Release Party | Analyzing production using Flamegraphs with Prashant Varanasi
Вставка
- Опубліковано 7 лют 2025
- Prashant Varanasi is a tech lead on the Software Networking team at Uber, working on service discovery and routing infrastructure. He started writing Go while at Google three years ago and it's been his go-to language ever since.
Go 1.9 Release Party Playlist → goo.gle/36ft6Yl
Subscribe to The Go Channel → goo.gle/go-sub...
#GoLang #GoReleaseParty #GoSF
This was the best overview on Go profiling that I've seen. Thanks Prashant!
Amazing talk!!
Thank you!
Is there a link to the slides?
What "CPU Profiler" means is it is blind to I/O. (Bad Thing)
Here's the problem with flame-graphs: a typical speed issue is that routine A calls routine B in 100 different places in the code, and it calls it for a really poor reason. It _should_ call it less often, if at all. So it wastes a lot of time, but you would never see it in a flame-graph because each slice of time is really narrow.
The technique that does work is no tool at all, except a debugger. Just run the app under a debugger, halt it manually at a random time, and look at the call stack. Do this 10 times.
Any slowdown worth fixing will show up in 10 stack samples. And once you _do_ fix it to reduce the time, other problems which were smaller are a bigger fraction, thus easier to see, because the total is reduced but they still take the same time.
good!
Very nice video man like and subscribe meen
The material in this talk is pretty basic and artificially made up for beginner coders. The talk does not teach anything new.