Snowflake · Diagnose before optimising

Query Performance Doctor

Why is this workload slow, and what evidence supports the recommended change?

Operational failure

Performance work often starts with a preferred remedy rather than a measured cause. Teams resize warehouses without separating queueing, spill, scan, pruning and join behaviour.

Read-only by default · evidence required · human approval before change

From source evidence to a reviewable finding.

01 · Source evidence

QUERY_HISTORY
Query and operator statistics where available
WAREHOUSE_METERING_HISTORY
Table and clustering metadata
Workload labels and reviewed query sets

02 · Analysis path

01
Group comparable workloads
02
Detect performance symptoms
03
Build and test cause hypotheses
04
Compare architecture and configuration options
05
Define before-and-after evidence plan

03 · Product outputs

Spill, pruning, scan, join and queueing findings
Clustering and search-optimisation recommendations
Materialisation candidates
Workload sizing recommendations
Before-and-after measurement plan

Security contract

The boundary is part of the product.

Data residency: Snowflake Native mode keeps the product workflow in the customer Snowflake account. Microsoft Foundry mode moves selected metadata or telemetry into the customer-controlled Azure boundary.
Permissions required: A dedicated scoped Snowflake role · Read access only to the listed metadata and telemetry views
Write scope: No write capability by default. Any remediation requires human approval and separately granted permissions.
Audit trail: Model calls, retrievals, source evidence, findings, recommendations and approvals are recorded for review.

Deployment choice

Native by default. Azure when the enterprise standard requires it.

Default for sensitive data

Snowflake Native App

Runs in the customer Snowflake account.
Uses Snowpark Container Services when a container runtime is required.
Customer controls installation, grants, compute and network policy.
Private listing first; Marketplace availability is not implied.

For Azure-standardised enterprises

Microsoft Foundry

Agent runtime, model routing, evaluation and tracing in customer Azure.
Snowflake access through customer-controlled OAuth or key-pair authentication.
Selected metadata or telemetry crosses into the designed Azure boundary.
Identity, network, retention and regional placement are explicit design decisions.

Pilot evidence

Decide from a baseline, not a demo.

01
In-scope workload coverage
02
Root causes confirmed by evidence
03
Improvement measured on reviewed queries
04
Recommendations accepted by platform owners

Evaluate Query Performance Doctor against your account.

Describe the decision, platform or operating risk that is slowing you down. We will identify the evidence required and whether consulting, training or a scoped agent is the right response.

Discuss the context