Opt signal trace

apache · Dec 19, 2024 · 6b52eba · 6b52eba
1 parent c44cb27
commit 6b52eba
Show file tree

Hide file tree

Showing 15 changed files with 536 additions and 335 deletions.
diff --git a/README_cn.md b/README_cn.md
@@ -38,6 +38,7 @@
     * [bthread or not](docs/cn/bthread_or_not.md)
     * [thread-local](docs/cn/thread_local.md)
     * [Execution Queue](docs/cn/execution_queue.md)
+    * [bthread tracer](docs/cn/bthread_tracer.md)
   * Client
     * [基础功能](docs/cn/client.md)
     * [错误码](docs/cn/error_code.md)

diff --git a/docs/cn/bthread_tracer.md b/docs/cn/bthread_tracer.md
@@ -63,7 +63,17 @@ jump_stack是bthread挂起或者运行的必经之路，也是STB的拦截点。
 3. 访问服务的内置服务：`http://ip:port/bthreads/<bthread_id>?st=1`或者代码里调用`bthread::stack_trace()`函数。
 4. 如果希望追踪pthread的调用栈，在对应pthread上调用`bthread::init_for_pthread_stack_trace()`函数获取一个伪bthread_t，然后使用步骤3即可获取pthread调用栈。
 
+下面是追踪bthread调用栈的输出示例：
+```shell
+#0 0x00007fdbbed500b5 __clock_gettime_2
+#1 0x000000000041f2b6 butil::cpuwide_time_ns()
+#2 0x000000000041f289 butil::cpuwide_time_us()
+#3 0x000000000041f1b9 butil::EveryManyUS::operator bool()
+#4 0x0000000000413289 (anonymous namespace)::spin_and_log()
+#5 0x00007fdbbfa58dc0 bthread::TaskGroup::task_runner()
+```
+
 # 相关flag
 
-- `enable_fast_unwind`：是否启用快速回溯功能，默认为true。大多数情况下，不需要关闭快速回溯功能。除非你关注的调用栈函数名转换失败，显示为`<unknown>`，则可以尝试关闭快速回溯功能，但这会导致性能下降。以包含30帧的调用栈举例，快速回溯只需要400~500us，而关闭快速回溯则需要4ms左右，性能下降了近10倍。
+- `enable_fast_unwind`：是否启用快速回溯功能，默认为true。大多数情况下，不需要关闭快速回溯功能。除非你关注的调用栈函数名转换失败，显示为`<unknown>`，则可以尝试关闭快速回溯功能，但这会导致性能下降。以包含30帧的调用栈举例，快速回溯基本上在200us以内就可以完成，而关闭快速回溯则需要4ms左右，性能下降了近20倍。
 - `signal_trace_timeout_ms`：信号追踪模式的超时时间，默认为50ms。虽然libunwind文档显示回溯功能是异步信号安全的，但是[gpertools社区发现libunwind在某些情况下会死锁](https://github.com/gperftools/gperftools/issues/775)，所以TaskTracer会设置了超时时间，超时后会放弃回溯，打破死锁。
diff --git a/src/brpc/builtin/bthreads_service.cpp b/src/brpc/builtin/bthreads_service.cpp
@@ -23,7 +23,7 @@
 #include "brpc/builtin/bthreads_service.h"
 
 namespace bthread {
-void print_task(std::ostream& os, bthread_t tid);
+extern void print_task(std::ostream& os, bthread_t tid);
 }
 
 

diff --git a/src/brpc/input_messenger.cpp b/src/brpc/input_messenger.cpp
@@ -287,8 +287,7 @@ int InputMessenger::ProcessNewMessage(
         // This unique_ptr prevents msg to be lost before transfering
         // ownership to last_msg
         DestroyingPtr<InputMessageBase> msg(pr.message());
-        QueueMessage(last_msg.release(), &num_bthread_created,
-                            m->_keytable_pool);
+        QueueMessage(last_msg.release(), &num_bthread_created, m->_keytable_pool);
         if (_handlers[index].process == NULL) {
             LOG(ERROR) << "process of index=" << index << " is NULL";
             continue;

diff --git a/src/brpc/reloadable_flags.cpp b/src/brpc/reloadable_flags.cpp
@@ -15,10 +15,6 @@
 // specific language governing permissions and limitations
 // under the License.
 
-
-#include <unistd.h>                    // write, _exit
-#include <gflags/gflags.h>
-#include "butil/macros.h"
 #include "brpc/reloadable_flags.h"
 
 namespace brpc {
@@ -62,37 +58,25 @@ bool NonNegativeInteger(const char*, int64_t val) {
     return val >= 0;
 }
 
-template <typename T>
-static bool RegisterFlagValidatorOrDieImpl(
-    const T* flag, bool (*validate_fn)(const char*, T val)) {
-    if (GFLAGS_NS::RegisterFlagValidator(flag, validate_fn)) {
-        return true;
-    }
-    // Error printed by gflags does not have newline. Add one to it.
-    char newline = '\n';
-    butil::ignore_result(write(2, &newline, 1));
-    _exit(1);
-}
-
 bool RegisterFlagValidatorOrDie(const bool* flag,
                                 bool (*validate_fn)(const char*, bool)) {
-    return RegisterFlagValidatorOrDieImpl(flag, validate_fn);
+    return butil::RegisterFlagValidatorOrDieImpl(flag, validate_fn);
 }
 bool RegisterFlagValidatorOrDie(const int32_t* flag,
                                 bool (*validate_fn)(const char*, int32_t)) {
-    return RegisterFlagValidatorOrDieImpl(flag, validate_fn);
+    return butil::RegisterFlagValidatorOrDieImpl(flag, validate_fn);
 }
 bool RegisterFlagValidatorOrDie(const int64_t* flag,
                                 bool (*validate_fn)(const char*, int64_t)) {
-    return RegisterFlagValidatorOrDieImpl(flag, validate_fn);
+    return butil::RegisterFlagValidatorOrDieImpl(flag, validate_fn);
 }
 bool RegisterFlagValidatorOrDie(const uint64_t* flag,
                                 bool (*validate_fn)(const char*, uint64_t)) {
-    return RegisterFlagValidatorOrDieImpl(flag, validate_fn);
+    return butil::RegisterFlagValidatorOrDieImpl(flag, validate_fn);
 }
 bool RegisterFlagValidatorOrDie(const double* flag,
                                 bool (*validate_fn)(const char*, double)) {
-    return RegisterFlagValidatorOrDieImpl(flag, validate_fn);
+    return butil::RegisterFlagValidatorOrDieImpl(flag, validate_fn);
 }
 
 } // namespace brpc
diff --git a/src/brpc/reloadable_flags.h b/src/brpc/reloadable_flags.h
@@ -19,28 +19,7 @@
 #ifndef BRPC_RELOADABLE_FLAGS_H
 #define BRPC_RELOADABLE_FLAGS_H
 
-// To brpc developers: This is a header included by user, don't depend
-// on internal structures, use opaque pointers instead.
-
-#include <stdint.h>
-
-// Register an always-true valiator to a gflag so that the gflag is treated as
-// reloadable by brpc. If a validator exists, abort the program.
-// You should call this macro within global scope. for example:
-//
-//   DEFINE_int32(foo, 0, "blah blah");
-//   BRPC_VALIDATE_GFLAG(foo, brpc::PassValidate);
-//
-// This macro does not work for string-flags because they're thread-unsafe to
-// modify directly. To emphasize this, you have to write the validator by
-// yourself and use GFLAGS_NS::GetCommandLineOption() to acess the flag.
-#define BRPC_VALIDATE_GFLAG(flag, validate_fn)                     \
-    namespace brpc_flags {}                                        \
-    const int register_FLAGS_ ## flag ## _dummy                    \
-                 __attribute__((__unused__)) =                     \
-        ::brpc::RegisterFlagValidatorOrDie(                        \
-            &FLAGS_##flag, (validate_fn))
-
+#include "butil/reloadable_flags.h"
 
 namespace brpc {
 
@@ -59,18 +38,18 @@ extern bool PositiveInteger(const char*, uint64_t);
 extern bool NonNegativeInteger(const char*, int32_t);
 extern bool NonNegativeInteger(const char*, int64_t);
 
-extern bool RegisterFlagValidatorOrDie(const bool* flag,
-                                  bool (*validate_fn)(const char*, bool));
-extern bool RegisterFlagValidatorOrDie(const int32_t* flag,
-                                  bool (*validate_fn)(const char*, int32_t));
-extern bool RegisterFlagValidatorOrDie(const uint32_t* flag,
-                                  bool (*validate_fn)(const char*, uint32_t));
-extern bool RegisterFlagValidatorOrDie(const int64_t* flag,
-                                  bool (*validate_fn)(const char*, int64_t));
-extern bool RegisterFlagValidatorOrDie(const uint64_t* flag,
-                                  bool (*validate_fn)(const char*, uint64_t));
-extern bool RegisterFlagValidatorOrDie(const double* flag,
-                                  bool (*validate_fn)(const char*, double));
+extern bool RegisterFlagValidatorOrDie(
+    const bool* flag, bool (*validate_fn)(const char*, bool));
+extern bool RegisterFlagValidatorOrDie(
+    const int32_t* flag, bool (*validate_fn)(const char*, int32_t));
+extern bool RegisterFlagValidatorOrDie(
+    const uint32_t* flag, bool (*validate_fn)(const char*, uint32_t));
+extern bool RegisterFlagValidatorOrDie(
+    const int64_t* flag, bool (*validate_fn)(const char*, int64_t));
+extern bool RegisterFlagValidatorOrDie(
+    const uint64_t* flag, bool (*validate_fn)(const char*, uint64_t));
+extern bool RegisterFlagValidatorOrDie(
+    const double* flag, bool (*validate_fn)(const char*, double));
 } // namespace brpc
 
 

diff --git a/src/brpc/shared_object.h b/src/brpc/shared_object.h
@@ -19,53 +19,11 @@
 #ifndef BRPC_SHARED_OBJECT_H
 #define BRPC_SHARED_OBJECT_H
 
-#include "butil/intrusive_ptr.hpp"                   // butil::intrusive_ptr
-#include "butil/atomicops.h"
-
+#include "butil/shared_object.h"
 
 namespace brpc {
 
-// Inherit this class to be intrusively shared. Comparing to shared_ptr,
-// intrusive_ptr saves one malloc (for shared_count) and gets better cache
-// locality when the ref/deref are frequent, in the cost of lack of weak_ptr
-// and worse interface.
-class SharedObject {
-friend void intrusive_ptr_add_ref(SharedObject*);
-friend void intrusive_ptr_release(SharedObject*);
-
-public:
-    SharedObject() : _nref(0) { }
-    int ref_count() const { return _nref.load(butil::memory_order_relaxed); }
-
-    // Add ref and returns the ref_count seen before added.
-    // The effect is basically same as butil::intrusive_ptr<T>(obj).detach()
-    // except that the latter one does not return the seen ref_count which is
-    // useful in some scenarios.
-    int AddRefManually()
-    { return _nref.fetch_add(1, butil::memory_order_relaxed); }
-
-    // Remove one ref, if the ref_count hit zero, delete this object.
-    // Same as butil::intrusive_ptr<T>(obj, false).reset(NULL)
-    void RemoveRefManually() {
-        if (_nref.fetch_sub(1, butil::memory_order_release) == 1) {
-            butil::atomic_thread_fence(butil::memory_order_acquire);
-            delete this;
-        }
-    }
-
-protected:
-    virtual ~SharedObject() { }
-private:
-    butil::atomic<int> _nref;
-};
-
-inline void intrusive_ptr_add_ref(SharedObject* obj) {
-    obj->AddRefManually();
-}
-
-inline void intrusive_ptr_release(SharedObject* obj) {
-    obj->RemoveRefManually();
-}
+using butil::SharedObject;
 
 } // namespace brpc
 

diff --git a/src/bthread/task_control.cpp b/src/bthread/task_control.cpp
@@ -216,7 +216,7 @@ int TaskControl::init(int concurrency) {
     }
 
 #ifdef BRPC_BTHREAD_TRACER
-    if (_task_tracer.Init() != 0) {
+    if (!_task_tracer.Init()) {
         LOG(ERROR) << "Fail to init TaskTracer";
         return -1;
     }

diff --git a/src/bthread/task_group.cpp b/src/bthread/task_group.cpp
@@ -660,8 +660,13 @@ void TaskGroup::sched_to(TaskGroup** pg, TaskMeta* next_meta) {
                 CHECK(cur_meta->stack == g->_main_stack);
             }
 #endif
+        } /* else because of ending_sched(including pthread_task->pthread_task). */
+#ifdef BRPC_BTHREAD_TRACER
+        else {
+            // _cur_meta: TASK_STATUS_FIRST_READY -> TASK_STATUS_RUNNING.
+            TaskTracer::set_running_status(g->tid(), g->_cur_meta);
         }
-        // else because of ending_sched(including pthread_task->pthread_task)
+#endif // BRPC_BTHREAD_TRACER
     } else {
         LOG(FATAL) << "bthread=" << g->current_tid() << " sched_to itself!";
     }
@@ -1013,13 +1018,15 @@ void print_task(std::ostream& os, bthread_t tid) {
            << "}\nhas_tls=" << has_tls
            << "\nuptime_ns=" << butil::cpuwide_time_ns() - cpuwide_start_ns
            << "\ncputime_ns=" << stat.cputime_ns
-            << "\nnswitch=" << stat.nswitch
+           << "\nnswitch=" << stat.nswitch
 #ifdef BRPC_BTHREAD_TRACER
            << "\nstatus=" << status
            << "\ntraced=" << traced
            << "\nworker_tid=" << worker_tid;
-#endif // BRPC_BTHREAD_TRACER
+#else
            ;
+           (void)status;(void)traced;(void)worker_tid;
+#endif // BRPC_BTHREAD_TRACER
     }
 }