loadkit

loadkit fires a synthetic HTTP burst in parallel with the chaos action it wraps. Useful to validate behavior under concurrent load — e.g. "does the service still respond 2xx while a pod is being evicted?".

Config block

load:
  vus: 10                    # required — number of parallel virtual users
  duration: 30s              # required — how long the burst lasts
  skipTlsVerify: false       # optional — disable TLS certificate verification for internal/self-signed endpoints
  requests:
    method: GET              # required — GET or POST
    url: https://my-app.example.com/healthz   # required
    interval: 100ms          # optional — per-VU delay between requests (default 100ms)
    contentType: application/json  # optional — sets Content-Type header (defaults to application/json when body is set)
    body: '{"key":"value"}'  # optional — request body for POST

Field reference

Field	Type	Default	Notes
load.vus	integer	—	Required, > 0. Number of parallel goroutines firing requests.
load.duration	duration	—	Required, > 0. Must be < scenario.interval if the module is periodic.
load.skipTlsVerify	boolean	false	Disables TLS verification. Use only for internal/self-signed endpoints you explicitly trust.
load.requests.method	enum	—	Required. GET or POST.
load.requests.url	URL	—	Required. Must start with http:// or https://.
load.requests.interval	duration	100ms	Per-VU sleep between two requests. Must be > 0.
load.requests.body	string	—	Optional request body. Typically used with POST.
load.requests.contentType	string	application/json	Content-Type header. Auto-set to application/json when body is present.

Behavior

When a wrapped module's Run is called, the middleware:

1. Immediately spawns vus goroutines bound to a context with a duration timeout.
2. Each goroutine loops: fire a request, sleep requests.interval, repeat — until the context expires.
3. The chaos action runs in parallel in the main goroutine.
4. The loadkit.Supervisor holds a WaitGroup so the burst is drained at shutdown.

Body handling is deliberately minimal: the response body is io.Copy-ed to io.Discard and closed. No retries, no assertions — this is a traffic generator, not a test client. Use testkit if you want to assert on the outcome.

Metrics

Every request is tracked via Prometheus:

Metric	Type	Labels
chaos_loading_http_active	gauge	name, method, url
chaos_load_requests_total	counter	name, method, url, status
chaos_load_request_duration_seconds	histogram	name, method, url

status is bucketed into 2xx / 3xx / 4xx / 5xx / 1xx / error (error covers network failures and context cancellation).

The name label is the module name — not the load config — so metrics join naturally with chaos_test_success.

Limitations

• One endpoint per module. Multi-endpoint load patterns require multiple modules.
• No custom headers / auth (yet). The generator targets public or token-in-URL endpoints.

These are deliberate tradeoffs — loadkit is meant to generate realistic-enough load during chaos, not to replace k6 / Gatling.

If skipTlsVerify is enabled, loadkit logs a warning and skips server certificate verification for that endpoint only.