001/* 002 * Copyright (C) 2012 The Guava Authors 003 * 004 * Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except 005 * in compliance with the License. You may obtain a copy of the License at 006 * 007 * http://www.apache.org/licenses/LICENSE-2.0 008 * 009 * Unless required by applicable law or agreed to in writing, software distributed under the License 010 * is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express 011 * or implied. See the License for the specific language governing permissions and limitations under 012 * the License. 013 */ 014package com.google.common.util.concurrent; 015 016import static com.google.common.base.Preconditions.checkArgument; 017import static com.google.common.base.Preconditions.checkNotNull; 018import static com.google.common.base.Preconditions.checkState; 019import static com.google.common.base.Predicates.equalTo; 020import static com.google.common.base.Predicates.in; 021import static com.google.common.base.Predicates.instanceOf; 022import static com.google.common.base.Predicates.not; 023import static com.google.common.util.concurrent.MoreExecutors.directExecutor; 024import static com.google.common.util.concurrent.Service.State.FAILED; 025import static com.google.common.util.concurrent.Service.State.NEW; 026import static com.google.common.util.concurrent.Service.State.RUNNING; 027import static com.google.common.util.concurrent.Service.State.STARTING; 028import static com.google.common.util.concurrent.Service.State.STOPPING; 029import static com.google.common.util.concurrent.Service.State.TERMINATED; 030import static java.util.concurrent.TimeUnit.MILLISECONDS; 031 032import com.google.common.annotations.Beta; 033import com.google.common.annotations.GwtIncompatible; 034import com.google.common.base.Function; 035import com.google.common.base.MoreObjects; 036import com.google.common.base.Stopwatch; 037import com.google.common.collect.Collections2; 038import com.google.common.collect.ImmutableCollection; 039import com.google.common.collect.ImmutableList; 040import com.google.common.collect.ImmutableMap; 041import com.google.common.collect.ImmutableMultimap; 042import com.google.common.collect.ImmutableSet; 043import com.google.common.collect.ImmutableSetMultimap; 044import com.google.common.collect.Lists; 045import com.google.common.collect.Maps; 046import com.google.common.collect.MultimapBuilder; 047import com.google.common.collect.Multimaps; 048import com.google.common.collect.Multiset; 049import com.google.common.collect.Ordering; 050import com.google.common.collect.SetMultimap; 051import com.google.common.util.concurrent.Service.State; 052import com.google.errorprone.annotations.CanIgnoreReturnValue; 053import com.google.j2objc.annotations.WeakOuter; 054import java.lang.ref.WeakReference; 055import java.util.Collections; 056import java.util.EnumSet; 057import java.util.List; 058import java.util.Map; 059import java.util.Map.Entry; 060import java.util.concurrent.Executor; 061import java.util.concurrent.TimeUnit; 062import java.util.concurrent.TimeoutException; 063import java.util.logging.Level; 064import java.util.logging.Logger; 065import javax.annotation.concurrent.GuardedBy; 066 067/** 068 * A manager for monitoring and controlling a set of {@linkplain Service services}. This class 069 * provides methods for {@linkplain #startAsync() starting}, {@linkplain #stopAsync() stopping} and 070 * {@linkplain #servicesByState inspecting} a collection of {@linkplain Service services}. 071 * Additionally, users can monitor state transitions with the {@linkplain Listener listener} 072 * mechanism. 073 * 074 * <p>While it is recommended that service lifecycles be managed via this class, state transitions 075 * initiated via other mechanisms do not impact the correctness of its methods. For example, if the 076 * services are started by some mechanism besides {@link #startAsync}, the listeners will be invoked 077 * when appropriate and {@link #awaitHealthy} will still work as expected. 078 * 079 * <p>Here is a simple example of how to use a {@code ServiceManager} to start a server. 080 * <pre> {@code 081 * class Server { 082 * public static void main(String[] args) { 083 * Set<Service> services = ...; 084 * ServiceManager manager = new ServiceManager(services); 085 * manager.addListener(new Listener() { 086 * public void stopped() {} 087 * public void healthy() { 088 * // Services have been initialized and are healthy, start accepting requests... 089 * } 090 * public void failure(Service service) { 091 * // Something failed, at this point we could log it, notify a load balancer, or take 092 * // some other action. For now we will just exit. 093 * System.exit(1); 094 * } 095 * }, 096 * MoreExecutors.directExecutor()); 097 * 098 * Runtime.getRuntime().addShutdownHook(new Thread() { 099 * public void run() { 100 * // Give the services 5 seconds to stop to ensure that we are responsive to shutdown 101 * // requests. 102 * try { 103 * manager.stopAsync().awaitStopped(5, TimeUnit.SECONDS); 104 * } catch (TimeoutException timeout) { 105 * // stopping timed out 106 * } 107 * } 108 * }); 109 * manager.startAsync(); // start all the services asynchronously 110 * } 111 * }}</pre> 112 * 113 * <p>This class uses the ServiceManager's methods to start all of its services, to respond to 114 * service failure and to ensure that when the JVM is shutting down all the services are stopped. 115 * 116 * @author Luke Sandberg 117 * @since 14.0 118 */ 119@Beta 120@GwtIncompatible 121public final class ServiceManager { 122 private static final Logger logger = Logger.getLogger(ServiceManager.class.getName()); 123 private static final ListenerCallQueue.Event<Listener> HEALTHY_EVENT = 124 new ListenerCallQueue.Event<Listener>() { 125 @Override 126 public void call(Listener listener) { 127 listener.healthy(); 128 } 129 130 @Override 131 public String toString() { 132 return "healthy()"; 133 } 134 }; 135 private static final ListenerCallQueue.Event<Listener> STOPPED_EVENT = 136 new ListenerCallQueue.Event<Listener>() { 137 @Override 138 public void call(Listener listener) { 139 listener.stopped(); 140 } 141 142 @Override 143 public String toString() { 144 return "stopped()"; 145 } 146 }; 147 148 /** 149 * A listener for the aggregate state changes of the services that are under management. Users 150 * that need to listen to more fine-grained events (such as when each particular 151 * {@linkplain Service service} starts, or terminates), should attach {@linkplain Service.Listener 152 * service listeners} to each individual service. 153 * 154 * @author Luke Sandberg 155 * @since 15.0 (present as an interface in 14.0) 156 */ 157 @Beta // Should come out of Beta when ServiceManager does 158 public abstract static class Listener { 159 /** 160 * Called when the service initially becomes healthy. 161 * 162 * <p>This will be called at most once after all the services have entered the 163 * {@linkplain State#RUNNING running} state. If any services fail during start up or 164 * {@linkplain State#FAILED fail}/{@linkplain State#TERMINATED terminate} before all other 165 * services have started {@linkplain State#RUNNING running} then this method will not be called. 166 */ 167 public void healthy() {} 168 169 /** 170 * Called when the all of the component services have reached a terminal state, either 171 * {@linkplain State#TERMINATED terminated} or {@linkplain State#FAILED failed}. 172 */ 173 public void stopped() {} 174 175 /** 176 * Called when a component service has {@linkplain State#FAILED failed}. 177 * 178 * @param service The service that failed. 179 */ 180 public void failure(Service service) {} 181 } 182 183 /** 184 * An encapsulation of all of the state that is accessed by the {@linkplain ServiceListener 185 * service listeners}. This is extracted into its own object so that {@link ServiceListener} could 186 * be made {@code static} and its instances can be safely constructed and added in the 187 * {@link ServiceManager} constructor without having to close over the partially constructed 188 * {@link ServiceManager} instance (i.e. avoid leaking a pointer to {@code this}). 189 */ 190 private final ServiceManagerState state; 191 private final ImmutableList<Service> services; 192 193 /** 194 * Constructs a new instance for managing the given services. 195 * 196 * @param services The services to manage 197 * 198 * @throws IllegalArgumentException if not all services are {@linkplain State#NEW new} or if there 199 * are any duplicate services. 200 */ 201 public ServiceManager(Iterable<? extends Service> services) { 202 ImmutableList<Service> copy = ImmutableList.copyOf(services); 203 if (copy.isEmpty()) { 204 // Having no services causes the manager to behave strangely. Notably, listeners are never 205 // fired. To avoid this we substitute a placeholder service. 206 logger.log( 207 Level.WARNING, 208 "ServiceManager configured with no services. Is your application configured properly?", 209 new EmptyServiceManagerWarning()); 210 copy = ImmutableList.<Service>of(new NoOpService()); 211 } 212 this.state = new ServiceManagerState(copy); 213 this.services = copy; 214 WeakReference<ServiceManagerState> stateReference = new WeakReference<>(state); 215 for (Service service : copy) { 216 service.addListener(new ServiceListener(service, stateReference), directExecutor()); 217 // We check the state after adding the listener as a way to ensure that our listener was added 218 // to a NEW service. 219 checkArgument(service.state() == NEW, "Can only manage NEW services, %s", service); 220 } 221 // We have installed all of our listeners and after this point any state transition should be 222 // correct. 223 this.state.markReady(); 224 } 225 226 /** 227 * Registers a {@link Listener} to be {@linkplain Executor#execute executed} on the given 228 * executor. The listener will not have previous state changes replayed, so it is suggested that 229 * listeners are added before any of the managed services are {@linkplain Service#startAsync 230 * started}. 231 * 232 * <p>{@code addListener} guarantees execution ordering across calls to a given listener but not 233 * across calls to multiple listeners. Specifically, a given listener will have its callbacks 234 * invoked in the same order as the underlying service enters those states. Additionally, at most 235 * one of the listener's callbacks will execute at once. However, multiple listeners' callbacks 236 * may execute concurrently, and listeners may execute in an order different from the one in which 237 * they were registered. 238 * 239 * <p>RuntimeExceptions thrown by a listener will be caught and logged. Any exception thrown 240 * during {@code Executor.execute} (e.g., a {@code RejectedExecutionException}) will be caught and 241 * logged. 242 * 243 * <p>For fast, lightweight listeners that would be safe to execute in any thread, consider 244 * calling {@link #addListener(Listener)}. 245 * 246 * @param listener the listener to run when the manager changes state 247 * @param executor the executor in which the listeners callback methods will be run. 248 */ 249 public void addListener(Listener listener, Executor executor) { 250 state.addListener(listener, executor); 251 } 252 253 /** 254 * Registers a {@link Listener} to be run when this {@link ServiceManager} changes state. The 255 * listener will not have previous state changes replayed, so it is suggested that listeners are 256 * added before any of the managed services are {@linkplain Service#startAsync started}. 257 * 258 * <p>{@code addListener} guarantees execution ordering across calls to a given listener but not 259 * across calls to multiple listeners. Specifically, a given listener will have its callbacks 260 * invoked in the same order as the underlying service enters those states. Additionally, at most 261 * one of the listener's callbacks will execute at once. However, multiple listeners' callbacks 262 * may execute concurrently, and listeners may execute in an order different from the one in which 263 * they were registered. 264 * 265 * <p>RuntimeExceptions thrown by a listener will be caught and logged. 266 * 267 * @param listener the listener to run when the manager changes state 268 */ 269 public void addListener(Listener listener) { 270 state.addListener(listener, directExecutor()); 271 } 272 273 /** 274 * Initiates service {@linkplain Service#startAsync startup} on all the services being managed. It 275 * is only valid to call this method if all of the services are {@linkplain State#NEW new}. 276 * 277 * @return this 278 * @throws IllegalStateException if any of the Services are not {@link State#NEW new} when the 279 * method is called. 280 */ 281 @CanIgnoreReturnValue 282 public ServiceManager startAsync() { 283 for (Service service : services) { 284 State state = service.state(); 285 checkState(state == NEW, "Service %s is %s, cannot start it.", service, state); 286 } 287 for (Service service : services) { 288 try { 289 state.tryStartTiming(service); 290 service.startAsync(); 291 } catch (IllegalStateException e) { 292 // This can happen if the service has already been started or stopped (e.g. by another 293 // service or listener). Our contract says it is safe to call this method if 294 // all services were NEW when it was called, and this has already been verified above, so we 295 // don't propagate the exception. 296 logger.log(Level.WARNING, "Unable to start Service " + service, e); 297 } 298 } 299 return this; 300 } 301 302 /** 303 * Waits for the {@link ServiceManager} to become {@linkplain #isHealthy() healthy}. The manager 304 * will become healthy after all the component services have reached the {@linkplain State#RUNNING 305 * running} state. 306 * 307 * @throws IllegalStateException if the service manager reaches a state from which it cannot 308 * become {@linkplain #isHealthy() healthy}. 309 */ 310 public void awaitHealthy() { 311 state.awaitHealthy(); 312 } 313 314 /** 315 * Waits for the {@link ServiceManager} to become {@linkplain #isHealthy() healthy} for no more 316 * than the given time. The manager will become healthy after all the component services have 317 * reached the {@linkplain State#RUNNING running} state. 318 * 319 * @param timeout the maximum time to wait 320 * @param unit the time unit of the timeout argument 321 * @throws TimeoutException if not all of the services have finished starting within the deadline 322 * @throws IllegalStateException if the service manager reaches a state from which it cannot 323 * become {@linkplain #isHealthy() healthy}. 324 */ 325 public void awaitHealthy(long timeout, TimeUnit unit) throws TimeoutException { 326 state.awaitHealthy(timeout, unit); 327 } 328 329 /** 330 * Initiates service {@linkplain Service#stopAsync shutdown} if necessary on all the services 331 * being managed. 332 * 333 * @return this 334 */ 335 @CanIgnoreReturnValue 336 public ServiceManager stopAsync() { 337 for (Service service : services) { 338 service.stopAsync(); 339 } 340 return this; 341 } 342 343 /** 344 * Waits for the all the services to reach a terminal state. After this method returns all 345 * services will either be {@linkplain Service.State#TERMINATED terminated} or 346 * {@linkplain Service.State#FAILED failed}. 347 */ 348 public void awaitStopped() { 349 state.awaitStopped(); 350 } 351 352 /** 353 * Waits for the all the services to reach a terminal state for no more than the given time. After 354 * this method returns all services will either be {@linkplain Service.State#TERMINATED 355 * terminated} or {@linkplain Service.State#FAILED failed}. 356 * 357 * @param timeout the maximum time to wait 358 * @param unit the time unit of the timeout argument 359 * @throws TimeoutException if not all of the services have stopped within the deadline 360 */ 361 public void awaitStopped(long timeout, TimeUnit unit) throws TimeoutException { 362 state.awaitStopped(timeout, unit); 363 } 364 365 /** 366 * Returns true if all services are currently in the {@linkplain State#RUNNING running} state. 367 * 368 * <p>Users who want more detailed information should use the {@link #servicesByState} method to 369 * get detailed information about which services are not running. 370 */ 371 public boolean isHealthy() { 372 for (Service service : services) { 373 if (!service.isRunning()) { 374 return false; 375 } 376 } 377 return true; 378 } 379 380 /** 381 * Provides a snapshot of the current state of all the services under management. 382 * 383 * <p>N.B. This snapshot is guaranteed to be consistent, i.e. the set of states returned will 384 * correspond to a point in time view of the services. 385 */ 386 public ImmutableMultimap<State, Service> servicesByState() { 387 return state.servicesByState(); 388 } 389 390 /** 391 * Returns the service load times. This value will only return startup times for services that 392 * have finished starting. 393 * 394 * @return Map of services and their corresponding startup time in millis, the map entries will be 395 * ordered by startup time. 396 */ 397 public ImmutableMap<Service, Long> startupTimes() { 398 return state.startupTimes(); 399 } 400 401 @Override 402 public String toString() { 403 return MoreObjects.toStringHelper(ServiceManager.class) 404 .add("services", Collections2.filter(services, not(instanceOf(NoOpService.class)))) 405 .toString(); 406 } 407 408 /** 409 * An encapsulation of all the mutable state of the {@link ServiceManager} that needs to be 410 * accessed by instances of {@link ServiceListener}. 411 */ 412 private static final class ServiceManagerState { 413 final Monitor monitor = new Monitor(); 414 415 @GuardedBy("monitor") 416 final SetMultimap<State, Service> servicesByState = 417 MultimapBuilder.enumKeys(State.class).linkedHashSetValues().build(); 418 419 @GuardedBy("monitor") 420 final Multiset<State> states = servicesByState.keys(); 421 422 @GuardedBy("monitor") 423 final Map<Service, Stopwatch> startupTimers = Maps.newIdentityHashMap(); 424 425 /** 426 * These two booleans are used to mark the state as ready to start. 427 * 428 * <p>{@link #ready}: is set by {@link #markReady} to indicate that all listeners have been 429 * correctly installed 430 * 431 * <p>{@link #transitioned}: is set by {@link #transitionService} to indicate that some 432 * transition has been performed. 433 * 434 * <p>Together, they allow us to enforce that all services have their listeners installed prior 435 * to any service performing a transition, then we can fail in the ServiceManager constructor 436 * rather than in a Service.Listener callback. 437 */ 438 @GuardedBy("monitor") 439 boolean ready; 440 441 @GuardedBy("monitor") 442 boolean transitioned; 443 444 final int numberOfServices; 445 446 /** 447 * Controls how long to wait for all the services to either become healthy or reach a state from 448 * which it is guaranteed that it can never become healthy. 449 */ 450 final Monitor.Guard awaitHealthGuard = new AwaitHealthGuard(); 451 452 @WeakOuter 453 final class AwaitHealthGuard extends Monitor.Guard { 454 AwaitHealthGuard() { 455 super(ServiceManagerState.this.monitor); 456 } 457 458 @Override 459 @GuardedBy("ServiceManagerState.this.monitor") 460 public boolean isSatisfied() { 461 // All services have started or some service has terminated/failed. 462 return states.count(RUNNING) == numberOfServices 463 || states.contains(STOPPING) 464 || states.contains(TERMINATED) 465 || states.contains(FAILED); 466 } 467 } 468 469 /** 470 * Controls how long to wait for all services to reach a terminal state. 471 */ 472 final Monitor.Guard stoppedGuard = new StoppedGuard(); 473 474 @WeakOuter 475 final class StoppedGuard extends Monitor.Guard { 476 StoppedGuard() { 477 super(ServiceManagerState.this.monitor); 478 } 479 480 @Override 481 @GuardedBy("ServiceManagerState.this.monitor") 482 public boolean isSatisfied() { 483 return states.count(TERMINATED) + states.count(FAILED) == numberOfServices; 484 } 485 } 486 487 /** The listeners to notify during a state transition. */ 488 final ListenerCallQueue<Listener> listeners = new ListenerCallQueue<>(); 489 490 /** 491 * It is implicitly assumed that all the services are NEW and that they will all remain NEW 492 * until all the Listeners are installed and {@link #markReady()} is called. It is our caller's 493 * responsibility to only call {@link #markReady()} if all services were new at the time this 494 * method was called and when all the listeners were installed. 495 */ 496 ServiceManagerState(ImmutableCollection<Service> services) { 497 this.numberOfServices = services.size(); 498 servicesByState.putAll(NEW, services); 499 } 500 501 /** 502 * Attempts to start the timer immediately prior to the service being started via 503 * {@link Service#startAsync()}. 504 */ 505 void tryStartTiming(Service service) { 506 monitor.enter(); 507 try { 508 Stopwatch stopwatch = startupTimers.get(service); 509 if (stopwatch == null) { 510 startupTimers.put(service, Stopwatch.createStarted()); 511 } 512 } finally { 513 monitor.leave(); 514 } 515 } 516 517 /** 518 * Marks the {@link State} as ready to receive transitions. Returns true if no transitions have 519 * been observed yet. 520 */ 521 void markReady() { 522 monitor.enter(); 523 try { 524 if (!transitioned) { 525 // nothing has transitioned since construction, good. 526 ready = true; 527 } else { 528 // This should be an extremely rare race condition. 529 List<Service> servicesInBadStates = Lists.newArrayList(); 530 for (Service service : servicesByState().values()) { 531 if (service.state() != NEW) { 532 servicesInBadStates.add(service); 533 } 534 } 535 throw new IllegalArgumentException( 536 "Services started transitioning asynchronously before " 537 + "the ServiceManager was constructed: " 538 + servicesInBadStates); 539 } 540 } finally { 541 monitor.leave(); 542 } 543 } 544 545 void addListener(Listener listener, Executor executor) { 546 listeners.addListener(listener, executor); 547 } 548 549 void awaitHealthy() { 550 monitor.enterWhenUninterruptibly(awaitHealthGuard); 551 try { 552 checkHealthy(); 553 } finally { 554 monitor.leave(); 555 } 556 } 557 558 void awaitHealthy(long timeout, TimeUnit unit) throws TimeoutException { 559 monitor.enter(); 560 try { 561 if (!monitor.waitForUninterruptibly(awaitHealthGuard, timeout, unit)) { 562 throw new TimeoutException( 563 "Timeout waiting for the services to become healthy. The " 564 + "following services have not started: " 565 + Multimaps.filterKeys(servicesByState, in(ImmutableSet.of(NEW, STARTING)))); 566 } 567 checkHealthy(); 568 } finally { 569 monitor.leave(); 570 } 571 } 572 573 void awaitStopped() { 574 monitor.enterWhenUninterruptibly(stoppedGuard); 575 monitor.leave(); 576 } 577 578 void awaitStopped(long timeout, TimeUnit unit) throws TimeoutException { 579 monitor.enter(); 580 try { 581 if (!monitor.waitForUninterruptibly(stoppedGuard, timeout, unit)) { 582 throw new TimeoutException( 583 "Timeout waiting for the services to stop. The following " 584 + "services have not stopped: " 585 + Multimaps.filterKeys(servicesByState, not(in(EnumSet.of(TERMINATED, FAILED))))); 586 } 587 } finally { 588 monitor.leave(); 589 } 590 } 591 592 ImmutableMultimap<State, Service> servicesByState() { 593 ImmutableSetMultimap.Builder<State, Service> builder = ImmutableSetMultimap.builder(); 594 monitor.enter(); 595 try { 596 for (Entry<State, Service> entry : servicesByState.entries()) { 597 if (!(entry.getValue() instanceof NoOpService)) { 598 builder.put(entry); 599 } 600 } 601 } finally { 602 monitor.leave(); 603 } 604 return builder.build(); 605 } 606 607 ImmutableMap<Service, Long> startupTimes() { 608 List<Entry<Service, Long>> loadTimes; 609 monitor.enter(); 610 try { 611 loadTimes = Lists.newArrayListWithCapacity(startupTimers.size()); 612 // N.B. There will only be an entry in the map if the service has started 613 for (Entry<Service, Stopwatch> entry : startupTimers.entrySet()) { 614 Service service = entry.getKey(); 615 Stopwatch stopWatch = entry.getValue(); 616 if (!stopWatch.isRunning() && !(service instanceof NoOpService)) { 617 loadTimes.add(Maps.immutableEntry(service, stopWatch.elapsed(MILLISECONDS))); 618 } 619 } 620 } finally { 621 monitor.leave(); 622 } 623 Collections.sort( 624 loadTimes, 625 Ordering.natural() 626 .onResultOf( 627 new Function<Entry<Service, Long>, Long>() { 628 @Override 629 public Long apply(Map.Entry<Service, Long> input) { 630 return input.getValue(); 631 } 632 })); 633 return ImmutableMap.copyOf(loadTimes); 634 } 635 636 /** 637 * Updates the state with the given service transition. 638 * 639 * <p>This method performs the main logic of ServiceManager in the following steps. 640 * <ol> 641 * <li>Update the {@link #servicesByState()} 642 * <li>Update the {@link #startupTimers} 643 * <li>Based on the new state queue listeners to run 644 * <li>Run the listeners (outside of the lock) 645 * </ol> 646 */ 647 void transitionService(final Service service, State from, State to) { 648 checkNotNull(service); 649 checkArgument(from != to); 650 monitor.enter(); 651 try { 652 transitioned = true; 653 if (!ready) { 654 return; 655 } 656 // Update state. 657 checkState( 658 servicesByState.remove(from, service), 659 "Service %s not at the expected location in the state map %s", 660 service, 661 from); 662 checkState( 663 servicesByState.put(to, service), 664 "Service %s in the state map unexpectedly at %s", 665 service, 666 to); 667 // Update the timer 668 Stopwatch stopwatch = startupTimers.get(service); 669 if (stopwatch == null) { 670 // This means the service was started by some means other than ServiceManager.startAsync 671 stopwatch = Stopwatch.createStarted(); 672 startupTimers.put(service, stopwatch); 673 } 674 if (to.compareTo(RUNNING) >= 0 && stopwatch.isRunning()) { 675 // N.B. if we miss the STARTING event then we may never record a startup time. 676 stopwatch.stop(); 677 if (!(service instanceof NoOpService)) { 678 logger.log(Level.FINE, "Started {0} in {1}.", new Object[] {service, stopwatch}); 679 } 680 } 681 // Queue our listeners 682 683 // Did a service fail? 684 if (to == FAILED) { 685 enqueueFailedEvent(service); 686 } 687 688 if (states.count(RUNNING) == numberOfServices) { 689 // This means that the manager is currently healthy. N.B. If other threads call isHealthy 690 // they are not guaranteed to get 'true', because any service could fail right now. 691 enqueueHealthyEvent(); 692 } else if (states.count(TERMINATED) + states.count(FAILED) == numberOfServices) { 693 enqueueStoppedEvent(); 694 } 695 } finally { 696 monitor.leave(); 697 // Run our executors outside of the lock 698 dispatchListenerEvents(); 699 } 700 } 701 702 void enqueueStoppedEvent() { 703 listeners.enqueue(STOPPED_EVENT); 704 } 705 706 void enqueueHealthyEvent() { 707 listeners.enqueue(HEALTHY_EVENT); 708 } 709 710 void enqueueFailedEvent(final Service service) { 711 listeners.enqueue( 712 new ListenerCallQueue.Event<Listener>() { 713 @Override 714 public void call(Listener listener) { 715 listener.failure(service); 716 } 717 718 @Override 719 public String toString() { 720 return "failed({service=" + service + "})"; 721 } 722 }); 723 } 724 725 /** Attempts to execute all the listeners in {@link #listeners}. */ 726 void dispatchListenerEvents() { 727 checkState( 728 !monitor.isOccupiedByCurrentThread(), 729 "It is incorrect to execute listeners with the monitor held."); 730 listeners.dispatch(); 731 } 732 733 @GuardedBy("monitor") 734 void checkHealthy() { 735 if (states.count(RUNNING) != numberOfServices) { 736 IllegalStateException exception = 737 new IllegalStateException( 738 "Expected to be healthy after starting. The following services are not running: " 739 + Multimaps.filterKeys(servicesByState, not(equalTo(RUNNING)))); 740 for (Service service : servicesByState.get(State.FAILED)) { 741 exception.addSuppressed(new FailedService(service)); 742 } 743 throw exception; 744 } 745 } 746 } 747 748 /** 749 * A {@link Service} that wraps another service and times how long it takes for it to start and 750 * also calls the {@link ServiceManagerState#transitionService(Service, State, State)}, to record 751 * the state transitions. 752 */ 753 private static final class ServiceListener extends Service.Listener { 754 final Service service; 755 // We store the state in a weak reference to ensure that if something went wrong while 756 // constructing the ServiceManager we don't pointlessly keep updating the state. 757 final WeakReference<ServiceManagerState> state; 758 759 ServiceListener(Service service, WeakReference<ServiceManagerState> state) { 760 this.service = service; 761 this.state = state; 762 } 763 764 @Override 765 public void starting() { 766 ServiceManagerState state = this.state.get(); 767 if (state != null) { 768 state.transitionService(service, NEW, STARTING); 769 if (!(service instanceof NoOpService)) { 770 logger.log(Level.FINE, "Starting {0}.", service); 771 } 772 } 773 } 774 775 @Override 776 public void running() { 777 ServiceManagerState state = this.state.get(); 778 if (state != null) { 779 state.transitionService(service, STARTING, RUNNING); 780 } 781 } 782 783 @Override 784 public void stopping(State from) { 785 ServiceManagerState state = this.state.get(); 786 if (state != null) { 787 state.transitionService(service, from, STOPPING); 788 } 789 } 790 791 @Override 792 public void terminated(State from) { 793 ServiceManagerState state = this.state.get(); 794 if (state != null) { 795 if (!(service instanceof NoOpService)) { 796 logger.log( 797 Level.FINE, 798 "Service {0} has terminated. Previous state was: {1}", 799 new Object[] {service, from}); 800 } 801 state.transitionService(service, from, TERMINATED); 802 } 803 } 804 805 @Override 806 public void failed(State from, Throwable failure) { 807 ServiceManagerState state = this.state.get(); 808 if (state != null) { 809 // Log before the transition, so that if the process exits in response to server failure, 810 // there is a higher likelihood that the cause will be in the logs. 811 boolean log = !(service instanceof NoOpService); 812 /* 813 * We have already exposed startup exceptions to the user in the form of suppressed 814 * exceptions. We don't need to log those exceptions again. 815 */ 816 log &= from != State.STARTING; 817 if (log) { 818 logger.log( 819 Level.SEVERE, 820 "Service " + service + " has failed in the " + from + " state.", 821 failure); 822 } 823 state.transitionService(service, from, FAILED); 824 } 825 } 826 } 827 828 /** 829 * A {@link Service} instance that does nothing. This is only useful as a placeholder to ensure 830 * that the {@link ServiceManager} functions properly even when it is managing no services. 831 * 832 * <p>The use of this class is considered an implementation detail of ServiceManager and as such 833 * it is excluded from {@link #servicesByState}, {@link #startupTimes}, {@link #toString} and all 834 * logging statements. 835 */ 836 private static final class NoOpService extends AbstractService { 837 @Override 838 protected void doStart() { 839 notifyStarted(); 840 } 841 842 @Override 843 protected void doStop() { 844 notifyStopped(); 845 } 846 } 847 848 /** This is never thrown but only used for logging. */ 849 private static final class EmptyServiceManagerWarning extends Throwable {} 850 851 private static final class FailedService extends Throwable { 852 FailedService(Service service) { 853 super( 854 service.toString(), 855 service.failureCause(), 856 false /* don't enable suppression */, 857 false /* don't calculate a stack trace. */); 858 } 859 } 860}