Files
Project_Velocity/infrastructure/ops_control_plane/README.md
2026-04-12 02:02:58 +05:30

79 lines
2.3 KiB
Markdown

# Desineuron Ops Control Plane
Internal Linux-hosted control surface for:
- AWS machine lifecycle
- S3-backed model ingest with generated manifests and checksums
- model hydration from S3
- runtime and estimated cost tracking
- ingress route management
- session logging and CSV export
Main deployment target:
- Linux box at `192.168.1.4`
Primary public route:
- `ops.desineuron.in`
Canonical S3 bucket:
- `desineuron-ops-control-plane-819079556187-us-east-1`
Related AWS nodes:
- ingress: `i-094df09acafb72494`
- current GPU worker: `i-0e4eab5fe67cf9abe`
Core runtime:
- FastAPI web + API surface
- background worker
- PostgreSQL
- Docker Compose
- systemd wrapper on Linux
Key files:
- `docker-compose.yml`
- `.env.example`
- `app/ops_control_plane/main.py`
- `app/ops_control_plane/worker.py`
- `app/ops_control_plane/cli.py`
- `manage_desineuron_routes.py`
- `install_linux_ops_control_plane.sh`
Runtime paths on Linux:
- stack root: `/opt/desineuron-ops-control-plane`
- env file: `/opt/desineuron-ops-control-plane/.env`
- exports: `/opt/desineuron-ops-control-plane/exports`
- state: `/opt/desineuron-ops-control-plane/state`
Access:
- login route: `https://ops.desineuron.in/login`
- operator logins are provisioned as email-style usernames
- admin password is stored in the protected `.env` file on Linux and should be retrieved locally rather than copied into repo notes
Validated live behaviors:
- market pricing API returns live on-demand and spot views
- session and cost tracking persist in PostgreSQL and export to CSV
- spot launch failures are recorded cleanly instead of crashing the UI
- on-demand GPU launch was validated with a `g6.xlarge` lifecycle test
- managed ingress route upsert/delete was validated through the helper on the `t4g.micro` ingress
- model ingest from Linux model library to S3 was validated with `ops-smoke-model`, including manifest generation and catalog registration
Operator retrieval commands:
- admin password:
- `sudo sed -n 's/^OPS_ADMIN_PASSWORD=//p' /opt/desineuron-ops-control-plane/.env`
- latest CSV export:
- `ls -lah /opt/desineuron-ops-control-plane/exports`
Installer safety note:
- `install_linux_ops_control_plane.sh` intentionally excludes runtime directories (`data/`, `exports/`, `logs/`, `state/`, `.env`) from code sync so redeploys do not corrupt Postgres state or overwrite secrets