Marketplaces & Logistics
Consumer SoftwareScaling a consumer VPN to millions of sessions a day
Session volume across regions was outgrowing the control plane. We re-engineered the authentication, session and billing services for horizontal scale and regional failover.
- 99.9%
- Uptime across regions
- 2×
- Throughput per server
- 50%
- Less on-call load
- 12 min
- To bring a new region online
Overview
The challenge and our approach
01The challenge
Every marketing push brought a wave of sign-ups and, a few hours later, a wave of connection failures. Servers were provisioned by hand, capacity was guessed a month ahead, and a regional outage was usually reported by customers before the monitoring noticed.
The account platform had grown the same way: billing, device limits and app configuration lived in one PHP application that two people were afraid to deploy.
02Our approach
We moved the fleet to declarative provisioning, so a server is described once and rebuilt identically anywhere. Capacity now follows real session load per region, adding nodes before users feel congestion and draining them when the peak passes.
Monitoring was rewritten around what a customer experiences — connection success, handshake time, throughput — from probes in every region. The account platform was split into billing and device services with a deploy pipeline anyone on the team can run.
- Client
- FastestVPN
- Region
- Global, 40+ locations
- Engagement
- Ongoing since 2023
- Services
- Cloud & DevOps, Product engineering
The build
What we built
Infrastructure first, then the account platform the apps depend on.
Declarative fleet
Every VPN node defined in code and rebuilt identically in any region in minutes.
Load-aware scaling
Capacity follows live sessions per region, ahead of congestion rather than after it.
Customer-side probes
Connection success and throughput measured from every region, paging on what users feel.
Smart server selection
Apps are steered to the fastest healthy node, not simply the nearest one.
Billing service
Subscriptions, renewals and refunds separated from the apps, with safe zero-downtime releases.
Hardened access
Short-lived credentials, audited admin actions and no shared keys on the fleet.
Results
Sign-up spikes stopped being incidents
Measured over twelve months across every region in the network.
99.9%
Regional uptime
Held through three of the largest campaigns in the company's history.
2×
Throughput per server
Tuned kernels and protocol settings doubled what each node carries.
50%
Less on-call load
Alerts fire on customer impact, so pages are fewer and all of them matter.
12 min
New region online
From decision to serving traffic, instead of a week of manual setup.
Tech stack
What it runs on
Apps & account
Flutter
PHP
MySQL
Redis
Fleet
Docker
Kubernetes
AWS
Delivery
GitHub
Git
Google Analytics
Throughput doubled and the on-call load halved. They understood the infrastructure better than the vendor we had been paying for it.
More case studies
Other work we are proud of
- Marketplaces & Logistics
Construction contractor · Oplyra ERP deployment · Construction & TradingConstruction contractorOne ledger for a contractor running six disconnected systems
Result:40%Less time closing the month
Facing a problem like this one?
Thirty minutes with a solution architect — walked through your own system, with an honest answer on whether we are the right team for it.