Add support for framehop as unwinder #267
Conversation
|
Welcome @Maaarcocr! It looks like this is your first PR to tikv/pprof-rs 🎉 |
Maaarcocr
force-pushed
the
master
branch
2 times, most recently
from
92c9b0d to
0557997
Compare
|
I've pushed a new commit that separates the perfmap implementation from the framehop unwinder, as they are different concerns. I hope this is better, we can now have perfmaps resolution even with the other unwinder if we want to. This also reduces the changes that were needed to make this work. |
Maaarcocr
force-pushed
the
master
branch
2 times, most recently
from
1cc6810 to
e3e2cc9
Compare
perfmaps are reloaded on demand in a bg thread Signed-off-by: Marco Concetto Rudilosso <marco.rudilosso@shopify.com>
ti-chi-bot
bot
added
dco-signoff: yes
dco-signoff: no
and removed
dco-signoff: no
dco-signoff: yes
labels
Signed-off-by: Marco Concetto Rudilosso <marco.rudilosso@shopify.com>
YangKeao
left a comment
YangKeao
left a comment
There was a problem hiding this comment.
This PR looks good! I'll find several devices to test it and approve before the next week.
| prost-codec = ["prost", "prost-derive", "prost-build", "sha2", "_protobuf"] | ||
| protobuf-codec = ["protobuf", "protobuf-codegen", "_protobuf"] | ||
| framehop-unwinder = ["framehop", "memmap2", "object"] | ||
| perfmaps = ["arc-swap"] |
There was a problem hiding this comment.
Just out of curiosity, could you tell me which JIT compiler you are using pprof-rs on? I'd like to have a try and test ❤️ .
There was a problem hiding this comment.
OH! I see, you are "profiling ruby applications from a native POV". Cool 👍
There was a problem hiding this comment.
yeah, this is for YJIT specifically and yes it's ruby!
| /// Define the MAX supported stack depth. TODO: make this variable mutable. | ||
| #[cfg(feature = "large-depth")] | ||
| pub const MAX_DEPTH: usize = 1024; |
There was a problem hiding this comment.
Nice! It's a good workaround to make compile-time constant configurable in the current rust construction method (though I sincerely hope it'll have better ways in the future) 👍.
Signed-off-by: Yang Keao <yangkeao@chunibyo.icu>
2 participants
This pr adds a new unwinder based on framehop, as it was suggested in #152 and some of the code here is inspired by https://github.com/sticnarf/runwind
While doing so, I made it such that any unwinder also supports resolving symbols from
perfmapsa commonly used format by JIT compilers.I've also made it possible to configure the maximum length of each stacktrace, as for my own usecases (profiling ruby applications from a native POV) 128 was too small.
In order to make framehop works I had to:
this changes have been tested on macos on ARM and linux on x86.