Feedbin
RSS on Heroku, then metal, then your own rack
Ben Ubois launched Feedbin two days before Google killed Reader — profitable in weeks, crushed by Heroku Postgres pricing, then rebuilt on SoftLayer metal, Sidekiq Pro, and eventually colocation at Hurricane Electric.
2013 launch
Google Reader dies, Heroku scales until it cannot
Ruby on Rails
Heroku
PostgreSQL
Sidekiq
Feedbin went live into a market Google had abandoned. Traffic spiked immediately; Ubois turned on more dynos and bigger Postgres plans until he hit Heroku’s ~$6,400/month database tier with nowhere left to go.
His first-year retrospective is blunt: Sidekiq made background refresh tractable; the web tier on Heroku did not.
Lesson
Managed platforms buy time, not infinite headroom. Know the top rung of the pricing ladder before you need it.
June 2013
21 servers, SoftLayer + DigitalOcean hybrid
DigitalOcean workers
SoftLayer bare metal
nginx load balancers
Redis
Sidekiq Pro
Route 53 failover
The server-move post describes 21 machines: SoftLayer for the core app and DB, DigitalOcean for parallel feed refresh workers chewing Sidekiq Pro queues, nginx pair for SSL termination, Redis for dedupe checks.
Ubois took a weekday outage because graphs said Feedbin was about to fall over — dedicated I/O beat throwing money at shared Postgres.
Lesson
Hybrid metal + cheap cloud workers is a valid pattern. Put parallel HTTP where VPSes shine; keep the DB on hardware you trust.
2019–today
Colocation at Hurricane Electric
Rails app
Colocated rack (HE Fremont)
Dedicated hardware
Self-managed networking
In September 2019 Ubois moved Feedbin into his own rack at Hurricane Electric Fremont 2 — months of prep for a half-hour cutover. More posts promised on the hardware setup.
The arc is cloud → dedicated → colo, each step driven by bill shock and I/O predictability, not resume-driven microservices.
Lesson
Leaving the cloud is not Luddism when your unit economics are RSS subscriptions, not VC rounds.
Sources
- Ben Ubois — Feedbin’s first year (Heroku → SoftLayer context)
- Ben Ubois — Feedbin server move (21-server architecture)
- Ben Ubois — Moving to colocation (2019)
Facts drawn from public engineering posts and interviews. Numbers are approximate where sources disagree — we're stack advisors, not historians.
Get the right stack for YOUR scale — not theirs
Feedbin's stack made sense for Feedbin. Yours won't — and copying theirs is how you end up with a $4k/month bill and three unused Kubernetes clusters.
Plan my stack — free