Hi Alexander, On Mon, Sep 11, 2023 at 03:44:16PM +0000, Stephan, Alexander wrote: > Hi Willy and Ilya, > > Sorry for the absence, I was mostly out-of-office the last week. I am really > sorry for causing this bug.
No worries, stuff like this happens, that's why we have regtests, CI, this development cycle, and we run all development versions in production on haproxy.org. I could have caught it as well when merging your patches and didn't either. > Thanks so much Willy for fixing it as well as for the pool inspection feature > I saw today, that looks really handy. Yep, actually I figured it could help a lot in some contexts like CI where we don't have the cores, or even some issue reports where we can't reasonably ask the user to send the core nor type complex commands. And I thought that this example was a perfect one since it triggered under vtest, so I rolled back to the commit, applied my patches on top of it and tried again. > I actually debugged the code for quite some time as well, I also tried > analyzers but sadly was not able to find it. That's the problem with such bugs that trigger at exactly one size. You'd have needed a fuzzer, and even then... Too much time needed to fall by pure luck on the problem. CI and prod are much more efficient ;-) > While looking through the CI history, I found > https://github.com/haproxy/haproxy/runs/16184951880, also related to TLVs and > only failing on BSD. > I guess, this is actual flakiness then? I didn't find any related bug fix yet. I think so because locally I'm seeing random tests fail on FreeBSD and I don't know why. We don't monitor this one often and already found it broken in the past. There might be a bug hitting this platform, or some flakiness of vtest there, I don't know for now. We'll figure it before the release anyway :-) Cheers, Willy
