Heap_lock голодание в сборщике мусора

Потоки сборщика мусора, похоже, голодали, ожидая Heap_lock. Вот трассировка стека потока GC

+   5480 _pthread_start  (in libsystem_pthread.dylib) + 168  [0x7fff9255191a]
+     5480 _pthread_body  (in libsystem_pthread.dylib) + 131  [0x7fff9255199d]
+       5480 java_start(Thread*)  (in libjvm.dylib) + 246  [0x110e82cd6]
+         5480 YieldingFlexibleGangWorker::loop()  (in libjvm.dylib) + 419  [0x110fbf8fb]
+           5480 Monitor::wait(bool, long, bool)  (in libjvm.dylib) + 375  [0x110e5c859]
+             5480 Monitor::IWait(Thread*, long)  (in libjvm.dylib) + 168  [0x110e5c5f0]
+               5480 ParkCommon(ParkEvent*, long)  (in libjvm.dylib) + 42  [0x110e5bd1a]
+                 5480 os::PlatformEvent::park()  (in libjvm.dylib) + 192  [0x110e7ea72]
+                   5480 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff92552728]
+                     5480 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff98080db6]

Мое приложение Java порождает около 300 потоков. Они также голодают. Вот пример трассировки стека:

+             5480 java_lang_String::create_from_str(char const*, Thread*)  (in libjvm.dylib) + 59  [0x110cf1ab9]
+               5480 java_lang_String::basic_create(int, Thread*)  (in libjvm.dylib) + 38  [0x110cef8a0]
+                 5480 InstanceKlass::allocate_instance(Thread*)  (in libjvm.dylib) + 31  [0x110cbd93f]
+                   5480 CollectedHeap::obj_allocate(KlassHandle, int, Thread*)  (in libjvm.dylib) + 38  [0x110cbff8a]
+                     5480 CollectedHeap::common_mem_allocate_noinit(KlassHandle, unsigned long, Thread*)  (in libjvm.dylib) + 140  [0x110ab2484]
+                       5480 GenCollectorPolicy::mem_allocate_work(unsigned long, bool, bool*)  (in libjvm.dylib) + 475  [0x110b9f313]
+                         5480 Monitor::lock(Thread*)  (in libjvm.dylib) + 132  [0x110e5c4e6]
+                           5480 Monitor::ILock(Thread*)  (in libjvm.dylib) + 310  [0x110e5c3dc]
+                             5480 ParkCommon(ParkEvent*, long)  (in libjvm.dylib) + 42  [0x110e5bd1a]
+                               5480 os::PlatformEvent::park()  (in libjvm.dylib) + 192  [0x110e7ea72]
+                                 5480 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff92552728]
+                                   5480 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff98080db6]

Голод начинается после нескольких часов работы моего приложения.

Я попробовал три сборщика: UseG1GC, UseParallelGC и UseConcMarkSweepGC с настройками по умолчанию и все еще получаю голод (не пробовал последовательный GC).

Я использую Java 1.8.0_92 (сборка 1.8.0_92-b14, Java HotSpot(TM) 64-разрядная серверная виртуальная машина, сборка 25.92-b14, смешанный режим) в OSX 10.11.5.

Достаточно места в куче Xmx6G; Я не вижу полной GC. Я увеличил стек для потоков GC XX:VMThreadStackSize до 10 МБ, но получил такое же голодание. Я также использовал более старую Java, то же голодание. Я не вижу ни одного фатального журнала ошибок ( http://www.oracle.com/technetwork/java/javase/felog-138657.html).

Пожалуйста, предложите, как отладить это голодание.

ОБНОВЛЕНИЕ 19 июня 2016 года:

Я сгруппировал висячие потоки по трассе стека, которую вы можете увидеть

Повесьте в GangWorker::loop()

5347 Thread_669608
+ 5347 thread_start  (in libsystem_pthread.dylib) + 13  [0x7fff930c3351]
+   5347 _pthread_start  (in libsystem_pthread.dylib) + 168  [0x7fff930c591a]
+     5347 _pthread_body  (in libsystem_pthread.dylib) + 131  [0x7fff930c599d]
+       5347 java_start(Thread*)  (in libjvm.dylib) + 246  [0x112485bbe]
+         5347 GangWorker::loop()  (in libjvm.dylib) + 179  [0x1125c194f]
+           5347 Monitor::wait(bool, long, bool)  (in libjvm.dylib) + 375  [0x11245f639]
+             5347 Monitor::IWait(Thread*, long)  (in libjvm.dylib) + 168  [0x11245f3d0]
+               5347 ParkCommon(ParkEvent*, long)  (in libjvm.dylib) + 42  [0x11245eafa]
+                 5347 os::PlatformEvent::park()  (in libjvm.dylib) + 192  [0x11248195a]
+                   5347 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff930c6728]
+                     5347 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff8dbbcdb6]

Зависать в ConcurrentG1RefineThread::wait_for_completed_buffers()

5347 Thread_669627
+ 5347 thread_start  (in libsystem_pthread.dylib) + 13  [0x7fff930c3351]
+   5347 _pthread_start  (in libsystem_pthread.dylib) + 168  [0x7fff930c591a]
+     5347 _pthread_body  (in libsystem_pthread.dylib) + 131  [0x7fff930c599d]
+       5347 java_start(Thread*)  (in libjvm.dylib) + 246  [0x112485bbe]
+         5347 ConcurrentG1RefineThread::run()  (in libjvm.dylib) + 90  [0x1121c054e]
+           5347 ConcurrentG1RefineThread::wait_for_completed_buffers()  (in libjvm.dylib) + 65  [0x1121c08e5]
+             5347 Monitor::wait(bool, long, bool)  (in libjvm.dylib) + 375  [0x11245f639]
+               5347 Monitor::IWait(Thread*, long)  (in libjvm.dylib) + 168  [0x11245f3d0]
+                 5347 ParkCommon(ParkEvent*, long)  (in libjvm.dylib) + 42  [0x11245eafa]
+                   5347 os::PlatformEvent::park()  (in libjvm.dylib) + 192  [0x11248195a]
+                     5347 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff930c6728]
+                       5347 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff8dbbcdb6]

Зависание потоков компилятора

5347 Thread_669680: Java: C2 CompilerThread0
+ 5347 thread_start  (in libsystem_pthread.dylib) + 13  [0x7fff930c3351]
+   5347 _pthread_start  (in libsystem_pthread.dylib) + 168  [0x7fff930c591a]
+     5347 _pthread_body  (in libsystem_pthread.dylib) + 131  [0x7fff930c599d]
+       5347 java_start(Thread*)  (in libjvm.dylib) + 246  [0x112485bbe]
+         5347 JavaThread::run()  (in libjvm.dylib) + 450  [0x112566344]
+           5347 JavaThread::thread_main_inner()  (in libjvm.dylib) + 155  [0x112564c57]
+             5347 CompileBroker::compiler_thread_loop()  (in libjvm.dylib) + 376  [0x1121b89ce]
+               5347 CompileQueue::get()  (in libjvm.dylib) + 122  [0x1121b87ac]
+                 5347 Monitor::wait(bool, long, bool)  (in libjvm.dylib) + 222  [0x11245f5a0]
+                   5347 Monitor::IWait(Thread*, long)  (in libjvm.dylib) + 168  [0x11245f3d0]
+                     5347 os::PlatformEvent::park(long)  (in libjvm.dylib) + 404  [0x112482844]
+                       5347 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff930c6728]
+                         5347 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff8dbbcdb6]

Кроме того, оставшиеся потоки имеют эти следы стека (одна строка на поток)

 _OSSpinLockLockSlow ! syscall_thread_switch _OSSpinLockLockSlow _pthread_lookup_thread pthread_kill os::SuspendedThreadTask::internal_do_task() os::SuspendedThreadTask::run() ThreadSampleClosure::sample_thread(JavaThread*,     5347 Thread_669693
 __accept BsdAttachListener::dequeue() AttachListener::dequeue() attach_listener_thread_entry(JavaThread*, JavaThread::thread_main_inner() JavaThread::run() java_start(Thread*) _pthread_body     5347 Thread_669711: Java: Attach Listener
 __psynch_cvwait _pthread_cond_wait Parker::park(bool, Unsafe_Park ??? ???       5347 Thread_669715: Java: RMI Scheduler(0)
 __psynch_cvwait _pthread_cond_wait Parker::park(bool, Unsafe_Park ??? ??? ??? ???     5347 Thread_669777: Java: log4j2.jmx.notif1
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ObjectMonitor::wait(long, ObjectSynchronizer::wait(Handle, JVM_MonitorWait ??? ???     5347 Thread_669603
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ObjectMonitor::wait(long, ObjectSynchronizer::wait(Handle, JVM_MonitorWait ??? ???     5347 Thread_669657: Java: Reference Handler
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ObjectMonitor::wait(long, ObjectSynchronizer::wait(Handle, JVM_MonitorWait ??? ???     5347 Thread_669658: Java: Finalizer
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned G1CollectedHeap::attempt_allocation(unsigned     5347 Thread_669669: Java: JFR request timer
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned G1CollectedHeap::attempt_allocation(unsigned     5347 Thread_669712: Java: RMI TCP Accept-0
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned G1CollectedHeap::attempt_allocation(unsigned     5347 Thread_669720: Java: JMX server connection timeout 29
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned G1CollectedHeap::attempt_allocation(unsigned     5347 Thread_683755: Java: RMI TCP Connection(65)-192.168.1.167
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned G1CollectedHeap::attempt_allocation(unsigned     5347 Thread_813902: Java: RMI TCP Connection(634)-192.168.1.167
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) SafepointSynchronize::begin() VMThread::loop()     5347 Thread_669656
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock_without_safepoint_check() SafepointSynchronize::block(JavaThread*) ThreadStateTransition::trans_and_fence(JavaThreadState,     5347 Thread_669676: Java: VM JFR Buffer Thread
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::IWait(Thread*, Monitor::wait(bool, ConcurrentMarkThread::sleepBeforeNextCycle() ConcurrentMarkThread::run()     5347 Thread_669645
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::IWait(Thread*, Monitor::wait(bool, ServiceThread::service_thread_entry(JavaThread*, JavaThread::thread_main_inner()     5347 Thread_669692: Java: Service Thread
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::IWait(Thread*, Monitor::wait(bool, SurrogateLockerThread::loop() JavaThread::thread_main_inner()     5347 Thread_669667: Java: Surrogate Locker Thread (Concurrent GC)
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::IWait(Thread*, Monitor::wait(bool, SuspendibleThreadSet::join() ConcurrentG1RefineThread::sample_young_list_rs_lengths()     5347 Thread_669626
 __semwait_signal pthread_join ContinueInNewThread0 ContinueInNewThread JVMInit JLI_Launch main apple_main     5347 Thread_669600
 __sigsuspend SR_handler(int, _sigtramp vframeStreamCommon::fill_from_frame() _pthread_lookup_thread pthread_kill closefd ???     5347 Thread_671069: Java: DelayClose
 kevent_qos _dispatch_mgr_invoke _dispatch_mgr_thread          5347 Thread_669601   DispatchQueue_2: com.apple.libdispatch-manager  (serial)
 mach_msg_trap mach_msg __CFRunLoopServiceMachPort __CFRunLoopRun CFRunLoopRunSpecific CreateExecutionEnvironment JLI_Launch main     5347 Thread_669599   DispatchQueue_1: com.apple.main-thread  (serial)
 semaphore_wait_trap check_pending_signals(bool) signal_thread_entry(JavaThread*, JavaThread::thread_main_inner() JavaThread::run() java_start(Thread*) _pthread_body _pthread_start     5347 Thread_669668: Java: Signal Dispatcher
 __psynch_cvwait _pthread_cond_wait os::PlatformEvent::park() ParkCommon(ParkEvent*, Monitor::ILock(Thread*) Monitor::lock(Thread*) G1CollectedHeap::attempt_allocation_slow(unsigned 5347 Thread_815183: Java: RMI TCP Connection(635)-192.168.1.167

Я пропустил свои 300 потоков, потому что Java не находится в безопасной точке (jstack не работает), поэтому голодание не может быть из-за моих (пользовательских) потоков. Истощение должно быть вызвано потоками виртуальных машин (все из которых я перечислил выше со следами стека).

ОБНОВЛЕНИЕ 24 июня 2016 года:

Я сделал jstack -F, и мои потоки заблокированы, некоторые из них в операциях с памятью, как этот

Thread 200451: (state = BLOCKED)
 - java.util.Arrays.copyOf(char[], int) @bci=1, line=3332 (Compiled frame)
 - java.lang.AbstractStringBuilder.expandCapacity(int) @bci=43, line=137 (Compiled frame)
 - java.lang.AbstractStringBuilder.ensureCapacityInternal(int) @bci=12, line=121 (Compiled frame)
 - java.lang.AbstractStringBuilder.append(java.lang.String) @bci=21, line=421 (Compiled frame)
 - java.lang.StringBuilder.append(java.lang.String) @bci=2, line=136 (Compiled frame)

но jstack не показывает потоки сборщика мусора. Чтобы получить это, я запускаю пример команды для того же процесса, и так же, как прежде, чем вы можете увидеть голодные потоки GC:

+ 5229 ???  (in <unknown binary>)  [0x11064ae07]
+   5229 OptoRuntime::new_array_nozero_C(Klass*, int, JavaThread*)  (in libjvm.dylib) + 54  [0x10f8da854]
+     5229 TypeArrayKlass::allocate_common(int, bool, Thread*)  (in libjvm.dylib) + 138  [0x10f975406]
+       5229 CollectedHeap::common_mem_allocate_noinit(KlassHandle, unsigned long, Thread*)  (in libjvm.dylib) + 105  [0x10f4b2bbb]
+         5229 CollectedHeap::allocate_from_tlab_slow(KlassHandle, Thread*, unsigned long)  (in libjvm.dylib) + 283  [0x10f59e2af]
+           5229 G1CollectedHeap::allocate_new_tlab(unsigned long)  (in libjvm.dylib) + 28  [0x10f65590e]
+             5229 G1CollectedHeap::attempt_allocation(unsigned long, unsigned int*, unsigned int*)  (in libjvm.dylib) + 182  [0x10f65e42a]
+               5229 G1CollectedHeap::attempt_allocation_slow(unsigned long, unsigned char, unsigned int*, unsigned int*)  (in libjvm.dylib) + 339  [0x10f6576ab]
+                 5229 Monitor::lock(Thread*)  (in libjvm.dylib) + 132  [0x10f85f2c6]
+                   5229 Monitor::ILock(Thread*)  (in libjvm.dylib) + 310  [0x10f85f1bc]
+                     5229 ParkCommon(ParkEvent*, long)  (in libjvm.dylib) + 42  [0x10f85eafa]
+                       5229 os::PlatformEvent::park()  (in libjvm.dylib) + 192  [0x10f88195a]
+                         5229 _pthread_cond_wait  (in libsystem_pthread.dylib) + 767  [0x7fff930c6728]
+                           5229 __psynch_cvwait  (in libsystem_kernel.dylib) + 10  [0x7fff8dbbcdb6]

Голод исчез, когда я запустил свое приложение всего с 30 потоками (вместо 300).

Нам нужно выяснить, какую блокировку ожидает __psynch_cvwait сборщика мусора, но jstack этого не говорит. Вероятно, проверка кода - лучший выбор.

0 ответов

Другие вопросы по тегам