-
Notifications
You must be signed in to change notification settings - Fork 588
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
FATAL check_ticks_consistency() issue #2967
Comments
Did you upload this trace to Pernosco? |
I've uploaded variants of this trace (previous recordings) to Pernosco, and they all seemed to fail. I'll upload this one now. |
Thanks. Is this reproducible if you record with |
If we record with |
I recorded a similar trace of the same program on another machine, and attempted to upload it:
The trace replays on this machine successfully. |
Hrm, ok, I'm not entirely sure why the |
The repo path evaluates to |
Can you post the output of |
The sources from the original trace: https://pastebin.com/xYaX7Whj |
I've been sparring quite a bit trying to get sources to show up. From the first trace: Yet according to the sources:
The sources are symlinked into the nix store. From project root: Perhaps that's the issue? IIRC, sources have to actually be copied into the directory for Pernosco to be able to reach them. |
Hmm. Copying the sources over into the local subdirectory (as opposed to using symlinks) yields the same issue. |
This is unexpected and is probably the problem... |
Line 550 in fd51787
|
Let's debug that. This patch to rr should help:
Then we (by whom I mean you :-) ) just need to work backwards to figure out which path that's supposed to be absolute is actually relative. The intent is that |
@rocallahan I'll try the patch tomorrow. Though it was ridiculously difficult to get sources working with nix, we managed to get a working Pernosco trace of a variant of our bug on our NixOS machine. This is cool because it allowed us to paste the working trace to another developer team (who hasn't heard of pernosco, I don't think) to help them fix our bug of interest: https://gitlab.freedesktop.org/monado/monado/-/issues/130#note_1068888 Pernosco has a viral component to it. I can easily see it becoming a standard practice/common courtesy to other teams in open source software development to always post a pernos.co trace alongside a Git Issue/bug report. @KaneTW and I were discussing this today, but someone from the nix-o-sphere needs to make some helper functions which make adding debug symbols and sources to a project as easy as # ...
package-dev = enableDebugSymbolsAndSources package;
# ... |
Will be interesting to see how that goes.
That is intentional :-) Yeah Nix does seem to be a source of frustration here and it would be great if there was a systematic solution that makes it easy to use Nix with |
So it seems like we're through all the pernosco-submit issues and we're back to the original "my recording doesn't replay" problem now? |
FWIW I discovered the probable source of this issue and fixed it: c168a27 |
Thanks! One other question: does rr support tracing programs that use vulkan? |
I expect that to not work. See #2507 |
I see. This might explain some of the weird behavior we've seen with these traces. The tracee in this instance is |
I would expect vulkan to blow up pretty obviously with some unsupported ioctls. But if the entire thing is done via fds and shared memory perhaps it could die subtly instead of obviously. |
Looks like there's some software rendering implementations of Vulkan out there being worked on (lavapipe & vallium):
Is getting them working with our tracee a step in the right direction towards getting RR traces working here? |
Sure, a CPU implementation of Vulkan should work fine with rr. |
I'm having this issue (also on a 5950x on an Ubuntu Impish system), recorded with and without
I guess that's probably related huh. Yep, the zen_workaround script fixed it. Awesome. ̶I̶ ̶g̶u̶e̶s̶s̶ ̶t̶h̶e̶ ̶d̶o̶w̶n̶s̶i̶d̶e̶ ̶i̶s̶,̶ ̶s̶t̶e̶p̶p̶i̶n̶g̶ ̶t̶h̶r̶o̶u̶g̶h̶ ̶J̶I̶T̶ ̶c̶o̶d̶e̶ ̶i̶s̶ ̶a̶ ̶c̶h̶a̶l̶l̶e̶n̶g̶e̶ ̶a̶s̶ ̶t̶h̶e̶r̶e̶'̶s̶ ̶n̶o̶ ̶d̶i̶s̶a̶s̶s̶e̶m̶b̶l̶y̶ ̶a̶v̶a̶i̶l̶a̶b̶l̶e̶.̶ No, I'm just an idiot who is too used to lldb, and haven't been using I hope you enjoyed reading @caitp's adventures in using rr |
On a Ryzen 5950 running NixOS, attempting to replay a trace:
Possibly relevant: the program I'm debugging attempts to connect to a Valve Index VR headset (though I believe unsuccessfully).
The text was updated successfully, but these errors were encountered: