So the Tailscale team were using a single text file as a database (as you do) and it worked great… until it didn’t.

Even with fast NVMe drives and splitting the database into two halves (important data vs. ephemeral data that we could lose on a tmpfs), things got slower and slower. We knew the day would come. The file reached a peak size of 150MB and we were writing it as quickly as the disk I/O would let us. Ain’t that just peachy?

So, migrate to MySQL or PostgreSQL, right? Maybe SQLite?

Nope, Crawshaw had other ideas.

I won’t ruin the surprise and tell you what they went with, but I will say it’s a widely deployed system amongst cloud natives…

But why?

Little bored. Not learning as much as I used to. I’ve been doing the same thing too long and need a change. It’d be nice to primarily work in Go rather than work on Go.

When I first joined Google it was a chaotic first couple years while I learned Google’s internal codebase, build system, a bunch of new languages, Borg, Bigtable, etc. Then I joined Android it was fun/learning chaos again. Go was the same when I joined and it was a new, fast-moving experiment. Now Go is very popular, stable…

Brad Fitzpatrick is going back to the drawing board on Go’s net/http Client and Transport. The linked repo is in its “problem collection” phase, but it’s worth following (or getting involved) now before the actual solution begins to take shape. What’s wrong with the status quo? The list of problems is extensive (and growing).