Comment by bravesoul2

I find most jobs I had fast becomes a big issue once things are too slow. Or expensive.

It's a retroactively fixed thing. Like imagine forgetting to make a UI, shipping just an API to a customer then thinking "oh shit, they need a UI they are not programmers". And only noticing from customer complaints. That is how performance is often treated.

This is probably because performance problems usually require load or unusual traffic patterns, which require sales, which require demos, which dont require performance tuning as there is one user!

If you want to speed your web service up first thing is invest time, maybe money in really good observability. Should be easy for anyone in the team to find a log, see what CPU is at etc. Then set up proxy metrics around speed you care about and talk about them every week and take actions.

Proxy metrics means you likely cant (well probably should not) check the speed that Harold can sum his spreadsheet every minute, but you can check the latency of the major calls involved. If something is slow but metrics look good then profiling might be needed.

Sometimes there is an easy speed up. Sometimes you need a new architecture! But at least you know what's happening.