Run Performance Profiler

Performance Profiler heatmap

Prerequisites

perfprofile.read and access to Azure Monitor metrics for the selected connection/scope.
Current workload inventory for workload mode; optional AI provider for narrative and connector for tickets.

Route

Open /performance; top-level views are 🔥 Profiler, 🚀 Fleet, and 🧹 Cleanup.

How to profile a workload or subscription and choose a window

Open Profiler and choose workload or subscription scope, not both.
Confirm connection and inventory.
Select a preset ISO-duration window such as one or seven days, or the supported custom range. Use short windows for spikes and longer windows for recurring patterns.
Select Run profile and monitor streamed progress; the run persists and can continue across navigation.
Select the completed row in Profile history when it is not already displayed.

Expected result: A persisted profile contains resource metrics, baseline states, bottlenecks, scorecard, narrative when available, and exact run window.

Verification: Confirm scope, run time, displayed start/end window, profiled/resource counts, scan-cap warnings, and connection status.

How to analyze the heatmap and all resources

Open Heatmap and start with highest bottleneck scores and red/amber cells.
Inspect metric value, threshold/baseline, resource, type, region, and available trend/detail.
Filter resource types or choose a resource to narrow the matrix.
Open All Resources for the full searchable/filterable virtualized resource list.
Correlate with deployments, scaling, logs, dependencies, and user-impact telemetry.
Treat green as “did not cross this baseline,” not proof of service health.

Expected result: A small set of candidate bottlenecks is supported by metric observations.

Verification: Reproduce important values in Azure Monitor for the same resource, aggregation, and time window.

How to use narrative, findings, tickets, evidence, and PDF

Read the AI narrative as a hypothesis and compare every claim with the matrix.
Select 🛡️ Register findings to create Performance-pillar findings from current bottlenecks.
For a specific bottleneck, choose 🎫 Ticket and the intended connector.
Select 🗄 Evidence to capture the currently viewed run as an immutable Evidence Locker snapshot.
Select 📄 PDF for the current or historical run; wait for generation or cancel the request.
Open Assessments/Evidence/external ticket and confirm the handoff.

Expected result: Validated bottlenecks have traceable findings, ticket/evidence records, or a report.

Verification: Match scope, run ID/time, resource, metric, threshold, and window in each handoff.

How to operate fleet profiling

Open 🚀 Fleet and review latest score, breaches, top bottleneck, staleness, and failed/never-profiled rows.
Filter/sort, select a bounded set, and launch supported mass profiling.
Let background runs continue; do not submit duplicates while a row is running.
Retry failed rows after checking throttling/access, then open each workload’s profile.

Expected result: Fleet rows update with terminal latest profiles and clear stale/error state.

Verification: Confirm each selected workload’s profile time/window and drill-down result.

How to use history and cleanup

Use Profile history to select comparable runs and download a run-specific PDF.
Move obsolete runs to Trash first; restore if required.
Open 🧹 Cleanup for bulk retention review.
Purge individual or all trashed runs only after evidence/report retention is satisfied.

Expected result: Useful history remains available and obsolete data follows recoverable-then-permanent deletion.

Verification: Restored runs reopen; purged runs do not; evidence snapshots remain separate records.

Safety and rollback

Profiling is read-only against Azure; it does not change thresholds/resources.
Metrics are delayed/aggregated and scan caps may omit resources.
AI causality is untrusted. Correlate before ticket/remediation.
Findings/tickets/evidence are records; correct them in their owning systems.
Trash is rollback for run deletion; purge is irreversible.

Troubleshooting

Symptom	Resolution
No resources	Check scope, inventory, connection, and metric-read access.
Many no-data cells	Check metric support, window, aggregation, provider delay, and permissions.
Refresh is slow/fails	Narrow scope/window, inspect streamed error, and allow Azure backoff.
Result is stale	Select current scope and run an explicit profile.
Score conflicts with experience	Review service-level telemetry, dependencies, and baseline suitability.
Evidence/PDF unavailable	Select a completed persisted run and verify storage/permission.