2 comments

[ 2.5 ms ] story [ 19.2 ms ] thread
Hey hey, Christine here from the Nylas crew. Happy to answer any questions anyone has about this!
Nice simple sampling technique.

I'm curious what this did for average request latency, as that's what really matters for the clients.

Since this effectively measures wall-clock time and the process can spend time waiting for replies from external resources (like local storage, memcache, DBs), which don't consume CPU, did you consider identifying and separating blocking calls into their own graphs?

It'd be nice to confirm/challenge the, "conventional wisdom" that app servers spend most of their time waiting for external resources.