Skip to content

RENDER_DEPLOY.md

Latest commit

 

History

History
222 lines (170 loc) · 5.85 KB

RENDER_DEPLOY.md

File metadata and controls

222 lines (170 loc) · 5.85 KB
Raw

Fear-Free Night Navigator — Render Deployment Guide

Project structure (what you push to GitHub)

fear-free-navigator/
├── data/
│   ├── edges_list.csv
│   ├── adjacency_matrix.npz
│   └── nodes_features.csv
├── outputs/                    ← commit this AFTER running train locally
│   ├── safety_model.pkl
│   └── css_cache.csv
├── src/
│   ├── features.py
│   ├── model.py
│   ├── router.py
│   ├── schemas.py
│   └── main.py                 ← includes AutoPing
├── tests/
│   └── test_router.py
├── requirements.txt
├── run.sh
└── render.yaml                 ← Render reads this automatically

Step 1 — Train locally first

Run training on your machine before touching Render. The model and CSS cache are large binary files — train once, commit, deploy.

# Install deps
pip install -r requirements.txt

# Train (80k sample, ~3-5 min)
python3 src/train_fast.py

# Or train on full 393k dataset (~10-15 min)
python3 src/train_fast.py --full

# Verify outputs exist
ls -lh outputs/
# You should see: safety_model.pkl, css_cache.csv

Step 2 — Push to GitHub

git init
git add .
git commit -m "feat: initial fear-free navigator with trained model"
git remote add origin https://github.com/YOUR_USERNAME/fear-free-navigator.git
git push -u origin main

Important: Make sure outputs/safety_model.pkl and outputs/css_cache.csv are included. Add this to .gitignore to keep the rest clean:

# .gitignore
.venv/
__pycache__/
*.pyc
outputs/roc_curve.png
outputs/feature_importance.png
outputs/ablation_plot.png
outputs/edges_with_coords.csv
outputs/demo.html

If css_cache.csv is over 100 MB, use Git LFS:

git lfs install
git lfs track "outputs/css_cache.csv"
git lfs track "outputs/safety_model.pkl"
git add .gitattributes
git commit -m "chore: track large files with git lfs"

Step 3 — Create Web Service on Render

  1. Go to https://dashboard.render.com → New → Web Service
  2. Connect your GitHub repo
  3. Fill in these settings:
Field Value
Name fear-free-navigator
Region Singapore (closest to Bengaluru data)
Branch main
Runtime Python 3
Build Command pip install -r requirements.txt
Start Command bash run.sh --skip-train
Instance Type Free (or Starter $7/mo for no sleep)
  1. Click Advanced → Add Environment Variable:
Key Value Notes
AUTOPING_ENABLED true Keeps free tier alive
PORT 8000 Render injects this automatically
PYTHON_VERSION 3.12.0 Pin your Python version
  1. Click Create Web Service

Render will:

  • Clone your repo
  • Run pip install -r requirements.txt
  • Run bash run.sh --skip-train (loads model + starts uvicorn)

Step 4 — Verify deployment

Once the deploy log shows [startup] Ready., test it:

# Replace with your actual Render URL
BASE_URL="https://fear-free-navigator.onrender.com"

# Health check
curl $BASE_URL/health

# Ping (AutoPing target)
curl $BASE_URL/ping

# Route request
curl -X POST $BASE_URL/route \
  -H "Content-Type: application/json" \
  -d '{
    "origin":      {"lat": 12.9758, "lon": 77.6011},
    "destination": {"lat": 12.9139, "lon": 77.6419},
    "departure_epoch": 1700000000,
    "profile": {
      "persona": "solo_woman",
      "safety_threshold": 0.65,
      "speed_weight": 0.3
    }
  }'

Your interactive docs are at: https://your-app.onrender.com/docs

How AutoPing works

Render Free Tier
  └─ spins down after 15 min of no requests
  └─ cold start takes ~30-60 sec (very bad UX)

AutoPing solution (inside main.py):
  1. At startup, asyncio.create_task(autoping_loop()) runs in background
  2. Every 13 minutes it sends GET /ping to itself
  3. /ping returns instantly without touching any state
  4. Render sees activity → dyno stays warm → no cold start

Environment variable control:
  AUTOPING_ENABLED=true   → autopings every 13 min  (default, use on free tier)
  AUTOPING_ENABLED=false  → disabled  (use on paid Starter plan — no sleep)

The RENDER_EXTERNAL_URL env variable is automatically set by Render to your service's public URL (e.g. https://fear-free-navigator.onrender.com). AutoPing reads it — no hardcoding needed.

Troubleshooting

Deploy fails at startup — "FileNotFoundError: css_cache.csv" → You forgot to commit outputs/. Run training locally first, then commit.

"pip install" fails on scipy or numpy → Add PYTHON_VERSION=3.12.0 to environment variables in Render dashboard.

Routes take > 5 sec → The free tier has 512 MB RAM. If the CSS cache is very large (>300 MB), upgrade to Starter ($7/mo) or reduce cache size by scoring fewer time bands.

AutoPing logs show failures → Render hasn't set RENDER_EXTERNAL_URL yet (first deploy only). It will resolve on the next deploy once the URL is assigned.

Cold starts still happening → Check that AUTOPING_ENABLED=true is set in your Render env vars. On free tier, if your service URL changes, update the env var accordingly.

Upgrade path

Tier Cost RAM Sleep Recommendation
Free $0 512 MB Yes (15 min) Use AutoPing — works fine
Starter $7/mo 512 MB Never Disable AutoPing, cleaner logs
Standard $25/mo 2 GB Never Use if CSS cache > 500 MB

Re-deploying after retraining

# Retrain with new data
python3 src/train_fast.py --full

# Commit updated outputs
git add outputs/safety_model.pkl outputs/css_cache.csv
git commit -m "model: retrain with updated edge data"
git push

# Render auto-deploys on push (if auto-deploy is on)
# Or click Manual Deploy in the Render dashboard