Julia Evans

Profiler week 4: callgrind support, containers, Mac progress!

Today is Monday and week 4 of working on my profiler is over! I’m 13 of the way through. Eep! My main goal for last week was to release Mac support, which hasn’t quite happened yet – it turns out that Mac systems programming is more complicated than I thought (getting a mac’s memory maps is really hard! and I got derailed a bit by a kernel bug).

last week (contributions!!)

exciting things that happened last week:

  • 2 new people contributed code to rbspy! (@liaden and @vasi!). liaden contributed a --duration flag and a --rate flag, so you can change the rate rbspy is sampling at. vasi contributed a new output format for rbspy (cachegrind format!) which I think will be useful – it lets you see. The cachegrind PR has some nice pictures.
  • also lots of folks made issues and suggestions and tried it out, which is awesome. All the issues are so helpful!
  • Added support for profiling processes running in containers! It seems to work well!
  • Learned A LOT more about Mac systems programming than I used to know. I think I should be able to merge a Mac support PR in the next day or two.

So far it seeems like Rust is easy enough that some non-Rust-programmers can come in and start contributing PRs to rbspy, which is really nice to see!

next week

On the schedule for this week:

  • finish up Mac support
  • fix a bug with getting the stacks when the top function in the stack is a C function that I can reproduce
  • possibly put together a website?
  • @liaden is working on support for profiling subprocesses (so you can point rbspy at your Unicorn process or something, and it’ll profile all of your web workers). I think that’ll be awesome.

a Rust flamegraph library?

Something that’s been on my mind but I haven’t really figured out is – rbspy is growing a few new visualization formats (flamegraphs! callgrind format!). I think it could be cool to build a Rust crate with support for different visualization formats, so that if people build other profilers then they can kinda have access to a consistent library of visualization formats.

I don’t really know if I have time for that though! For now I’m going to stay focused on lower level concerns.

MEMORY PROFILERS!!!

I’m not done with CPU profiling (there are still lots of rough edges to sort out!) but once I get things slightly more feature complete I think I kinda want to let the CPU profiler rest for a bit and work on prototyping something else while I give people a chance to try out the project.

Today I had 2 extremely helpful conversations about memory profilers and I feel excited about trying something out!

My thought is to maybe add a memprofile subcommand to rbspy, where you give it a PID and it does… something? I have 2 ideas for what it could do right now:

  • give you a memory profile of your program (a summary of objects / their types / their sizes)
  • start tracking new allocations and tell you where they’re happening and how much memory is being allocated right now.

I’m hoping to have time to do some work prototyping a memory profiler this week. We’ll see what happens in real life!