Per-container Docker stats in Overseer dashboard

[lbedner]

Context

The server runs 11 Docker containers on a 1.9GB DigitalOcean droplet (currently 72.5% memory). The Overseer dashboard shows host-level CPU/memory/disk via psutil, but there's no way to see which container is the memory hog. This adds per-container memory/CPU breakdown.

Approach

• Query Docker Engine API via Unix socket using httpx (already a dependency — no new packages)
• Add as a sub_component of backend alongside existing cpu/memory/disk metrics
• Display in a new "Containers" tab in the Server/Backend modal, sorted by memory usage descending
• Graceful degradation when Docker socket isn't available (dev machines)

Files to Modify

1. docker-compose.yml — Mount Docker socket for webserver

Add /var/run/docker.sock:/var/run/docker.sock:ro to webserver volumes

2. docker-compose.dev.yml — Same for dev

Add socket mount alongside existing code mount

3. app/core/config.py — Add settings

• DOCKER_SOCKET_PATH: str = "/var/run/docker.sock"
• DOCKER_STATS_CACHE_SECONDS: int = 10
• DOCKER_STATS_ENABLED: bool = True

4. app/services/system/health_docker.py — NEW: Docker stats health check

• check_docker_stats() -> ComponentStatus — main entry point
• Uses httpx.AsyncHTTPTransport(uds=socket_path) to hit Docker API
• GET /containers/json to list containers, then parallel GET /containers/{id}/stats?stream=false
• Parses memory (usage/limit/percent), CPU percent, PIDs, network I/O
• Returns ComponentStatus with per-container sub_components
• Friendly name mapping: sector-7g-worker-homer-1 → Homer
• 10-second cache (stats calls take ~1-2s)
• Returns INFO status (not UNHEALTHY) when socket missing

5. app/services/system/health.py — Wire into cached metrics pipeline

• Add "containers" to SYSTEM_METRICS set (line 37)
• Add check_docker_stats to system_metric_checks dict in _get_cached_system_metrics() (line 390)

6. app/components/frontend/dashboard/modals/containers_tab.py — NEW: Containers tab UI

• ContainersTab(ft.Container) — Flet component
• Summary metric cards: container count, total memory, peak CPU, top consumer
• Expandable table sorted by memory descending: status dot, name, memory MB, limit, mem%, CPU%, PIDs
• Expanded rows show: image, container ID, network RX/TX
• Uses existing controls: ExpandableDataTable, MetricCard, PrimaryText, SecondaryText
• Follows OverviewTab/RoutesTab/LifecycleTab pattern from backend_modal.py

7. app/components/frontend/dashboard/modals/backend_modal.py — Add tab

• Import ContainersTab
• Add ft.Tab(text="Containers", content=ContainersTab(backend_component)) at line 1143

8. app/components/frontend/dashboard/cards/server_card.py — Optional card hint

• Add "Top Container" metric showing the highest memory consumer name + MB

Verification

1. make serve locally — Containers tab shows in Server modal (may show "unavailable" without Docker socket)
2. Deploy to server — tab shows all 11 containers sorted by memory, expandable for details
3. uv run pytest — existing tests still pass
4. Test graceful degradation: without socket mount, tab shows info message instead of error