Component
Spec
Capacity
App instances
2x apps-d-1vcpu-2gb
4 gunicorn workers each, 8 total
PostgreSQL
db-s-1vcpu-1gb
~100 concurrent connections
Redis/Valkey
db-s-1vcpu-1gb
~10,000 ops/sec
Monitoring
s-2vcpu-4gb droplet
Prometheus + Grafana + Loki
From k6 load testing against production:
Tier
Concurrent Users
Requests/sec
P95 Latency
Error Rate
Bronze
50
~100 req/s
< 500ms
< 1%
Silver
200
~270 req/s
< 3s
< 5%
Gold
500
~330 req/s
< 5s
< 5%
Database queries — the primary bottleneck. Mitigated by Redis caching (30-60s TTL on read-heavy endpoints)Gunicorn workers — 4 workers per instance can handle ~40 concurrent requests before queuing. Monitored via http_requests_in_flight metricSingle-threaded GIL — Python's GIL limits CPU-bound work per worker. CPU spikes from chaos testing confirm this
Upgrade to apps-d-2vcpu-4gb instances — doubles worker count to 8 per instance
Upgrade Postgres to db-s-2vcpu-4gb — more connections, larger shared_buffers
Horizontal (for sustained growth)
Increase instance_count in .do/app.yaml — App Platform handles load balancing
Current: 2 instances. Can scale to 10+ without config changes
Each additional instance adds ~165 req/s capacity
Instance Count
Est. Throughput
Est. Max Users
2 (current)
~330 req/s
~500
4
~660 req/s
~1,000
8
~1,300 req/s
~2,000
Component
Monthly Cost
App Platform (2x apps-d-1vcpu-2gb)
$24
Managed Postgres (db-s-1vcpu-1gb)
$15
Managed Redis (db-s-1vcpu-1gb)
$15
Monitoring Droplet (s-2vcpu-4gb)
$24
Staging App Platform (basic-xxs)
$5
Total ~$83/mo
Scaling to 4 instances adds ~$12/mo. Database upgrades add ~$30/mo each.
These Grafana alerts fire before capacity is exhausted:
High CPU > 80% — indicates need for more instances or larger instancesHigh Memory > 85% — indicates memory pressure, potential OOMIn-flight requests > 50 — indicates request queuing, add instancesP95 latency > 2s — indicates saturation, scale horizontally