Teams running affiliate ops, marketplace accounts, or data collection at volume hit the same wall: concurrency without isolation burns accounts. Anti-detect profiles (e.g. Multilogin) plus disciplined ops beat raw headless farms.
Architecture layers
| Layer | Role | Failure mode |
|---|---|---|
| Job queue | Rate-limit tasks, retry with jitter | Thundering herd on one profile |
| Profile pool | 1 account = 1 isolated browser identity | Cookie/fingerprint cross-link |
| Proxy pool | Sticky IP per profile tier | Subnet ban affects whole pool |
| Workers | Playwright via CDP attach | RAM exhaustion, API 429 |
| Observability | Block rate, latency, API errors | Scaling blind into bans |
Scaling rules
- Horizontal first — add worker machines before maxing profiles on one host.
- Profile tiers — warm-up profiles (light browsing) vs production (full flows) vs burn (disposable after flag).
- Cooldown after blocks — pause profile 24–72h; do not immediately swap proxy on same storage.
- API budget — respect Multilogin launch/stop rate limits; batch with semaphores.
- Human-like schedules — stagger starts; avoid 50 profiles hitting login at :00.
Monitoring metrics
- Success rate per profile (7-day rolling)
- Captcha/block events per 100 sessions
- Profile launch P95 latency
- Proxy health (timeout %, ASN changes)
- Canvas hash stability (same profile, two launches)
Related setup
Multilogin automation setup
Queue worker recipe
Profile pool manager
Proxy alignment
Fingerprint checklist
Fix automation blocks
FAQ
How many Multilogin profiles can run at once?
Depends on RAM, API limits, and target site tolerance. Start with 5–10 per worker and measure block rate before doubling.
One proxy per profile or shared pool?
One sticky proxy per production profile is safest. Shared pools only for low-risk tasks with strict session isolation.
Disclosure: MLX-MMO affiliated with Multilogin. SAAS50 / MIN50.