Simulating Outages¶

Test how your chain handles failures before they happen in production.

Basic Simulation¶

fallbackrabbit test my-chain.yaml

This runs 5 test prompts with all providers healthy, showing latency and success rates.

Outage Scenarios¶

Create an outage scenario file to inject specific failures:

# outage-scenario.yaml
outages:
  - provider: GPT-4
    error_type: rate_limit
    duration_prompts: 3
    status_code: 429

  - provider: Claude
    error_type: timeout
    duration_prompts: 2

  - provider: Gemini
    error_type: server_error
    duration_prompts: 5
    status_code: 503

Run with the scenario:

fallbackrabbit test my-chain.yaml --outages outage-scenario.yaml

Outage Fields¶

Field	Type	Required	Description
`provider`	string	✅	Provider name to affect
`error_type`	string	✅	Error type to simulate
`duration_prompts`	int	✅	How many prompts the outage lasts
`status_code`	int	❌	Specific HTTP status code
`error_message`	string	❌	Custom error message

Simulation Output¶

The simulator reports:

Per-prompt results — Which provider handled each prompt, latency, retries
Chain report — Success rate, average latency, total tokens, provider usage breakdown
Fallback path — Visual representation of which failovers occurred
Timing — Retries used, total wait time per prompt

Verbose Mode¶

fallbackrabbit test my-chain.yaml --outages outage-scenario.yaml --verbose

Shows detailed per-prompt breakdown including error messages and retry history.