← back to the graveyard
SS-011 Backend-as-a-service · Facebook 2017

Parse — The Backend Facebook Bought, Then Switched Off Under Half a Million Apps

Lifespan
2011–2017 · 6 yrs
Peak Users
~500K+ apps (2014)
Killed By
Facebook (strategic exit)
Status
Shut Down

Summary

Parse was a mobile backend-as-a-service — a hosted set of tools that let an app developer skip building and running servers entirely — and on January 28, 2017 Facebook switched it off. Founded in 2011 by Tikhon Bernstam, Ilya Sukhar, James Yu, and Kevin Lacker, Parse sold a deceptively simple promise: write your iOS or Android app, point it at Parse for data storage, user logins, and push notifications, and never think about a database, a server, or a sysadmin again. For a generation of mobile startups and solo developers, that promise was irresistible, and Parse grew fast.

In April 2013 Facebook acquired the company in a deal widely reported at roughly $85 million, its first serious move into paid developer tools. Under Facebook's banner Parse kept growing; by 2014 it was reported to power around 500,000 mobile apps, and at shutdown TechCrunch noted that at one point some 600,000 apps relied on the platform. For nearly three years it looked like a rare example of a developer service that an acquisition had strengthened rather than smothered.

Then, on January 28, 2016, Facebook announced that Parse would be wound down over exactly one year and shut for good on January 28, 2017. There was no scandal, no security breach, no collapse in usage cited — only a corporate decision to focus elsewhere. Facebook gave developers a year, a database-migration guide, and, unusually, the platform's own source code, open-sourced the same day as Parse Server so that anyone could stand up their own copy. It was about as graceful as a shutdown gets, and it still stranded hundreds of thousands of apps whose owners had to scramble to migrate or go dark.

That is the lasting weight of the Parse story. Tens of thousands of developers had taken Facebook at its word and built their products on infrastructure they did not control. When the owner's priorities changed, the polite one-year window did nothing to change the basic fact: every one of those teams now had to rebuild the foundation of a shipping product, on a deadline they did not set, for a reason that had nothing to do with them.

Timeline

2011
Founding
Tikhon Bernstam, Ilya Sukhar, James Yu, and Kevin Lacker — alumni of Google and Y Combinator — start Parse to give mobile developers a hosted backend.
2012
Early traction
Parse reports its tools in use by roughly 20,000 mobile developers, with monthly growth cited around 40 percent.
April 25, 2013
Facebook acquires Parse
Facebook buys the company in a deal widely reported at about $85 million — its first paid business-to-business service for developers.
2013–2014
Growth under Facebook
Parse adds analytics, hosting, and cloud code; by 2014 it is reported to power around 500,000 apps.
2015
Quiet plateau
Inside Facebook the platform sits outside the company's core advertising and social mission, a developer service with no obvious tie to the ad business.
January 28, 2016
The wind-down notice
Facebook announces Parse will shut down in exactly one year, on January 28, 2017, citing a desire to focus its efforts.
January 28, 2016
Parse Server open-sourced
On the same day, Facebook releases Parse Server, an open-source version of the API, so developers can self-host instead of leaving the ecosystem entirely.
2016
The migration year
Facebook publishes a database-migration guide; developers move data to MongoDB, stand up Parse Server, or rewrite onto rivals like Firebase and AWS.
January 28, 2017
Lights out
The hosted Parse service is retired; apps still pointed at it stop working.
2017 onward
The open-source afterlife
Parse Platform continues as a community-run open-source project, with its own documentation, blog, and forum, long outliving the hosted business.

The Backend in a Box

Parse arrived at the exact moment mobile development became a gold rush and most of the prospectors had no interest in operations. Building an app's front end — the screens a user taps — was hard enough; behind it sat a second, invisible product nobody wanted to build twice: a database, an authentication system, push-notification plumbing, and servers to run it all, kept patched and scaled at three in the morning. Parse offered to be that whole back half as a service. A developer dropped in a software kit, called a few methods, and got cloud data storage, user accounts, and push notifications without provisioning a single machine.

For startups racing to ship and for the vast population of independent developers building one app at a time, this was transformative. It compressed weeks of backend engineering into an afternoon and removed the entire category of server operations from the to-do list. Parse was, in the cleanest sense, infrastructure — the load-bearing layer beneath an app, the part the end user never sees but without which nothing works. By the time Facebook came calling in 2013, Parse had tens of thousands of developers and a reputation as the friendliest way to put an app online.

The acquisition seemed to validate everything. Facebook, flush from its IPO and hungry to matter to mobile developers, paid a reported ~$85 million and promised investment, not absorption. For a while it delivered, layering on hosting and analytics while the app count climbed toward and past half a million. The danger in that arrangement was invisible precisely because it was working: every one of those apps had quietly made a multinational with its own agenda the landlord of its foundation.

The Quiet Logic of a Cull

Nothing about Parse failed. There was no breach, no exodus, no embarrassing headline — the platform was running hundreds of thousands of apps when Facebook decided to end it. What changed was Facebook, and the calculus a developer-tools business looks like from inside a company whose fortune is advertising. Parse generated revenue from developers, not from ads or social engagement, which made it orthogonal to the mission and, in a period of sharpening focus, expendable. A profitable-enough side business that does not feed the core is exactly the kind of thing a large company prunes.

So Facebook pruned it, and did so about as decently as the format allows. Where Google Reader's users got roughly three months and an OPML file, Parse's developers got a full year, a detailed migration guide, and — the genuinely unusual part — the platform's own code, released as the open-source Parse Server on the same day as the shutdown notice. The message was coherent: we are leaving this business, here is the door, and here is a kit to rebuild what you had somewhere we no longer have to run it.

Decency, however, does not refund the work. A one-year window is generous against the clock and brutal against the calendar of a team that has a live product, paying users, and a roadmap of its own. Migrating a backend is not a weekend chore; it means re-pointing every app, moving every record, re-testing every login and push notification, and shipping updates to users who must install them or break. The grace was real, and so was the cost, and the cost fell entirely on the developers who had trusted the platform — not on the owner who changed its mind.

January 28, 2017

The shutdown landed on the anniversary of its own announcement, a year to the day, which gave the whole affair an unusually orderly feel. By then most of the serious operations had moved. Some teams stood up Parse Server on their own infrastructure, swapping a managed convenience for the server work Parse had been invented to abolish — which was, in its way, the point and the irony at once. Others used the moment to migrate onto rivals: Google's Firebase, which inherited a great deal of Parse's mindshare, or Amazon's services, or any of the backends that sprang up to catch the refugees.

The apps that did not move simply stopped. A backend service fails differently from a desktop program: when the servers go dark, the app on the phone keeps launching but can no longer log a user in, fetch data, or send a notification — a working shell with nothing behind it. For abandoned side projects, dormant startups, and developers who had moved on without noticing the deadline, January 28 was the day their apps quietly became inert. There is no count of how many, which is its own small epitaph.

What survived was the code. Parse Server, open-sourced in 2016, outlived the company that paid $85 million for the original and the company that switched it off, carried forward by a community that runs it to this day. The hosted business is gone; the platform, oddly, is not. It is a tidy demonstration that open-sourcing a product on the way out is the most useful apology a platform owner can offer — and also that it is no substitute for not needing to apologize.

The Five Factors

01
You don't control the platform you build on
Every app on Parse rented its foundation from a company whose priorities it could not influence. When the landlord's strategy changed, the tenants' carefully shipped products became the landlord's housekeeping. Building on someone else's backend means accepting that the floor can be sold out from under you on their schedule.
02
Orthogonal businesses get pruned
Parse made money from developers, not from advertising or social engagement, which made it irrelevant to Facebook's core mission no matter how well it ran. A profitable side business that does not feed the parent's central revenue line is a perennial candidate for the cull, performance notwithstanding.
03
A graceful shutdown still relocates the cost, it does not erase it
A one-year window, a migration guide, and open-sourced code are about as good as exits get — and they still left hundreds of thousands of teams to rebuild a live product on a deadline they did not choose. Grace softens the blow; it does not pay for the work.
04
Infrastructure is exactly what you must not rent casually
The deeper a service sits in your stack, the costlier it is to replace, because everything above it depends on it. Parse sat at the very bottom — the database and auth layer — which is why its removal forced a foundation rebuild rather than a swap of a feature.
05
Open-sourcing is the most honest off-ramp a platform can give
By releasing Parse Server, Facebook let the platform outlive the business and gave stranded developers a way to keep their architecture intact on their own machines. It is the difference between abandonment and migration — and the rare shutdown move that leaves something running.

Aftermath

The migration largely worked, in the narrow sense that the year and the tools were enough for teams that were paying attention. Many moved to Firebase, which absorbed much of Parse's role as the default friendly backend; others self-hosted Parse Server, trading managed ease for control; others rebuilt on AWS or wound their apps down on purpose. The apps that vanished were mostly the ones already neglected — but every active product that survived did so by spending engineering time it would rather have spent on its actual business.

The lasting mark is twofold. Parse Server endures as a living open-source project, proof that a shuttered platform can leave usable bones behind. And Parse joined Google Reader in the small canon of shutdowns that taught the industry a lesson it keeps having to relearn: a platform you do not own is a dependency, not a foundation, and the friendliness of the exit does not change the arithmetic of the trust. Among developers, "what if it goes the way of Parse?" became a real question asked of every managed backend since — which is perhaps the most useful thing a dead platform can become.

Lessons

  1. Treat any platform you build on as a dependency you may have to replace, not a foundation you own; the deeper it sits in your stack, the more that warning matters.
  2. A free or hosted service from a giant is the giant's to discontinue; read its relationship to the owner's core business, because the orthogonal extras get pruned first.
  3. Value a clean exit — a long window, a migration path, open-sourced code — but never mistake it for safety; grace relocates the cost of a shutdown onto you, it does not remove it.
  4. Keep your data portable and your architecture loosely coupled, so that re-pointing your backend is a migration and not a rewrite.
  5. If you must depend on someone else's infrastructure, prefer the kind you can self-host or fork; the ability to take the code with you is the only real insurance against a strategic exit.

References