Closing a service without losing its people
A month after we took simplehoney’s wishlist service offline, people were still writing to ask why their lists wouldn’t load. I thought we had done enough to tell them the service was closing. Their messages made it clear that we hadn’t. As CTO and co-founder, I wish I had handled it better.
We had more than a million products in the database, tens of thousands of searches, and hundreds of thousands of products wished for. People had spent time collecting things there. The server shutdown happened on our schedule; their reasons to return happened on theirs.
These are lessons from a 2013 shutdown. The original infrastructure choices are historical examples. Data handling and retention need a plan specific to the service; the time periods below are planning considerations, not universal rules.
Preserve what can remain useful
For content that was already public, a static copy can preserve access without keeping the entire application running. Private account data needs a separate, authenticated export path. A static copy must not turn a private wishlist into a public page.
In our case, we gradually moved from larger Amazon instances to smaller ones. I also considered static HTML generated with Jekyll. My original hosting suggestions included Joyent micro instances and caching with Varnish; the underlying task is to serve the preserved pages and their assets with as little infrastructure as the site requires.
A practical sequence is:
- Inventory the public pages and decide which should remain accessible.
- Work from a separate copy of the code. Update navigation and presentation so the preserved pages work without the application backend.
- Generate those pages from the inventory. A crawl with
curlorwgetcan capture server-rendered pages, but it does not reproduce application behavior by itself. - Copy the required images and other assets, preserving their paths or updating the references together.
- Check the result as a visitor, including links, missing pages, and any action that used to require a server. Add caching where the traffic and hosting setup call for it.
Advertising or affiliate income might help cover the smaller hosting bill. I would treat that as a possibility to measure, not assume the archive will pay for itself.
Give people a way to take their data
Prepare a usable export for each person. JSON was the simple format I proposed for our data. Package it, store it securely, and make sure the right person can retrieve it during the announced access period. An export sitting on a server is not much help if the person who needs it can’t find or open it.
I originally suggested keeping access available beyond sixty to ninety days, then considering whether the site could be switched off after roughly six months. I would be more careful about treating traffic as the only signal. A seasonal product may need a full annual cycle before its users have a reason to return.
Our app launched around Black Friday. Someone who saved a gift idea could reasonably come back when it was time to shop again. That pattern should have been part of our shutdown plan.
Finish the data cleanup deliberately
The shutdown also needs a decision about what happens to the data after the export period. Inventory the live database, the export files, and the copies kept elsewhere. Define what will be retained, why, who can access it, and when it will be removed. Don’t assume that removing names from a database makes all the remaining information anonymous or suitable for another project.
My original note treated cleanup too simply: export the data, remove identifying information, and keep the rest for analysis. I would want a clearer boundary around what people entrusted to the service before using any of it again.
I had a lot of fun building simplehoney, and I hope people give me another chance to build something they love. The messages arriving after the shutdown are part of that work too. They tell me where the ending we planned didn’t match the ending people experienced.