Synapse was generally considered to be inefficient and a huge memory hog, so they have been experimenting with various ways to rewrite it in Go, saying[1] at the end of last year:
"Synapse is in a relatively stable state currently ... we’re starting to hit some fundamental limitations of the architecture: ... python’s single-threadedness and memory inefficiency; ...; the fact the app papers over SQL problems by caching everything in RAM (resulting in synapse’s high RAM requirements); ..."
But in January[2] it was discovered, during the investigation of another problem, that the vast majority of the memory usage was due to a bug:
"It’s also revealed the root cause of why Synapse’s RAM usage is quite so bad – it turns out that it actually idles at around 200MB with default caching, but there’s a particular codepath which causes it to spike temporarily by 1GB or so – and that RAM is then not released back to the OS."
I wonder if they would still have started the Go rewrite if it hadn't been for that bug. (It seems the new implementation has a better architechture and is substantially faster, so I guess it's still a good idea though, but maybe not as urgent as before?)
[1] https://matrix.org/blog/2016/12/26/the-matrix-holiday-specia...
[2] https://matrix.org/blog/2017/01/06/synapse-0-18-6-is-out-ple...