src/core/lib/iomgr/executor.cc

   1 /*
   2  *
   3  * Copyright 2015 gRPC authors.
   4  *
   5  * Licensed under the Apache License, Version 2.0 (the "License");
   6  * you may not use this file except in compliance with the License.
   7  * You may obtain a copy of the License at
   8  *
   9  *     http://www.apache.org/licenses/LICENSE-2.0
  10  *
  11  * Unless required by applicable law or agreed to in writing, software
  12  * distributed under the License is distributed on an "AS IS" BASIS,
  13  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  14  * See the License for the specific language governing permissions and
  15  * limitations under the License.
  16  *
  17  */
  18
  19 #include <grpc/support/port_platform.h>
  20
  21 #include "src/core/lib/iomgr/executor.h"
  22
  23 #include <string.h>
  24
  25 #include <grpc/support/alloc.h>
  26 #include <grpc/support/cpu.h>
  27 #include <grpc/support/log.h>
  28 #include <grpc/support/sync.h>
  29
  30 #include "src/core/lib/debug/stats.h"
  31 #include "src/core/lib/gpr/tls.h"
  32 #include "src/core/lib/gpr/useful.h"
  33 #include "src/core/lib/gprpp/memory.h"
  34 #include "src/core/lib/iomgr/exec_ctx.h"
  35 #include "src/core/lib/iomgr/iomgr.h"
  36
  37 #define MAX_DEPTH 2
  38
  39 #define EXECUTOR_TRACE(format, ...)                     \
  40   if (executor_trace.enabled()) {                       \
  41     gpr_log(GPR_INFO, "EXECUTOR " format, __VA_ARGS__); \
  42   }
  43
  44 #define EXECUTOR_TRACE0(str)            \
  45   if (executor_trace.enabled()) {       \
  46     gpr_log(GPR_INFO, "EXECUTOR " str); \
  47   }
  48
  49 namespace grpc_core {
  50 namespace {
  51
  52 GPR_TLS_DECL(g_this_thread_state);
  53
  54 Executor* executors[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)];
  55
  56 void default_enqueue_short(grpc_closure* closure, grpc_error* error) {
  57   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Enqueue(
  58       closure, error, true /* is_short */);
  59 }
  60
  61 void default_enqueue_long(grpc_closure* closure, grpc_error* error) {
  62   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Enqueue(
  63       closure, error, false /* is_short */);
  64 }
  65
  66 void resolver_enqueue_short(grpc_closure* closure, grpc_error* error) {
  67   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Enqueue(
  68       closure, error, true /* is_short */);
  69 }
  70
  71 void resolver_enqueue_long(grpc_closure* closure, grpc_error* error) {
  72   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Enqueue(
  73       closure, error, false /* is_short */);
  74 }
  75
  76 const grpc_closure_scheduler_vtable
  77     vtables_[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)]
  78             [static_cast<size_t>(ExecutorJobType::NUM_JOB_TYPES)] = {
  79                 {{&default_enqueue_short, &default_enqueue_short,
  80                   "def-ex-short"},
  81                  {&default_enqueue_long, &default_enqueue_long, "def-ex-long"}},
  82                 {{&resolver_enqueue_short, &resolver_enqueue_short,
  83                   "res-ex-short"},
  84                  {&resolver_enqueue_long, &resolver_enqueue_long,
  85                   "res-ex-long"}}};
  86
  87 grpc_closure_scheduler
  88     schedulers_[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)]
  89                [static_cast<size_t>(ExecutorJobType::NUM_JOB_TYPES)] = {
  90                    {{&vtables_[static_cast<size_t>(ExecutorType::DEFAULT)]
  91                               [static_cast<size_t>(ExecutorJobType::SHORT)]},
  92                     {&vtables_[static_cast<size_t>(ExecutorType::DEFAULT)]
  93                               [static_cast<size_t>(ExecutorJobType::LONG)]}},
  94                    {{&vtables_[static_cast<size_t>(ExecutorType::RESOLVER)]
  95                               [static_cast<size_t>(ExecutorJobType::SHORT)]},
  96                     {&vtables_[static_cast<size_t>(ExecutorType::RESOLVER)]
  97                               [static_cast<size_t>(ExecutorJobType::LONG)]}}};
  98
  99 }  // namespace
 100
 101 TraceFlag executor_trace(false, "executor");
 102
 103 Executor::Executor(const char* name) : name_(name) {
 104   adding_thread_lock_ = GPR_SPINLOCK_STATIC_INITIALIZER;
 105   gpr_atm_rel_store(&num_threads_, 0);
 106   max_threads_ = GPR_MAX(1, 2 * gpr_cpu_num_cores());
 107 }
 108
 109 void Executor::Init() { SetThreading(true); }
 110
 111 size_t Executor::RunClosures(const char* executor_name,
 112                              grpc_closure_list list) {
 113   size_t n = 0;
 114
 115   // In the executor, the ExecCtx for the thread is declared in the executor
 116   // thread itself, but this is the point where we could start seeing
 117   // application-level callbacks. No need to create a new ExecCtx, though,
 118   // since there already is one and it is flushed (but not destructed) in this
 119   // function itself.
 120   grpc_core::ApplicationCallbackExecCtx callback_exec_ctx(
 121       GRPC_APP_CALLBACK_EXEC_CTX_FLAG_IS_INTERNAL_THREAD);
 122
 123   grpc_closure* c = list.head;
 124   while (c != nullptr) {
 125     grpc_closure* next = c->next_data.next;
 126     grpc_error* error = c->error_data.error;
 127 #ifndef NDEBUG
 128     EXECUTOR_TRACE("(%s) run %p [created by %s:%d]", executor_name, c,
 129                    c->file_created, c->line_created);
 130     c->scheduled = false;
 131 #else
 132     EXECUTOR_TRACE("(%s) run %p", executor_name, c);
 133 #endif
 134     c->cb(c->cb_arg, error);
 135     GRPC_ERROR_UNREF(error);
 136     c = next;
 137     n++;
 138     grpc_core::ExecCtx::Get()->Flush();
 139   }
 140
 141   return n;
 142 }
 143
 144 bool Executor::IsThreaded() const {
 145   return gpr_atm_acq_load(&num_threads_) > 0;
 146 }
 147
 148 void Executor::SetThreading(bool threading) {
 149   gpr_atm curr_num_threads = gpr_atm_acq_load(&num_threads_);
 150   EXECUTOR_TRACE("(%s) SetThreading(%d) begin", name_, threading);
 151
 152   if (threading) {
 153     if (curr_num_threads > 0) {
 154       EXECUTOR_TRACE("(%s) SetThreading(true). curr_num_threads == 0", name_);
 155       return;
 156     }
 157
 158     GPR_ASSERT(num_threads_ == 0);
 159     gpr_atm_rel_store(&num_threads_, 1);
 160     gpr_tls_init(&g_this_thread_state);
 161     thd_state_ = static_cast<ThreadState*>(
 162         gpr_zalloc(sizeof(ThreadState) * max_threads_));
 163
 164     for (size_t i = 0; i < max_threads_; i++) {
 165       gpr_mu_init(&thd_state_[i].mu);
 166       gpr_cv_init(&thd_state_[i].cv);
 167       thd_state_[i].id = i;
 168       thd_state_[i].name = name_;
 169       thd_state_[i].thd = grpc_core::Thread();
 170       thd_state_[i].elems = GRPC_CLOSURE_LIST_INIT;
 171     }
 172
 173     thd_state_[0].thd =
 174         grpc_core::Thread(name_, &Executor::ThreadMain, &thd_state_[0]);
 175     thd_state_[0].thd.Start();
 176   } else {  // !threading
 177     if (curr_num_threads == 0) {
 178       EXECUTOR_TRACE("(%s) SetThreading(false). curr_num_threads == 0", name_);
 179       return;
 180     }
 181
 182     for (size_t i = 0; i < max_threads_; i++) {
 183       gpr_mu_lock(&thd_state_[i].mu);
 184       thd_state_[i].shutdown = true;
 185       gpr_cv_signal(&thd_state_[i].cv);
 186       gpr_mu_unlock(&thd_state_[i].mu);
 187     }
 188
 189     /* Ensure no thread is adding a new thread. Once this is past, then no
 190      * thread will try to add a new one either (since shutdown is true) */
 191     gpr_spinlock_lock(&adding_thread_lock_);
 192     gpr_spinlock_unlock(&adding_thread_lock_);
 193
 194     curr_num_threads = gpr_atm_no_barrier_load(&num_threads_);
 195     for (gpr_atm i = 0; i < curr_num_threads; i++) {
 196       thd_state_[i].thd.Join();
 197       EXECUTOR_TRACE("(%s) Thread %" PRIdPTR " of %" PRIdPTR " joined", name_,
 198                      i + 1, curr_num_threads);
 199     }
 200
 201     gpr_atm_rel_store(&num_threads_, 0);
 202     for (size_t i = 0; i < max_threads_; i++) {
 203       gpr_mu_destroy(&thd_state_[i].mu);
 204       gpr_cv_destroy(&thd_state_[i].cv);
 205       RunClosures(thd_state_[i].name, thd_state_[i].elems);
 206     }
 207
 208     gpr_free(thd_state_);
 209     gpr_tls_destroy(&g_this_thread_state);
 210
 211     // grpc_iomgr_shutdown_background_closure() will close all the registered
 212     // fds in the background poller, and wait for all pending closures to
 213     // finish. Thus, never call Executor::SetThreading(false) in the middle of
 214     // an application.
 215     // TODO(guantaol): create another method to finish all the pending closures
 216     // registered in the background poller by grpc_core::Executor.
 217     grpc_iomgr_shutdown_background_closure();
 218   }
 219
 220   EXECUTOR_TRACE("(%s) SetThreading(%d) done", name_, threading);
 221 }
 222
 223 void Executor::Shutdown() { SetThreading(false); }
 224
 225 void Executor::ThreadMain(void* arg) {
 226   ThreadState* ts = static_cast<ThreadState*>(arg);
 227   gpr_tls_set(&g_this_thread_state, reinterpret_cast<intptr_t>(ts));
 228
 229   grpc_core::ExecCtx exec_ctx(GRPC_EXEC_CTX_FLAG_IS_INTERNAL_THREAD);
 230
 231   size_t subtract_depth = 0;
 232   for (;;) {
 233     EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: step (sub_depth=%" PRIdPTR ")",
 234                    ts->name, ts->id, subtract_depth);
 235
 236     gpr_mu_lock(&ts->mu);
 237     ts->depth -= subtract_depth;
 238     // Wait for closures to be enqueued or for the executor to be shutdown
 239     while (grpc_closure_list_empty(ts->elems) && !ts->shutdown) {
 240       ts->queued_long_job = false;
 241       gpr_cv_wait(&ts->cv, &ts->mu, gpr_inf_future(GPR_CLOCK_MONOTONIC));
 242     }
 243
 244     if (ts->shutdown) {
 245       EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: shutdown", ts->name, ts->id);
 246       gpr_mu_unlock(&ts->mu);
 247       break;
 248     }
 249
 250     GRPC_STATS_INC_EXECUTOR_QUEUE_DRAINED();
 251     grpc_closure_list closures = ts->elems;
 252     ts->elems = GRPC_CLOSURE_LIST_INIT;
 253     gpr_mu_unlock(&ts->mu);
 254
 255     EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: execute", ts->name, ts->id);
 256
 257     grpc_core::ExecCtx::Get()->InvalidateNow();
 258     subtract_depth = RunClosures(ts->name, closures);
 259   }
 260 }
 261
 262 void Executor::Enqueue(grpc_closure* closure, grpc_error* error,
 263                        bool is_short) {
 264   bool retry_push;
 265   if (is_short) {
 266     GRPC_STATS_INC_EXECUTOR_SCHEDULED_SHORT_ITEMS();
 267   } else {
 268     GRPC_STATS_INC_EXECUTOR_SCHEDULED_LONG_ITEMS();
 269   }
 270
 271   do {
 272     retry_push = false;
 273     size_t cur_thread_count =
 274         static_cast<size_t>(gpr_atm_acq_load(&num_threads_));
 275
 276     // If the number of threads is zero(i.e either the executor is not threaded
 277     // or already shutdown), then queue the closure on the exec context itself
 278     if (cur_thread_count == 0) {
 279 #ifndef NDEBUG
 280       EXECUTOR_TRACE("(%s) schedule %p (created %s:%d) inline", name_, closure,
 281                      closure->file_created, closure->line_created);
 282 #else
 283       EXECUTOR_TRACE("(%s) schedule %p inline", name_, closure);
 284 #endif
 285       grpc_closure_list_append(grpc_core::ExecCtx::Get()->closure_list(),
 286                                closure, error);
 287       return;
 288     }
 289
 290     if (grpc_iomgr_add_closure_to_background_poller(closure, error)) {
 291       return;
 292     }
 293
 294     ThreadState* ts = (ThreadState*)gpr_tls_get(&g_this_thread_state);
 295     if (ts == nullptr) {
 296       ts = &thd_state_[GPR_HASH_POINTER(grpc_core::ExecCtx::Get(),
 297                                         cur_thread_count)];
 298     } else {
 299       GRPC_STATS_INC_EXECUTOR_SCHEDULED_TO_SELF();
 300     }
 301
 302     ThreadState* orig_ts = ts;
 303     bool try_new_thread = false;
 304
 305     for (;;) {
 306 #ifndef NDEBUG
 307       EXECUTOR_TRACE(
 308           "(%s) try to schedule %p (%s) (created %s:%d) to thread "
 309           "%" PRIdPTR,
 310           name_, closure, is_short ? "short" : "long", closure->file_created,
 311           closure->line_created, ts->id);
 312 #else
 313       EXECUTOR_TRACE("(%s) try to schedule %p (%s) to thread %" PRIdPTR, name_,
 314                      closure, is_short ? "short" : "long", ts->id);
 315 #endif
 316
 317       gpr_mu_lock(&ts->mu);
 318       if (ts->queued_long_job) {
 319         // if there's a long job queued, we never queue anything else to this
 320         // queue (since long jobs can take 'infinite' time and we need to
 321         // guarantee no starvation). Spin through queues and try again
 322         gpr_mu_unlock(&ts->mu);
 323         size_t idx = ts->id;
 324         ts = &thd_state_[(idx + 1) % cur_thread_count];
 325         if (ts == orig_ts) {
 326           // We cycled through all the threads. Retry enqueue again by creating
 327           // a new thread
 328           //
 329           // TODO (sreek): There is a potential issue here. We are
 330           // unconditionally setting try_new_thread to true here. What if the
 331           // executor is shutdown OR if cur_thread_count is already equal to
 332           // max_threads ?
 333           // (Fortunately, this is not an issue yet (as of july 2018) because
 334           // there is only one instance of long job in gRPC and hence we will
 335           // not hit this code path)
 336           retry_push = true;
 337           try_new_thread = true;
 338           break;
 339         }
 340
 341         continue;  // Try the next thread-state
 342       }
 343
 344       // == Found the thread state (i.e thread) to enqueue this closure! ==
 345
 346       // Also, if this thread has been waiting for closures, wake it up.
 347       // - If grpc_closure_list_empty() is true and the Executor is not
 348       //   shutdown, it means that the thread must be waiting in ThreadMain()
 349       // - Note that gpr_cv_signal() won't immediately wakeup the thread. That
 350       //   happens after we release the mutex &ts->mu a few lines below
 351       if (grpc_closure_list_empty(ts->elems) && !ts->shutdown) {
 352         GRPC_STATS_INC_EXECUTOR_WAKEUP_INITIATED();
 353         gpr_cv_signal(&ts->cv);
 354       }
 355
 356       grpc_closure_list_append(&ts->elems, closure, error);
 357
 358       // If we already queued more than MAX_DEPTH number of closures on this
 359       // thread, use this as a hint to create more threads
 360       ts->depth++;
 361       try_new_thread = ts->depth > MAX_DEPTH &&
 362                        cur_thread_count < max_threads_ && !ts->shutdown;
 363
 364       ts->queued_long_job = !is_short;
 365
 366       gpr_mu_unlock(&ts->mu);
 367       break;
 368     }
 369
 370     if (try_new_thread && gpr_spinlock_trylock(&adding_thread_lock_)) {
 371       cur_thread_count = static_cast<size_t>(gpr_atm_acq_load(&num_threads_));
 372       if (cur_thread_count < max_threads_) {
 373         // Increment num_threads (safe to do a store instead of a cas because we
 374         // always increment num_threads under the 'adding_thread_lock')
 375         gpr_atm_rel_store(&num_threads_, cur_thread_count + 1);
 376
 377         thd_state_[cur_thread_count].thd = grpc_core::Thread(
 378             name_, &Executor::ThreadMain, &thd_state_[cur_thread_count]);
 379         thd_state_[cur_thread_count].thd.Start();
 380       }
 381       gpr_spinlock_unlock(&adding_thread_lock_);
 382     }
 383
 384     if (retry_push) {
 385       GRPC_STATS_INC_EXECUTOR_PUSH_RETRIES();
 386     }
 387   } while (retry_push);
 388 }
 389
 390 // Executor::InitAll() and Executor::ShutdownAll() functions are called in the
 391 // the grpc_init() and grpc_shutdown() code paths which are protected by a
 392 // global mutex. So it is okay to assume that these functions are thread-safe
 393 void Executor::InitAll() {
 394   EXECUTOR_TRACE0("Executor::InitAll() enter");
 395
 396   // Return if Executor::InitAll() is already called earlier
 397   if (executors[static_cast<size_t>(ExecutorType::DEFAULT)] != nullptr) {
 398     GPR_ASSERT(executors[static_cast<size_t>(ExecutorType::RESOLVER)] !=
 399                nullptr);
 400     return;
 401   }
 402
 403   executors[static_cast<size_t>(ExecutorType::DEFAULT)] =
 404       grpc_core::New<Executor>("default-executor");
 405   executors[static_cast<size_t>(ExecutorType::RESOLVER)] =
 406       grpc_core::New<Executor>("resolver-executor");
 407
 408   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Init();
 409   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Init();
 410
 411   EXECUTOR_TRACE0("Executor::InitAll() done");
 412 }
 413
 414 grpc_closure_scheduler* Executor::Scheduler(ExecutorType executor_type,
 415                                             ExecutorJobType job_type) {
 416   return &schedulers_[static_cast<size_t>(executor_type)]
 417                      [static_cast<size_t>(job_type)];
 418 }
 419
 420 grpc_closure_scheduler* Executor::Scheduler(ExecutorJobType job_type) {
 421   return Executor::Scheduler(ExecutorType::DEFAULT, job_type);
 422 }
 423
 424 void Executor::ShutdownAll() {
 425   EXECUTOR_TRACE0("Executor::ShutdownAll() enter");
 426
 427   // Return if Executor:SshutdownAll() is already called earlier
 428   if (executors[static_cast<size_t>(ExecutorType::DEFAULT)] == nullptr) {
 429     GPR_ASSERT(executors[static_cast<size_t>(ExecutorType::RESOLVER)] ==
 430                nullptr);
 431     return;
 432   }
 433
 434   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Shutdown();
 435   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Shutdown();
 436
 437   // Delete the executor objects.
 438   //
 439   // NOTE: It is important to call Shutdown() on all executors first before
 440   // calling Delete() because it is possible for one executor (that is not
 441   // shutdown yet) to call Enqueue() on a different executor which is already
 442   // shutdown. This is legal and in such cases, the Enqueue() operation
 443   // effectively "fails" and enqueues that closure on the calling thread's
 444   // exec_ctx.
 445   //
 446   // By ensuring that all executors are shutdown first, we are also ensuring
 447   // that no thread is active across all executors.
 448
 449   grpc_core::Delete<Executor>(
 450       executors[static_cast<size_t>(ExecutorType::DEFAULT)]);
 451   grpc_core::Delete<Executor>(
 452       executors[static_cast<size_t>(ExecutorType::RESOLVER)]);
 453   executors[static_cast<size_t>(ExecutorType::DEFAULT)] = nullptr;
 454   executors[static_cast<size_t>(ExecutorType::RESOLVER)] = nullptr;
 455
 456   EXECUTOR_TRACE0("Executor::ShutdownAll() done");
 457 }
 458
 459 bool Executor::IsThreaded(ExecutorType executor_type) {
 460   GPR_ASSERT(executor_type < ExecutorType::NUM_EXECUTORS);
 461   return executors[static_cast<size_t>(executor_type)]->IsThreaded();
 462 }
 463
 464 bool Executor::IsThreadedDefault() {
 465   return Executor::IsThreaded(ExecutorType::DEFAULT);
 466 }
 467
 468 void Executor::SetThreadingAll(bool enable) {
 469   EXECUTOR_TRACE("Executor::SetThreadingAll(%d) called", enable);
 470   for (size_t i = 0; i < static_cast<size_t>(ExecutorType::NUM_EXECUTORS);
 471        i++) {
 472     executors[i]->SetThreading(enable);
 473   }
 474 }
 475
 476 void Executor::SetThreadingDefault(bool enable) {
 477   EXECUTOR_TRACE("Executor::SetThreadingDefault(%d) called", enable);
 478   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->SetThreading(enable);
 479 }
 480
 481 }  // namespace grpc_core