I’ve seen this happen a couple of times recently.

https://build.chromium.org/p/client.crashpad/builders/crashpad_try_win_x86_wow64_rel/builds/308/steps/run%20tests/logs/stdio

[ RUN      ] CrashpadHandler.ExtensibilityCalloutsWork
e:\b\build\slave\crashpad_try_win_x86_wow64_rel\build\crashpad\handler\crashpad_handler_test.cc(88): error: Expected equality of these values:
  reports.size()
    Which is: 0
  1u
    Which is: 1
[24792:25484:20171010,195857.246:ERROR process_info.cc:350] VirtualQueryEx: Access is denied. (5)
[24792:25484:20171010,195857.246:ERROR process_info.cc:571] ReadMemoryInfo failed
[24792:25484:20171010,195857.246:ERROR win_multiprocess_with_temp_dir.cc:94] ProcessInfo::Initialize, pid: 22376
[  FAILED  ] CrashpadHandler.ExtensibilityCalloutsWork (73 ms)

https://build.chromium.org/p/client.crashpad/builders/crashpad_try_win_x86_wow64_rel/builds/267/steps/run%20tests/logs/stdio

[ RUN      ] CrashpadHandler.ExtensibilityCalloutsWork
[24812:24440:20171002,185723.158:ERROR file_io_win.cc:278] GetFileSizeEx: The handle is invalid. (6)
e:\b\build\slave\crashpad_try_win_x86_wow64_rel\build\crashpad\handler\crashpad_handler_test.cc(88): error:       Expected: reports.size()
      Which is: 0
To be equal to: 1u
      Which is: 1
[24156:25112:20171002,185723.162:ERROR process_info.cc:118] ReadProcessMemory bool __cdecl crashpad::`anonymous-namespace'::ReadStruct<struct crashpad::process_types::PEB<struct crashpad::process_types::internal::Traits32>>(void *,unsigned __int64,struct crashpad::process_types::PEB<struct crashpad::process_types::internal::Traits32> *): Only part of a ReadProcessMemory or WriteProcessMemory request was completed. (299)
[24156:25112:20171002,185723.162:ERROR process_info.cc:566] ReadProcessData failed
[24156:25112:20171002,185723.162:ERROR win_multiprocess_with_temp_dir.cc:94] ProcessInfo::Initialize, pid: 24812
[  FAILED  ] CrashpadHandler.ExtensibilityCalloutsWork (62 ms)

The logged messages don’t appear to be specific to the extensibility interface, but the common code is alao tested by end_to_end_test.py which we haven’t seen this sort of flake from.