22:17 <agentzh> fche: i'll talk about tracing nginx/openresty/perl/python/redis/etc on this year's nginx conf: https://www.nginx.com/nginxconf/2018/agenda/#day-1-yichun-zhang

22:17 <agentzh> systemtap will be the center :)

22:18 <agentzh> *will be at the center

22:20 <fche> sounds good

22:21 <fche> haven't had any new thoughts about the @vma() one ... do you have a sense of whether there is any conceivable safety (crashability) hazard with it?

22:22 <agentzh> not any that i'm aware of. it's just converting integer values itself, not accessing any memory.

22:22 <agentzh> *any target memory

22:22 <fche> no 0-dereference type of possibilities?

22:22 <agentzh> nope

22:22 <agentzh> it's just integer additions (in case of PIC/PIE).

22:22 <agentzh> or no-op for non-PIC/PIE cases.

22:25 <agentzh> fche: out of curiosity, are you guys using virtualization solutions like qemu to debug weird kernel-space issues?

22:25 <fche> all the time

22:25 <agentzh> nice.

22:25 <agentzh> with kvm?

22:25 <fche> but haven't had any crashes in quite some time, so am running tests on my main workstation much fo the time

22:25 <fche> yeah kvm is a favorite

22:26 <agentzh> i'm still thinking about that CPU lockup issue while running the stap test suite with high job count.

22:26 <agentzh> it worries me.

22:26 <fche> yeah

22:27 <agentzh> do you also try vbox?

22:27 <agentzh> virtualbox i mean

22:27 <fche> nope

22:27 <agentzh> too big? ;)

22:27 <fche> hm that rings a bell, with some virtualization engines, kernel kprobes don't work as reliably as with others

22:27 <agentzh> or too oracle?

22:27 <fche> or at least didn't in the past

22:28 <agentzh> oh, so it might be something specific to vmware?

22:28 <fche> possible

22:28 <agentzh> i was using vmware.

22:29 <agentzh> if it's indeed the case, then it would be quite relieving.

22:29 <agentzh> few people run vmware for production anyway.

22:36 <agentzh> fche: do you have any detailed info on the kprobes instabilities on certain virt engines?

22:37 <agentzh> that would be very helpful for me.

22:37 <fche> sorry, nothing recent

22:37 <agentzh> okay, np.

22:37 <fche> but ISTR kprobes on large-smp boxes were implicated

22:37 <fche> so probably some sort of icache flushing matter. .... dunno sorry

22:37 <agentzh> i'll make note anyway. thanks.

22:39 <agentzh> maybe i should start by attempts of reproducing the issue on bare metals.

22:40 <fche> that's be great

22:40 <fche> must be off, see ya