Kill Switch Feature Flag Pattern

Kill Switch Pattern

A kill switch is a feature flag designed to instantly disable a feature or service when problems occur. Unlike regular feature flags that enable new features, kill switches are always on in production and get turned off during incidents.

Configuration Example

{
  "payment-processing-enabled": {
    "name": "Payment Processing Enabled",
    "description": "KILL SWITCH: Disable to stop all payment processing during incidents",
    "type": "boolean",
    "enabled": true,
    "defaultValue": true,
    "targeting": []
  }
}

Key Difference from Regular Flags

Aspect	Regular Flag	Kill Switch
Default state	Off (feature disabled)	On (feature enabled)
Action	Turn on to enable	Turn off to disable
Urgency	Planned launches	Emergency response
Lifetime	Temporary (remove after launch)	Permanent
Naming	enable-feature-x	feature-x-enabled or circuit-breaker-x

Implementation in Code

// Kill switch pattern in application code
if (featureFlags.isEnabled("payment-processing-enabled")) {
  processPayment(order);
} else {
  // Graceful degradation
  queuePaymentForLater(order);
  showMaintenanceMessage();
}

Multiple Kill Switches

Complex systems often have layered kill switches:

{
  "external-api-calls-enabled": { "defaultValue": true },
  "payment-processing-enabled": { "defaultValue": true },
  "email-notifications-enabled": { "defaultValue": true },
  "search-indexing-enabled": { "defaultValue": true },
  "analytics-tracking-enabled": { "defaultValue": true }
}

Incident Response Runbook

1. Identify the failing component
2. Toggle the corresponding kill switch to OFF
3. Verify the system stabilizes
4. Investigate the root cause
5. Deploy a fix
6. Toggle the kill switch back to ON
7. Monitor for recurrence

Best Practices

• Label clearly: Prefix descriptions with "KILL SWITCH:" so on-call engineers find them quickly
• No targeting rules: Kill switches should affect all users immediately
• Test the off path: Regularly verify that the degraded experience works correctly
• Fast propagation: Use streaming SDKs to minimize the delay between toggling and effect
• Document dependencies: Note which services are affected by each kill switch