OK, if you have more or less reliable way to reproduce the problem, please file a bug at https://github.com/jvm-profiling-tools/async-profiler/issues
BTW, obj_size + tlab_size does not make sense, since tlab_size already includes all objects allocated in TLAB. Regards, Andrei понедельник, 12 февраля 2018 г., 20:20:54 UTC+3 пользователь Maxim Terletsky написал: > > I doubt that is the case (rewrite libasyncProfiler.dylib while the library > is already loaded into the target process) ... It happens on developer > machines (Mac), not production (Linux) - at least we don't see it in > Production :) > > Are you sure log https://pastebin.com/kJptx2nA is not complete? > I will try to enable it again for developers and ask them to send me more > logs.. > > On Monday, February 12, 2018 at 5:53:25 PM UTC+2, Andrei Pangin wrote: >> >> Hi Maxim, >> >> Does it also fail with original unmodified libasyncProfiler? I guess you >> rewrite libasyncProfiler.dylib while the library is already loaded into the >> target process - can it be the case? >> The crash log seems incomplete, I can't tell for sure what's wrong. >> >> воскресенье, 11 февраля 2018 г., 9:46:33 UTC+3 пользователь Maxim >> Terletsky написал: >>> >>> @Andrei Pangin >>> Hi, >>> Sometimes we saw the following exception thrown >>> https://pastebin.com/kJptx2nA >>> RDI=0x000000015a4e7110: _ZN11AllocTracer13signalHandlerEiP9__siginfoPv+0 >>> in /Users/....../libasyncProfiler.dylib at 0x000000015a4e6000 >>> >>> The code, probably very familiar to you :) , is >>> https://pastebin.com/CPmysehe - the only diff with the original is >>> >>> 1. u64 tlab_size = frame.arg1(); >>> 2. Profiler::_instance.recordSample(ucontext, obj_size + tlab_size, >>> BCI_KLASS, alloc_class); >>> >>> which I added as you suggested. >>> Do you know what could be the issue? >>> >> -- You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group. To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-sympathy+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.