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   do {                                                    \
  41     if (GRPC_TRACE_FLAG_ENABLED(executor_trace)) {        \
  42       gpr_log(GPR_INFO, "EXECUTOR " format, __VA_ARGS__); \
  43     }                                                     \
  44   } while (0)
  45
  46 #define EXECUTOR_TRACE0(str)                       \
  47   do {                                             \
  48     if (GRPC_TRACE_FLAG_ENABLED(executor_trace)) { \
  49       gpr_log(GPR_INFO, "EXECUTOR " str);          \
  50     }                                              \
  51   } while (0)
  52
  53 namespace grpc_core {
  54 namespace {
  55
  56 GPR_TLS_DECL(g_this_thread_state);
  57
  58 Executor* executors[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)];
  59
  60 void default_enqueue_short(grpc_closure* closure, grpc_error* error) {
  61   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Enqueue(
  62       closure, error, true /* is_short */);
  63 }
  64
  65 void default_enqueue_long(grpc_closure* closure, grpc_error* error) {
  66   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Enqueue(
  67       closure, error, false /* is_short */);
  68 }
  69
  70 void resolver_enqueue_short(grpc_closure* closure, grpc_error* error) {
  71   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Enqueue(
  72       closure, error, true /* is_short */);
  73 }
  74
  75 void resolver_enqueue_long(grpc_closure* closure, grpc_error* error) {
  76   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Enqueue(
  77       closure, error, false /* is_short */);
  78 }
  79
  80 const grpc_closure_scheduler_vtable
  81     vtables_[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)]
  82             [static_cast<size_t>(ExecutorJobType::NUM_JOB_TYPES)] = {
  83                 {{&default_enqueue_short, &default_enqueue_short,
  84                   "def-ex-short"},
  85                  {&default_enqueue_long, &default_enqueue_long, "def-ex-long"}},
  86                 {{&resolver_enqueue_short, &resolver_enqueue_short,
  87                   "res-ex-short"},
  88                  {&resolver_enqueue_long, &resolver_enqueue_long,
  89                   "res-ex-long"}}};
  90
  91 grpc_closure_scheduler
  92     schedulers_[static_cast<size_t>(ExecutorType::NUM_EXECUTORS)]
  93                [static_cast<size_t>(ExecutorJobType::NUM_JOB_TYPES)] = {
  94                    {{&vtables_[static_cast<size_t>(ExecutorType::DEFAULT)]
  95                               [static_cast<size_t>(ExecutorJobType::SHORT)]},
  96                     {&vtables_[static_cast<size_t>(ExecutorType::DEFAULT)]
  97                               [static_cast<size_t>(ExecutorJobType::LONG)]}},
  98                    {{&vtables_[static_cast<size_t>(ExecutorType::RESOLVER)]
  99                               [static_cast<size_t>(ExecutorJobType::SHORT)]},
 100                     {&vtables_[static_cast<size_t>(ExecutorType::RESOLVER)]
 101                               [static_cast<size_t>(ExecutorJobType::LONG)]}}};
 102
 103 }  // namespace
 104
 105 TraceFlag executor_trace(false, "executor");
 106
 107 Executor::Executor(const char* name) : name_(name) {
 108   adding_thread_lock_ = GPR_SPINLOCK_STATIC_INITIALIZER;
 109   gpr_atm_rel_store(&num_threads_, 0);
 110   max_threads_ = GPR_MAX(1, 2 * gpr_cpu_num_cores());
 111 }
 112
 113 void Executor::Init() { SetThreading(true); }
 114
 115 size_t Executor::RunClosures(const char* executor_name,
 116                              grpc_closure_list list) {
 117   size_t n = 0;
 118
 119   // In the executor, the ExecCtx for the thread is declared in the executor
 120   // thread itself, but this is the point where we could start seeing
 121   // application-level callbacks. No need to create a new ExecCtx, though,
 122   // since there already is one and it is flushed (but not destructed) in this
 123   // function itself.
 124   grpc_core::ApplicationCallbackExecCtx callback_exec_ctx(
 125       GRPC_APP_CALLBACK_EXEC_CTX_FLAG_IS_INTERNAL_THREAD);
 126
 127   grpc_closure* c = list.head;
 128   while (c != nullptr) {
 129     grpc_closure* next = c->next_data.next;
 130     grpc_error* error = c->error_data.error;
 131 #ifndef NDEBUG
 132     EXECUTOR_TRACE("(%s) run %p [created by %s:%d]", executor_name, c,
 133                    c->file_created, c->line_created);
 134     c->scheduled = false;
 135 #else
 136     EXECUTOR_TRACE("(%s) run %p", executor_name, c);
 137 #endif
 138     c->cb(c->cb_arg, error);
 139     GRPC_ERROR_UNREF(error);
 140     c = next;
 141     n++;
 142     grpc_core::ExecCtx::Get()->Flush();
 143   }
 144
 145   return n;
 146 }
 147
 148 bool Executor::IsThreaded() const {
 149   return gpr_atm_acq_load(&num_threads_) > 0;
 150 }
 151
 152 void Executor::SetThreading(bool threading) {
 153   gpr_atm curr_num_threads = gpr_atm_acq_load(&num_threads_);
 154   EXECUTOR_TRACE("(%s) SetThreading(%d) begin", name_, threading);
 155
 156   if (threading) {
 157     if (curr_num_threads > 0) {
 158       EXECUTOR_TRACE("(%s) SetThreading(true). curr_num_threads == 0", name_);
 159       return;
 160     }
 161
 162     GPR_ASSERT(num_threads_ == 0);
 163     gpr_atm_rel_store(&num_threads_, 1);
 164     gpr_tls_init(&g_this_thread_state);
 165     thd_state_ = static_cast<ThreadState*>(
 166         gpr_zalloc(sizeof(ThreadState) * max_threads_));
 167
 168     for (size_t i = 0; i < max_threads_; i++) {
 169       gpr_mu_init(&thd_state_[i].mu);
 170       gpr_cv_init(&thd_state_[i].cv);
 171       thd_state_[i].id = i;
 172       thd_state_[i].name = name_;
 173       thd_state_[i].thd = grpc_core::Thread();
 174       thd_state_[i].elems = GRPC_CLOSURE_LIST_INIT;
 175     }
 176
 177     thd_state_[0].thd =
 178         grpc_core::Thread(name_, &Executor::ThreadMain, &thd_state_[0]);
 179     thd_state_[0].thd.Start();
 180   } else {  // !threading
 181     if (curr_num_threads == 0) {
 182       EXECUTOR_TRACE("(%s) SetThreading(false). curr_num_threads == 0", name_);
 183       return;
 184     }
 185
 186     for (size_t i = 0; i < max_threads_; i++) {
 187       gpr_mu_lock(&thd_state_[i].mu);
 188       thd_state_[i].shutdown = true;
 189       gpr_cv_signal(&thd_state_[i].cv);
 190       gpr_mu_unlock(&thd_state_[i].mu);
 191     }
 192
 193     /* Ensure no thread is adding a new thread. Once this is past, then no
 194      * thread will try to add a new one either (since shutdown is true) */
 195     gpr_spinlock_lock(&adding_thread_lock_);
 196     gpr_spinlock_unlock(&adding_thread_lock_);
 197
 198     curr_num_threads = gpr_atm_no_barrier_load(&num_threads_);
 199     for (gpr_atm i = 0; i < curr_num_threads; i++) {
 200       thd_state_[i].thd.Join();
 201       EXECUTOR_TRACE("(%s) Thread %" PRIdPTR " of %" PRIdPTR " joined", name_,
 202                      i + 1, curr_num_threads);
 203     }
 204
 205     gpr_atm_rel_store(&num_threads_, 0);
 206     for (size_t i = 0; i < max_threads_; i++) {
 207       gpr_mu_destroy(&thd_state_[i].mu);
 208       gpr_cv_destroy(&thd_state_[i].cv);
 209       RunClosures(thd_state_[i].name, thd_state_[i].elems);
 210     }
 211
 212     gpr_free(thd_state_);
 213     gpr_tls_destroy(&g_this_thread_state);
 214
 215     // grpc_iomgr_shutdown_background_closure() will close all the registered
 216     // fds in the background poller, and wait for all pending closures to
 217     // finish. Thus, never call Executor::SetThreading(false) in the middle of
 218     // an application.
 219     // TODO(guantaol): create another method to finish all the pending closures
 220     // registered in the background poller by grpc_core::Executor.
 221     grpc_iomgr_shutdown_background_closure();
 222   }
 223
 224   EXECUTOR_TRACE("(%s) SetThreading(%d) done", name_, threading);
 225 }
 226
 227 void Executor::Shutdown() { SetThreading(false); }
 228
 229 void Executor::ThreadMain(void* arg) {
 230   ThreadState* ts = static_cast<ThreadState*>(arg);
 231   gpr_tls_set(&g_this_thread_state, reinterpret_cast<intptr_t>(ts));
 232
 233   grpc_core::ExecCtx exec_ctx(GRPC_EXEC_CTX_FLAG_IS_INTERNAL_THREAD);
 234
 235   size_t subtract_depth = 0;
 236   for (;;) {
 237     EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: step (sub_depth=%" PRIdPTR ")",
 238                    ts->name, ts->id, subtract_depth);
 239
 240     gpr_mu_lock(&ts->mu);
 241     ts->depth -= subtract_depth;
 242     // Wait for closures to be enqueued or for the executor to be shutdown
 243     while (grpc_closure_list_empty(ts->elems) && !ts->shutdown) {
 244       ts->queued_long_job = false;
 245       gpr_cv_wait(&ts->cv, &ts->mu, gpr_inf_future(GPR_CLOCK_MONOTONIC));
 246     }
 247
 248     if (ts->shutdown) {
 249       EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: shutdown", ts->name, ts->id);
 250       gpr_mu_unlock(&ts->mu);
 251       break;
 252     }
 253
 254     GRPC_STATS_INC_EXECUTOR_QUEUE_DRAINED();
 255     grpc_closure_list closures = ts->elems;
 256     ts->elems = GRPC_CLOSURE_LIST_INIT;
 257     gpr_mu_unlock(&ts->mu);
 258
 259     EXECUTOR_TRACE("(%s) [%" PRIdPTR "]: execute", ts->name, ts->id);
 260
 261     grpc_core::ExecCtx::Get()->InvalidateNow();
 262     subtract_depth = RunClosures(ts->name, closures);
 263   }
 264 }
 265
 266 void Executor::Enqueue(grpc_closure* closure, grpc_error* error,
 267                        bool is_short) {
 268   bool retry_push;
 269   if (is_short) {
 270     GRPC_STATS_INC_EXECUTOR_SCHEDULED_SHORT_ITEMS();
 271   } else {
 272     GRPC_STATS_INC_EXECUTOR_SCHEDULED_LONG_ITEMS();
 273   }
 274
 275   do {
 276     retry_push = false;
 277     size_t cur_thread_count =
 278         static_cast<size_t>(gpr_atm_acq_load(&num_threads_));
 279
 280     // If the number of threads is zero(i.e either the executor is not threaded
 281     // or already shutdown), then queue the closure on the exec context itself
 282     if (cur_thread_count == 0) {
 283 #ifndef NDEBUG
 284       EXECUTOR_TRACE("(%s) schedule %p (created %s:%d) inline", name_, closure,
 285                      closure->file_created, closure->line_created);
 286 #else
 287       EXECUTOR_TRACE("(%s) schedule %p inline", name_, closure);
 288 #endif
 289       grpc_closure_list_append(grpc_core::ExecCtx::Get()->closure_list(),
 290                                closure, error);
 291       return;
 292     }
 293
 294     if (grpc_iomgr_add_closure_to_background_poller(closure, error)) {
 295       return;
 296     }
 297
 298     ThreadState* ts = (ThreadState*)gpr_tls_get(&g_this_thread_state);
 299     if (ts == nullptr) {
 300       ts = &thd_state_[GPR_HASH_POINTER(grpc_core::ExecCtx::Get(),
 301                                         cur_thread_count)];
 302     } else {
 303       GRPC_STATS_INC_EXECUTOR_SCHEDULED_TO_SELF();
 304     }
 305
 306     ThreadState* orig_ts = ts;
 307     bool try_new_thread = false;
 308
 309     for (;;) {
 310 #ifndef NDEBUG
 311       EXECUTOR_TRACE(
 312           "(%s) try to schedule %p (%s) (created %s:%d) to thread "
 313           "%" PRIdPTR,
 314           name_, closure, is_short ? "short" : "long", closure->file_created,
 315           closure->line_created, ts->id);
 316 #else
 317       EXECUTOR_TRACE("(%s) try to schedule %p (%s) to thread %" PRIdPTR, name_,
 318                      closure, is_short ? "short" : "long", ts->id);
 319 #endif
 320
 321       gpr_mu_lock(&ts->mu);
 322       if (ts->queued_long_job) {
 323         // if there's a long job queued, we never queue anything else to this
 324         // queue (since long jobs can take 'infinite' time and we need to
 325         // guarantee no starvation). Spin through queues and try again
 326         gpr_mu_unlock(&ts->mu);
 327         size_t idx = ts->id;
 328         ts = &thd_state_[(idx + 1) % cur_thread_count];
 329         if (ts == orig_ts) {
 330           // We cycled through all the threads. Retry enqueue again by creating
 331           // a new thread
 332           //
 333           // TODO (sreek): There is a potential issue here. We are
 334           // unconditionally setting try_new_thread to true here. What if the
 335           // executor is shutdown OR if cur_thread_count is already equal to
 336           // max_threads ?
 337           // (Fortunately, this is not an issue yet (as of july 2018) because
 338           // there is only one instance of long job in gRPC and hence we will
 339           // not hit this code path)
 340           retry_push = true;
 341           try_new_thread = true;
 342           break;
 343         }
 344
 345         continue;  // Try the next thread-state
 346       }
 347
 348       // == Found the thread state (i.e thread) to enqueue this closure! ==
 349
 350       // Also, if this thread has been waiting for closures, wake it up.
 351       // - If grpc_closure_list_empty() is true and the Executor is not
 352       //   shutdown, it means that the thread must be waiting in ThreadMain()
 353       // - Note that gpr_cv_signal() won't immediately wakeup the thread. That
 354       //   happens after we release the mutex &ts->mu a few lines below
 355       if (grpc_closure_list_empty(ts->elems) && !ts->shutdown) {
 356         GRPC_STATS_INC_EXECUTOR_WAKEUP_INITIATED();
 357         gpr_cv_signal(&ts->cv);
 358       }
 359
 360       grpc_closure_list_append(&ts->elems, closure, error);
 361
 362       // If we already queued more than MAX_DEPTH number of closures on this
 363       // thread, use this as a hint to create more threads
 364       ts->depth++;
 365       try_new_thread = ts->depth > MAX_DEPTH &&
 366                        cur_thread_count < max_threads_ && !ts->shutdown;
 367
 368       ts->queued_long_job = !is_short;
 369
 370       gpr_mu_unlock(&ts->mu);
 371       break;
 372     }
 373
 374     if (try_new_thread && gpr_spinlock_trylock(&adding_thread_lock_)) {
 375       cur_thread_count = static_cast<size_t>(gpr_atm_acq_load(&num_threads_));
 376       if (cur_thread_count < max_threads_) {
 377         // Increment num_threads (safe to do a store instead of a cas because we
 378         // always increment num_threads under the 'adding_thread_lock')
 379         gpr_atm_rel_store(&num_threads_, cur_thread_count + 1);
 380
 381         thd_state_[cur_thread_count].thd = grpc_core::Thread(
 382             name_, &Executor::ThreadMain, &thd_state_[cur_thread_count]);
 383         thd_state_[cur_thread_count].thd.Start();
 384       }
 385       gpr_spinlock_unlock(&adding_thread_lock_);
 386     }
 387
 388     if (retry_push) {
 389       GRPC_STATS_INC_EXECUTOR_PUSH_RETRIES();
 390     }
 391   } while (retry_push);
 392 }
 393
 394 // Executor::InitAll() and Executor::ShutdownAll() functions are called in the
 395 // the grpc_init() and grpc_shutdown() code paths which are protected by a
 396 // global mutex. So it is okay to assume that these functions are thread-safe
 397 void Executor::InitAll() {
 398   EXECUTOR_TRACE0("Executor::InitAll() enter");
 399
 400   // Return if Executor::InitAll() is already called earlier
 401   if (executors[static_cast<size_t>(ExecutorType::DEFAULT)] != nullptr) {
 402     GPR_ASSERT(executors[static_cast<size_t>(ExecutorType::RESOLVER)] !=
 403                nullptr);
 404     return;
 405   }
 406
 407   executors[static_cast<size_t>(ExecutorType::DEFAULT)] =
 408       grpc_core::New<Executor>("default-executor");
 409   executors[static_cast<size_t>(ExecutorType::RESOLVER)] =
 410       grpc_core::New<Executor>("resolver-executor");
 411
 412   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Init();
 413   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Init();
 414
 415   EXECUTOR_TRACE0("Executor::InitAll() done");
 416 }
 417
 418 grpc_closure_scheduler* Executor::Scheduler(ExecutorType executor_type,
 419                                             ExecutorJobType job_type) {
 420   return &schedulers_[static_cast<size_t>(executor_type)]
 421                      [static_cast<size_t>(job_type)];
 422 }
 423
 424 grpc_closure_scheduler* Executor::Scheduler(ExecutorJobType job_type) {
 425   return Executor::Scheduler(ExecutorType::DEFAULT, job_type);
 426 }
 427
 428 void Executor::ShutdownAll() {
 429   EXECUTOR_TRACE0("Executor::ShutdownAll() enter");
 430
 431   // Return if Executor:SshutdownAll() is already called earlier
 432   if (executors[static_cast<size_t>(ExecutorType::DEFAULT)] == nullptr) {
 433     GPR_ASSERT(executors[static_cast<size_t>(ExecutorType::RESOLVER)] ==
 434                nullptr);
 435     return;
 436   }
 437
 438   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->Shutdown();
 439   executors[static_cast<size_t>(ExecutorType::RESOLVER)]->Shutdown();
 440
 441   // Delete the executor objects.
 442   //
 443   // NOTE: It is important to call Shutdown() on all executors first before
 444   // calling Delete() because it is possible for one executor (that is not
 445   // shutdown yet) to call Enqueue() on a different executor which is already
 446   // shutdown. This is legal and in such cases, the Enqueue() operation
 447   // effectively "fails" and enqueues that closure on the calling thread's
 448   // exec_ctx.
 449   //
 450   // By ensuring that all executors are shutdown first, we are also ensuring
 451   // that no thread is active across all executors.
 452
 453   grpc_core::Delete<Executor>(
 454       executors[static_cast<size_t>(ExecutorType::DEFAULT)]);
 455   grpc_core::Delete<Executor>(
 456       executors[static_cast<size_t>(ExecutorType::RESOLVER)]);
 457   executors[static_cast<size_t>(ExecutorType::DEFAULT)] = nullptr;
 458   executors[static_cast<size_t>(ExecutorType::RESOLVER)] = nullptr;
 459
 460   EXECUTOR_TRACE0("Executor::ShutdownAll() done");
 461 }
 462
 463 bool Executor::IsThreaded(ExecutorType executor_type) {
 464   GPR_ASSERT(executor_type < ExecutorType::NUM_EXECUTORS);
 465   return executors[static_cast<size_t>(executor_type)]->IsThreaded();
 466 }
 467
 468 bool Executor::IsThreadedDefault() {
 469   return Executor::IsThreaded(ExecutorType::DEFAULT);
 470 }
 471
 472 void Executor::SetThreadingAll(bool enable) {
 473   EXECUTOR_TRACE("Executor::SetThreadingAll(%d) called", enable);
 474   for (size_t i = 0; i < static_cast<size_t>(ExecutorType::NUM_EXECUTORS);
 475        i++) {
 476     executors[i]->SetThreading(enable);
 477   }
 478 }
 479
 480 void Executor::SetThreadingDefault(bool enable) {
 481   EXECUTOR_TRACE("Executor::SetThreadingDefault(%d) called", enable);
 482   executors[static_cast<size_t>(ExecutorType::DEFAULT)]->SetThreading(enable);
 483 }
 484
 485 }  // namespace grpc_core