Configuration

Ruptura is primarily configured via environment variables. A ruptura.yaml config file is optional — pass it with --config /path/to/ruptura.yaml.

Environment variables

These are the most important variables for getting started:

Env var	Default	Description
RUPTURA_API_KEY	(empty)	Bearer token for all API requests. If empty, authentication is disabled (dev only).
RUPTURA_INGEST_RPS	1000	Token-bucket rate limit (requests/second) on the ingest server
RUPTURA_LOG_LEVEL	info	Log verbosity: debug | info | warn | error

CLI flags

Flag	Default	Description
--port	8080	REST API + Prometheus remote_write server port
--otlp-port	4317	OTLP ingest server port (metrics, logs, traces)
--storage	/var/lib/ruptura/data	BadgerDB data directory
--config	(none)	Path to optional ruptura.yaml config file

Full config file reference

# ruptura.yaml

mode: connected          # connected | stateless | shadow

ingest:
  http_port: 8080        # REST API + Prometheus remote_write
  otlp_port: 4317        # OTLP HTTP ingest (metrics, logs, traces)
  grpc_port: 9090        # gRPC ingest

eventbus:
  driver: none           # none | nats | kafka
  # nats:
  #   url: "nats://localhost:4222"
  #   stream: ruptura-events
  # kafka:
  #   brokers: ["localhost:9092"]
  #   topic: ruptura.events

ensemble:
  adaptive: true         # Online MAE-based weight adaptation (recommended)
  # weights updated every 60s over a 1-hour sliding window when adaptive: true

predictor:
  rupture_threshold: 3.0          # FusedR ≥ threshold → Warning
  confidence_thresholds:
    auto_action: 0.85             # confidence required for Tier-1 auto-action
    alert: 0.60                   # confidence required to raise an alert

baselines:
  observation_window: 96          # observations before adaptive baselines activate
                                  # default 96 × 15s interval ≈ 24h

actions:
  execution_mode: suggest         # shadow | suggest | auto
  safety:
    rate_limit_per_hour: 6        # max Tier-1 actions per target per hour
    cooldown_seconds: 300         # min gap between actions on the same target
    namespace_allowlist:          # only act on pods in these namespaces
      - production
      - staging

auth:
  api_key: ""                     # set via RUPTURA_API_KEY env var

storage:
  path: /var/lib/ruptura/data     # BadgerDB data directory
  retention:
    kpis_days: 400                # long-term KPI retention

Key options

mode

Value	Behaviour
connected	Full rupture detection, actions, and XAI
stateless	No persistence — useful for testing ingest
shadow	Actions computed but never executed (dry-run)

actions.execution_mode

Value	Behaviour
shadow	Actions logged only — nothing executed
suggest	Actions queued, waiting for /api/v2/actions/{id}/approve
auto	Tier-1 actions executed automatically with safety gates

For production, start with suggest to validate Ruptura's recommendations before enabling auto.

baselines.observation_window

The number of 15-second observations before Ruptura switches from global thresholds to per-workload adaptive baselines. At the default of 96 (≈24h), you will see more initial alerts during the learning period — this is expected. After the window, false positives from batch jobs and normally-loaded services drop significantly.

ensemble.adaptive

When true, Ruptura computes per-model MAE over a 1-hour sliding window and normalises prediction weights every 60 seconds. Models with lower error receive more weight. The five models are: CA-ILR, ARIMA, Holt-Winters, MAD, EWMA.

Minimal production setup (Helm)

helm install ruptura helm \
  --namespace ruptura-system \
  --create-namespace \
  --set apiKey=$(openssl rand -hex 32) \
  --set persistence.size=20Gi \
  --set serviceMonitor.enabled=true

Minimal production setup (Docker)

docker run -d \
  --name ruptura \
  -p 8080:8080 \
  -p 4317:4317 \
  -v ruptura-data:/var/lib/ruptura/data \
  -e RUPTURA_API_KEY=$(openssl rand -hex 32) \
  -e RUPTURA_INGEST_RPS=5000 \
  -e RUPTURA_LOG_LEVEL=info \
  ghcr.io/benfradjselim/ruptura:6.7.0