#008

On GOV.UK we handle draft content by having separate databases for draft and live content, and running two sets of frontend apps: one for draft and one for live. The draft apps are protected behind signon, our single sign-on service. Historically, our content item API, which I mentioned last week, hasn’t worked for draft content; I fixed that. It turned out to be a surprisingly complex problem, involving puppet, nginx, firewall rules, and DNS.

I fixed a discrepancy in our local-links-manager app, which is what handles linking to local government services from GOV.UK. I’m not really sure if I actually fixed the problem, or just fixed the symptoms, but it’s good enough for now.

I did a bit of refactoring of our Jenkins configuration, to help avoid implicit dependencies between Jenkins jobs and how our machines are currently set up.

I finally deleted our logging VMs, which I started a month ago. This isn’t the first time I’ve retired something—if I keep this up, I should have switched off all of GOV.UK in about 10 years.

I took Thursday and Friday off. I had planned to go to Code Mesh, but in the 5 months since I submitted my request, HR didn’t manage to buy the tickets. Not going to the conference was disappointing, but having some time off was still nice.

I did some reading:

• I read The Checklist Manifesto, which was really good. It’s about how checklists became standard in surgery, which sounds like the driest topic imaginable, but it’s full of interesting anecdotes from aviation, construction, finance, and the odd surgeon who realised the benefits before they became widespread. It’s got me wondering what things I should have checklists for.

• I finished Children of Dune. Unfortunately the rest of my Dune books are buried in a loft in Hull, so I can’t move on to my favourite Dune book: God Emperor of Dune.

I’ve now moved on to Lolita, which I last read in December 2012.

I started writing up my session notes from Call of Cthulhu.

A replacement motherboard for nyarlathotep, my broken server, arrived on Monday; but was itself busted. I’ve sent it back, and I guess it will be next week before I get things working again.

I think my flat is cursed, as azathoth, my desktop computer, also died. So I went from having a working desktop computer and a working server, to both being dead in 11 days.

Fortunately, it was the CPU that had died (somehow), so I was able to use the CPU that I’d bought for nyarlathotep—the same model as the one in my desktop. So now I have a working desktop again, but I’ll need to buy another processor for nyarlathotep. Fortunately 2012-era processors aren’t that expensive. Certainly much cheaper than building a new desktop computer with modern hardware.

I’ve started looking into Elixir, and read through the guides. It looks pretty nice, and I’m tempted to write a little HTTP router for it. Such a router would inspect the request path and do one of:

• Act as a reverse proxy to another server, if the request method is allowed.
• Perform an “internal redirect”, change the request path and re-route, without sending a 3xx to the client.
• Perform an “external redirect”, send a 3xx to the client.
• Return a 410.

A path without a route would return a 404. The 404, 405, 410, and 500 responses would use a static file as the response body.

Ideally, the collection of routes would live in a database or a file, and a signal would trigger the router to re-load them into memory, probably as a trie, and atomically switch to using the new routes. This is similar to the GOV.UK router, but written in Elixir rather than Go.