Thread unsafe signal handling in main_common_test #6083
Description
It seems that the libevent signal handler might be invoked on an arbitrary thread. This then cause issues with main_common_test when we enable TSAN on libevent in #6061. E.g. in AdminRequestGetStatsAndKill:
WARNING: ThreadSanitizer: data race (pid=21713)
Read of size 8 at 0x00000455e1c8 by main thread:
#0 evsig_handler /build/tmp/_bazel_bazel/b570b5ccd0454dc9af9f65ab1833764d/execroot/envoy/external/com_github_libevent_libevent/signal.c:385:6 (main_common_test+0x386f1c2)
#1 __tsan::CallUserSignalHandler(__tsan::ThreadState*, bool, bool, bool, int, __sanitizer::__sanitizer_siginfo*, void*) <null> (main_common_test+0x116abb0)
#2 Envoy::AdminRequestTest_AdminRequestGetStatsAndKill_Test::TestBody() /proc/self/cwd/test/exe/main_common_test.cc:249:3 (main_common_test+0x11edde6)
#3 void testing::internal::HandleSehExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2424:10 (main_common_test+0x3a1eaa6)
#4 void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2460:14 (main_common_test+0x3a03f01)
#5 testing::Test::Run() /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2499:5 (main_common_test+0x39e406b)
#6 testing::TestInfo::Run() /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2675:11 (main_common_test+0x39e51a9)
#7 testing::TestSuite::Run() /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2803:28 (main_common_test+0x39e5c14)
#8 testing::internal::UnitTestImpl::RunAllTests() /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:5241:44 (main_common_test+0x39f8b14)
#9 bool testing::internal::HandleSehExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2424:10 (main_common_test+0x3a24976)
#10 bool testing::internal::HandleExceptionsInMethodIfSupported<testing::internal::UnitTestImpl, bool>(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:2460:14 (main_common_test+0x3a07f17)
#11 testing::UnitTest::Run() /proc/self/cwd/external/com_google_googletest/googletest/src/gtest.cc:4843:10 (main_common_test+0x39f857e)
#12 RUN_ALL_TESTS() /proc/self/cwd/external/com_google_googletest/googletest/include/gtest/gtest.h:2499:46 (main_common_test+0x296f087)
#13 Envoy::TestRunner::RunTests(int, char**) /proc/self/cwd/./test/test_runner.h:57:12 (main_common_test+0x296eca4)
#14 main /proc/self/cwd/test/main.cc:39:10 (main_common_test+0x296df4b)
Previous write of size 8 at 0x00000455e1c8 by thread T5 (mutexes: write M636550213812770000, write M2072):
#0 evsig_set_base_ /build/tmp/_bazel_bazel/b570b5ccd0454dc9af9f65ab1833764d/execroot/envoy/external/com_github_libevent_libevent/signal.c:123:13 (main_common_test+0x386e4d0)
#1 event_base_loop /build/tmp/_bazel_bazel/b570b5ccd0454dc9af9f65ab1833764d/execroot/envoy/external/com_github_libevent_libevent/event.c:1901:3 (main_common_test+0x385e084)
#2 Envoy::Event::DispatcherImpl::run(Envoy::Event::Dispatcher::RunType) /proc/self/cwd/source/common/event/dispatcher_impl.cc:171:3 (main_common_test+0x2a6a74e)
#3 Envoy::Server::InstanceImpl::run() /proc/self/cwd/source/server/server.cc:475:16 (main_common_test+0x1fc0d52)
#4 Envoy::MainCommonBase::run() /proc/self/cwd/source/exe/main_common.cc:114:14 (main_common_test+0x126363f)
#5 Envoy::MainCommon::run() /proc/self/cwd/bazel-out/k8-dbg/bin/source/exe/_virtual_includes/envoy_main_common_lib/exe/main_common.h:86:29 (main_common_test+0x11f4c8f)
#6 Envoy::AdminRequestTest::startEnvoy()::'lambda'()::operator()() const /proc/self/cwd/test/exe/main_common_test.cc:195:35 (main_common_test+0x124a350)
#7 std::_Function_handler<void (), Envoy::AdminRequestTest::startEnvoy()::'lambda'()>::_M_invoke(std::_Any_data const&) /usr/lib/gcc/x86_64-linux-gnu/7.4.0/../../../../include/c++/7.4.0/bits/std_function.h:316:2 (main_common_test+0x1249faa)
#8 std::function<void ()>::operator()() const /usr/lib/gcc/x86_64-linux-gnu/7.4.0/../../../../include/c++/7.4.0/bits/std_function.h:706:14 (main_common_test+0x13f990e)
#9 Envoy::Thread::ThreadImplPosix::ThreadImplPosix(std::function<void ()>)::$_0::operator()(void*) const /proc/self/cwd/source/common/common/posix/thread_impl.cc:38:35 (main_common_test+0x38add98)
#10 Envoy::Thread::ThreadImplPosix::ThreadImplPosix(std::function<void ()>)::$_0::__invoke(void*) /proc/self/cwd/source/common/common/posix/thread_impl.cc:37:33 (main_common_test+0x38add28)
Location is global 'evsig_base' of size 8 at 0x00000455e1c8 (main_common_test+0x00000455e1c8)
evsig_base seems like its correctly modeled as atomic, I think we need to wait until the main server thread is running before performing the kill.