Nomad Orchestration

Job Structure

Platform Jobs

Run in the default namespace:

• chelar-api — Go API (port 8080)
• chelar-dashboard — Next.js dashboard (port 3000)
• chelar-chromium — Shared Chromium browser service

Tenant Jobs

Prefixed with tenant-{id} (e.g., tenant-abc). Each job runs a single Docker container with:

• Memory: 1,024 MB (OpenClaw) or 256 MB (ZeroClaw)
• CPU: 300 MHz minimum
• Network: Bridge mode with dynamic port mapping
• Volumes: JuiceFS bind mount at /data/tenants/{slug}

Nomad Integration

The Go API uses hashicorp/nomad/api to:

1. Register jobs — POST /v1/jobs with the computed job spec
2. Query status — GET /v1/job/{id} for container state
3. Stop/start — DELETE /v1/job/{id} or re-register
4. Logs — GET /v1/client/fs/logs/{allocID}

Key files:

• api/internal/orchestrator/nomad.go — job spec builder
• api/internal/orchestrator/orchestrator.go — tenant lifecycle

Health Monitoring

The API polls Nomad allocation status and maps it to dashboard states:

• running → Running
• pending → Starting
• dead (no job) → Suspended
• dead (job exists) → Stopped/Error

Idle Suspension

A background goroutine checks for tenants with no recent activity (72h) and no proactive features. It stops the Nomad job but preserves the job definition for quick restart.