Cost Optimization Best Practices¶

Azure Functions cost control is primarily an execution-model problem: trigger volume, scale behavior, telemetry ingestion, and storage transaction patterns determine spend more than code size. This guide provides practical controls to keep cost predictable without reducing reliability.

Why This Matters¶

Pick plan economics that match workload shape, not assumptions.

Recommended Practices¶

Use the Consumption free grant intentionally¶

For Consumption, each subscription receives monthly free usage:

• 1,000,000 executions
• 400,000 GB-seconds

Design low-volume workloads to stay inside this envelope when possible.

Practical design implications¶

• Consolidate low-throughput timer and queue functions into one app where operationally appropriate.
• Minimize unnecessary retries by making handlers idempotent and dependency-aware.
• Avoid chatty trigger patterns that produce high invocation count with tiny business value.

Application Insights cost traps and controls¶

Telemetry is frequently the largest cost line item in otherwise cheap serverless workloads.

Common trap¶

• Collecting all traces and dependencies at production volume with no sampling.

Recommended control pattern¶

1. Enable sampling.
2. Keep exceptions and critical requests unsampled.
3. Reduce noisy custom traces.
4. Set retention and daily cap intentionally.

{
  "version": "2.0",
  "logging": {
    "applicationInsights": {
      "samplingSettings": {
        "isEnabled": true,
        "maxTelemetryItemsPerSecond": 5,
        "excludedTypes": "Exception"
      }
    }
  }
}

Basic vs Standard logging workspace choices¶

• Use Basic for high-volume logs where deep analytics is not required.
• Use Standard (Analytics) for richer KQL, cross-table joins, and long investigations.
• Split routing when needed: critical production diagnostics in Standard, bulk low-value logs in Basic.

Storage account cost awareness¶

AzureWebJobsStorage can generate transactions even when functions are idle.

Why this occurs:

• Trigger polling and checkpoint checks.
• Lease coordination for listeners.
• Host state updates and heartbeat activities.

Controls¶

• Separate host storage from business data storage.
• Avoid sharing one storage account across many function apps.
• Monitor transaction metrics and throttling events.

Use scale limits as explicit cost guardrails¶

Set a maximum instance count for event-driven apps to prevent runaway scaling.

For Consumption and Premium:

az resource update \
    --resource-group "<resource-group>" \
    --name "<app-name>" \
    --resource-type "Microsoft.Web/sites" \
    --set properties.siteConfig.functionAppScaleLimit=20

For Flex Consumption, set scaleAndConcurrency.maximumInstanceCount (or equivalent portal setting) aligned to dependency capacity.

Guardrail design rule¶

• Start from downstream limits (database RU/s, API quota, queue visibility timeout).
• Derive maximum safe concurrent instances.
• Set functionAppScaleLimit below the failure threshold.

Durable Functions cost considerations¶

Durable orchestration reliability depends on storage-backed history; this has direct cost impact.

Cost drivers¶

• Large orchestration histories.
• Frequent checkpoint writes.
• Never-purged completed/failed instances.

Controls¶

• Keep orchestrations granular and deterministic.
• Use periodic history purge (durable purge-history) by age and status.
• Limit fan-out size based on downstream throughput budgets.

Right-size plan selection by workload pattern¶

Cost monitoring and alerting¶

Operational cost control requires continuous monitoring, not monthly review.

Minimum monitoring stack¶

• Azure Cost Management budget alerts (monthly and forecast).
• Application Insights ingestion and table growth trends.
• Storage transaction and throttling metrics for host accounts.
• Queue depth/age alerts tied to scale limits.

Alert policy examples¶

1. Budget threshold at 50%, 80%, and 100% of monthly target.
2. Daily Application Insights ingestion anomaly alert.
3. Storage transaction surge alert for AzureWebJobsStorage.
4. Queue age growth while instance count pinned at scale limit.

flowchart TD
    A[Budget threshold breached] --> B[Open cost incident]
    B --> C{Driver signal}
    C -->|App Insights ingestion spike| D[Check sampling, cap, noisy traces]
    C -->|Storage transaction surge| E[Inspect AzureWebJobsStorage listeners and sharing]
    C -->|Queue age grows at scale cap| F[Recalculate max instance and dependency budget]
    D --> G[Apply control changes]
    E --> G
    F --> G
    G --> H[Escalate if forecast still exceeds budget]

Common Mistakes / Anti-Patterns¶

Premium for low-volume workloads¶

• Impact: baseline dominates bill.
• Fix: move to Consumption or Flex unless strict warm latency is required.

No scale limit on queue/event triggers¶

• Impact: burst can cause runaway cost and dependency throttling.
• Fix: set functionAppScaleLimit/maximum instance count and tune trigger concurrency.

Application Insights without sampling¶

• Impact: telemetry ingestion spikes under load.
• Fix: enable sampling and preserve critical telemetry types.

Unused always-ready instances on Flex/Premium¶

• Impact: paying baseline for idle capacity.
• Fix: right-size always-ready/minimum instances from observed traffic percentiles.

Validation Checklist¶

Plan and architecture¶

• Validate plan choice from real trigger profile and networking needs.
• Confirm cold start expectation is documented for stakeholders.
• Confirm always-ready/minimum instances are justified by latency SLO.

Runtime and scaling¶

• Set functionAppScaleLimit or Flex maximum instance count.
• Tune trigger batch/concurrency to downstream limits.
• Validate retry policies to avoid cost-amplifying retry storms.

Observability and storage¶

• Enable Application Insights sampling in production.
• Set telemetry retention and daily caps deliberately.
• Isolate AzureWebJobsStorage for high-throughput apps.
• Define Durable Functions purge schedule.

Governance¶

• Configure budgets and anomaly alerts.
• Review cost monthly with deployment and traffic changes.
• Keep rollback option for plan changes and scale settings.

flowchart TD
    A[Start with workload shape] --> B{Private networking required?}
    B -->|No| C{Low or bursty volume?}
    C -->|Yes| D[Consumption]
    C -->|No| E[Dedicated]
    B -->|Yes| F{Always-on low latency required?}
    F -->|Yes| G[Premium]
    F -->|No| H[Flex Consumption]
    D --> I[Set scale limit and sampling]
    E --> I
    G --> I
    H --> I
    I --> J[Monitor budget, telemetry, storage transactions]

See Also¶

• Operations: Cost Optimization
• Platform: Hosting
• Platform: Scaling
• Operations: Monitoring
• Operations: Alerts

Sources¶

• Azure Functions Consumption plan costs (Microsoft Learn)
• Azure Functions pricing (Microsoft Learn)