The proportion of read/write may skew towards reads, but Wikipedia still is an application where any user can create state visible to all other users. It's not as simple as this comment makes it out to be.
But how quickly must those writes be reflected in the reads of others? If you can accept a few minutes of latency there, I imagine things would get easier
In order for wikipedia's anyone can edit to work, its really important that when someone makes a bad edit to a popular article that it can be removed immediately. This is important both to get things fixed quickly and to make it less of a juicy target so less people vandalize (no fun to vandalize if it doesnt stay up).
I suspect latency in the minutes for cache updates would be unaceptable to wikipedia users
Power users use very different workflows than read-only users. You can serve pages from a 30-minute-old cache to the 99.9% of passive readers and it doesn't hurt that much. Editors use "Recent Changes" to monitor edits, and that's much easier to render in real time because the audience is comparatively minuscule.
Yes but if someone replaces the picture on the trump article with goatse, and non power users get this version for 30 minutes until the cache clears - they are going to be pretty pissed and start yelling to power users & just generally cause a PR disaster.
Additionally if vandals know their vandalism will stay for 30 min, they are much more likely to do it, which is a vicious cycle
Aren't articles like Trump write protected? If you want to edit, you shouldn't be able to, unless you have an account that's not brand new. You will be banned very quickly as soon as you start putting goatse on most visited pages.
Also, the white house PR team is actively watching and editing political figures articles. They will sort it out too.
First off, keep in mind that even scaling a broadcast publication can be complex. Sure one can bolt on fastly or s3 but cache invalidation is never a simple problem.
Next "power users" as others put it are not a single set of editors. It's more of a social network with multiple levels of trust. The idea of a wiki is that all users have write access, even if those changes are moderated to have different levels of latency.
Of course there are ways to engineer the system, but at that point one is, well, engineering a system. And WMF is doing so on a shoestring compared to other comparable levels of traffic.
Is WMF creating new paradigms of computing? Probably not. But they are doing a good job, IMHO.
It must be immediate for logged-in users, but not others. It’s an excellent thing that Wikipedia doesn’t nudge people to log in all the time, and I susped 95% of users are not logged in.