#155
Sep 6, 2021
This was written a day late as I was busy over the weekend.
Work
This was a quiet week. Bank holiday on Monday, and the other backend developer on the team has been off, so it’s just been me and whoever I can get PR reviews from.
I’ve continued working on the integration with the new Digital Identity auth service, and also on our approach to caching in the presence of personalisation (I am really growing to dislike that word), which can be summed up as “personalisation is progressive enhancement”.
Auth integration
Possibly the most impactful thing I did was to organise my integration to-do list and to make a second list of things I want done by the end of the sprint (as the full list is not possible).
So rather than one big entry like “finish account management & security” I now have “TODO (for us): use correct account management & security URLs in each environment” and “TODO (for DI): finish account management & security frontends, and pick final URLs”. We are not the DI team, so the more things I can note down as “(for DI)”, the better.
There are a few “TODO (both)” entries, for larger features which need some co-design before we can work out exactly what each team has to implement.
I then worked through a bunch of the “TODO (for us)” entries, and we’re on track to meet my sprint goals, which is nice.
Caching personalised pages
I think there are three ways a page might depend on the user’s session (or lack thereof), and by thinking about what ways a particular page should use, we can avoid losing caching entirely.
There’s core functionality: things like authenticating, credential management, resuming things you’ve saved, and so on. This is best rendered entirely server-side (or in the CDN), as it has to work for all users, and plain HTML and CSS with no JavaScript is good for that. This stuff is typically uncacheable.
There’s navigation: things like showing the appropriate “sign in” or “sign out” link in the header. The defining feature here is that this depends on whether the user has a session or not, but not on who the user is. This is part of why I dislike the term “personalisation”, as it’s not personalised.
There’s everything else: which is a fuzzy category for things which depend on who the user is, but which aren’t core functionality, and so could potentially be implemented in more cache-friendly ways, like by JavaScript. This is what we can treat as a progressive enhancement.
This is a helpful way to think about the problem, because it transforms a hard technical problem (“how do we rearchitect GOV.UK for a world where almost nothing is cacheable?”) to a few less hard technical and design problems. Those problems are:
“How do we rearchitect GOV.UK for a world where core account functionality is uncacheable?” — well, we probably don’t need to. Most things aren’t core functionality, so this stuff has comparatively little load. Just slightly scaling up should do it (if that’s even necessary).
“How do we handle caching of pages which have account navigation elements?” — we can cache two copies of the page, one for logged in users and one for logged out users, and use some CDN magic to ensure that we’re not having a cache miss for each individual user.
“How do we design for the case where JavaScript is unavailable or slow?” — if everything non-core and non-navigational is implemented client-side, we need good default states and to avoid (or minimise) a “flash of unpersonalised content”.
Each way a page might depend on the user’s session has a different question. But that question is scoped to a more specific problem, and we still keep caching for most of GOV.UK. This is progress!
Books
This week I didn’t finish any books, but I read some of volume 1 of Magistellus Bad Trip, which a friend is the English translator for. The text is fine. It’s light novel-y, but that’s to be expected. But the story is garbage. I gave it a shot, but the suspension of disbelief required was just… too much.
The premise is that there’s an online VR stock trading game, set in a virtual city-state, but also with guns, AI paramilitary troops and, generally, all the stuff you need to make a game about stock trading fun. You can drive fast cars, shoot people to get their stuff, eat virtual food (which, thanks to the VR tech, is actually tasty but physically unfulfilling), spend it in virtual strip clubs, and so on. That’s cool. I like that setting.
Where it goes bad, though, is that it explains that this VR game has become super influential in reality too. If you ruin an individual in the game, well, if they’ve invested enough of their own money (there’s a one-to-one conversion ratio with the Yen), then you ruin them in reality too because now their money is gone! Taking it up a notch, there are entire countries whose economies depend on this game.
That could be an interesting premise, if the VR game weren’t so gamey! The plot of the first volume is about a “cheat level” shotgun, which is so powerful you could become the richest person in the game, ruining everyone else, and therefore causing a global financial collapse. And, while the stakes are high, and the characters say that the stakes are high, they still all act like this is just a game.
It’s just too much. I did steal the name “Magistellus” for my Traveller character though.
Gaming
Due to a rearrangement, on Tuesday I played a fourth RPG session in four days: Stars Without Number on Saturday, D&D on Sunday, Call of Cthulhu on Monday, and Traveller on Tuesday. That was fun, but it’s nice having some of my evenings to myself.
I’m definitely enjoying having a weekly Traveller game. Even though we’re not getting more done in a session than in my other games, it just feels faster-paced not needing to wait a fortnight between sessions. Though, I don’t think I’d want to play a single game for six months before changing to something else. Having other groups playing different games is good too.
Miscellaneous
On the 1st, my home server monitoring alerted me of a potential disk failure. That was worrisome, but thankfully it turned out to be not so bad. I’m still not entirely sure what the problem was, but it went away after cleaning dust out of the case and ensuring the SATA cables were firmly connected.
I wrote up an incident report covering my debugging steps.