQL, Snowflake) into the lakehouse with no ETL — true zero-ETL for the covered sources.

CSA-in-a-Box advantage: Full control over SHIR and on-prem connectivity; Gov-compatible.

Transformation¶

Fabric advantage: Lineage is automatic across Fabric items — no manual Purview registration step. Single notebook environment.

CSA-in-a-Box advantage: Best-in-class Spark (Databricks Photon benchmarks ahead of Fabric Spark for tuned workloads); real Unity Catalog if you need it; ML ecosystem maturity.

Governance and catalog¶

Same on both: Microsoft Purview is the governance plane. The coverage in Fabric is wider out-of-the-box (every Fabric item is catalog-aware), but you can reach the same coverage on CSA-in-a-Box by disciplined Purview registration.

BI and serving¶

Fabric advantage (significant): Direct Lake is genuinely unique. If your primary requirement is Power BI on a lakehouse with sub-second refresh, Fabric is the right answer. CSA-in-a-Box does not match Direct Lake performance — it cannot, by architecture.

Real-time and streaming¶

Fabric advantage: Reflex (Data Activator) is unique — declarative rules over streaming data with no-code action triggers.

Machine learning and AI¶

Fabric advantage: The Copilot and Data Agents experience is built in — no glue code. AI Skills are a managed RAG pattern.

CSA-in-a-Box advantage: Full control over the Copilot pipeline (this very widget is the proof). Bring-your-own model, bring-your-own guardrails. Gov-compatible Azure OpenAI today.

Deployment, IaC, CI/CD¶

Same on both: Both can be Git-driven. CSA-in-a-Box is more mature (everything goes through PR + CI today); Fabric Git integration is catching up but adds platform-specific quirks.

Compliance and Gov-cloud posture¶

This is the headline differentiator. Fabric is Commercial-only as of 2026-05-08. Federal, defence, and many state customers cannot adopt Fabric today — and CSA-in-a-Box is what they should build on.

Cost model¶

For a single Power BI-centric workload, Fabric's bundled pricing typically wins. For a workload that's heavy on Spark + ML + 24/7 streaming, separate-service pricing in CSA-in-a-Box is often cheaper because you tune each component.

Migration paths¶

Today's CSA-in-a-Box → Fabric (when Gov GA arrives)¶

This is what ADR-0010 is built for. Most migrations are item-level moves rather than rewrites:

The repo's adr/ directory documents the per-component forward path; together they constitute a migration runbook.

Microsoft Fabric → CSA-in-a-Box (Gov customer leaving Fabric pilot)¶

Less common but happens — a customer pilots Fabric in Commercial, then needs to ship the production workload in Gov:

The asymmetry is real: Fabric's unique features (Direct Lake, Reflex, Data Agents) don't map cleanly back to PaaS. If you start in Fabric, you accept that some artifacts will need re-implementation if you ever need to leave.

When to choose which (executive summary)¶

Choose Microsoft Fabric if all of these are true:

• You're on Azure Commercial (not Gov)
• Power BI is central to your delivery
• You want the simplest possible operational footprint (single CU pool, one workspace per domain)
• You don't need fine-grained Spark tuning
• You're comfortable with Microsoft-SaaS coupling

Choose CSA-in-a-Box if any of these are true:

• You need Azure Government (FedRAMP High / IL4 / IL5) today
• You have an existing Synapse Dedicated or Databricks investment to evolve, not replace
• Spark / ML workloads are heavy enough to want Photon + Unity Catalog
• You need cross-cloud read scenarios that exceed OneLake shortcut capabilities
• You want everything in Bicep + Git + GitHub Actions with zero SaaS-only surfaces
• Your security posture requires per-resource isolation (per-RG, per-NSG, per-Key-Vault) that Fabric workspaces don't expose

Choose both: A common pattern is Fabric for interactive analytics + Power BI in Commercial, plus CSA-in-a-Box for the Gov environment or for Spark/ML pipelines that feed Fabric via OneLake shortcuts. The two are not mutually exclusive — many CSA-in-a-Box customers will end up with hybrid topologies.

Related¶

• ADR: 0010 — Microsoft Fabric as strategic target
• Decision tree: Fabric vs Databricks vs Synapse
• Issue tracking Gov GA: #177 Microsoft Fabric availability + setup guidance for Azure Government
• Companion site: Supercharge Microsoft Fabric
• Service availability tracker: GOV_SERVICE_MATRIX.md

See also¶

• ← Previous: Microsoft Fabric Platform Guide
• → Next: Migration paths
• ⌂ Index: Documentation home