I built an AIOps platform that auto-heals Docker services as a student

Описание и применение

WATCHDOG — платформа AIOps, автоматически обнаруживающая и устраняющая сбои Docker‑сервисов, сокращая время простоя до 1.4–1.7 с.

Задача

Traditional monitoring alerts when something breaks, requiring a human to investigate and fix it manually, often at inconvenient times. WATCHDOG automates the entire incident response loop, reducing manual intervention and downtime.

Результат

WATCHDOG achieved 97% reliability in tests, handled 108 events/sec with zero failures, and restored services in 1.4–1.7 seconds, passing all RBAC, multi-tenant isolation, and WAL recovery tests.

Рабочий процесс

• 1. Collect metrics from Docker services.
• 2. Run six parallel detection algorithms (Z-score, Isolation Forest, LSTM, multivariate correlation, seasonal, drift).
• 3. If anomaly confidence ≥65% and severity HIGH/CRITICAL, predict failure and forecast capacity.
• 4. Perform root cause analysis using causal graph and dependency mapping.
• 5. Pass potential heal action through gates: confidence, severity, cooldown, rate limit, circuit breaker.
• 6. Execute Docker SDK restart if all gates pass.
• 7. Log detection, decision, and action to cryptographic audit chain.

Доказательства

40/40 reliability tests passing (97%)

Источник

1.4–1.7s verified auto-heal recovery

Источник