Watch an HTTP endpoint

Use a watch-preview binary from the source checkout. This example polls every five seconds, opens an incident after three actual 5xx responses, and recovers after two nonmatching observations. HTTP 4xx responses are nonmatching here; choose a different condition if those should also open an incident.

apiVersion: ding.ing/v1alpha1
kind: Destination
metadata:
  id: ops-webhook
spec:
  type: webhook
  urlRef: {env: OPS_WEBHOOK_URL}
---
apiVersion: ding.ing/v1alpha1
kind: Watch
metadata:
  id: api-health
  name: API health
spec:
  source:
    type: http
    url: https://api.example.com/health
    every: 5s
    timeout: 2s
  condition:
    field: http.status
    operator: gte
    value: 500
  policy:
    trigger: transition
    consecutive: 3
    recoverAfter: 2
    onUnknown: hold-incident
  destinations:
    - ref: ops-webhook
      events: [firing, recovered, source_error, source_recovered]

Download the complete manifest. Replace the example URL with your endpoint. Set OPS_WEBHOOK_URL in the daemon's environment before starting it, or change the destination to console output for a local trial.

./ding validate examples/watches/api-health.yaml --json
./ding apply examples/watches/api-health.yaml --state-dir ./ding-state --dry-run --json
./ding apply examples/watches/api-health.yaml --state-dir ./ding-state
./ding watch inspect api-health --state-dir ./ding-state --json

Test without a live service

./ding test examples/watches/api-health.yaml \
  --events testdata/watches/api-health.jsonl --json

The fixture returns one firing and one recovery. It performs no source I/O or delivery. Its inputs are 500, 502, 503, 200, 200, five seconds apart.

A timeout is unknown input, not an HTTP 5xx observation. A successful 304 advances freshness without counting as a new condition sample. Redirects are rejected. Source semantics explain projection and limits.