1 |
/* |
2 |
* Written by Doug Lea with assistance from members of JCP JSR-166 |
3 |
* Expert Group and released to the public domain, as explained at |
4 |
* http://creativecommons.org/licenses/publicdomain |
5 |
*/ |
6 |
|
7 |
package jsr166y; |
8 |
|
9 |
import java.util.ArrayList; |
10 |
import java.util.Arrays; |
11 |
import java.util.Collection; |
12 |
import java.util.Collections; |
13 |
import java.util.List; |
14 |
import java.util.concurrent.AbstractExecutorService; |
15 |
import java.util.concurrent.Callable; |
16 |
import java.util.concurrent.ExecutorService; |
17 |
import java.util.concurrent.Future; |
18 |
import java.util.concurrent.RejectedExecutionException; |
19 |
import java.util.concurrent.RunnableFuture; |
20 |
import java.util.concurrent.TimeUnit; |
21 |
import java.util.concurrent.TimeoutException; |
22 |
import java.util.concurrent.atomic.AtomicInteger; |
23 |
import java.util.concurrent.locks.LockSupport; |
24 |
import java.util.concurrent.locks.ReentrantLock; |
25 |
|
26 |
/** |
27 |
* An {@link ExecutorService} for running {@link ForkJoinTask}s. |
28 |
* A {@code ForkJoinPool} provides the entry point for submissions |
29 |
* from non-{@code ForkJoinTask} clients, as well as management and |
30 |
* monitoring operations. |
31 |
* |
32 |
* <p>A {@code ForkJoinPool} differs from other kinds of {@link |
33 |
* ExecutorService} mainly by virtue of employing |
34 |
* <em>work-stealing</em>: all threads in the pool attempt to find and |
35 |
* execute subtasks created by other active tasks (eventually blocking |
36 |
* waiting for work if none exist). This enables efficient processing |
37 |
* when most tasks spawn other subtasks (as do most {@code |
38 |
* ForkJoinTask}s). When setting <em>asyncMode</em> to true in |
39 |
* constructors, {@code ForkJoinPool}s may also be appropriate for use |
40 |
* with event-style tasks that are never joined. |
41 |
* |
42 |
* <p>A {@code ForkJoinPool} is constructed with a given target |
43 |
* parallelism level; by default, equal to the number of available |
44 |
* processors. The pool attempts to maintain enough active (or |
45 |
* available) threads by dynamically adding, suspending, or resuming |
46 |
* internal worker threads, even if some tasks are stalled waiting to |
47 |
* join others. However, no such adjustments are guaranteed in the |
48 |
* face of blocked IO or other unmanaged synchronization. The nested |
49 |
* {@link ManagedBlocker} interface enables extension of the kinds of |
50 |
* synchronization accommodated. |
51 |
* |
52 |
* <p>In addition to execution and lifecycle control methods, this |
53 |
* class provides status check methods (for example |
54 |
* {@link #getStealCount}) that are intended to aid in developing, |
55 |
* tuning, and monitoring fork/join applications. Also, method |
56 |
* {@link #toString} returns indications of pool state in a |
57 |
* convenient form for informal monitoring. |
58 |
* |
59 |
* <p> As is the case with other ExecutorServices, there are three |
60 |
* main task execution methods summarized in the following |
61 |
* table. These are designed to be used by clients not already engaged |
62 |
* in fork/join computations in the current pool. The main forms of |
63 |
* these methods accept instances of {@code ForkJoinTask}, but |
64 |
* overloaded forms also allow mixed execution of plain {@code |
65 |
* Runnable}- or {@code Callable}- based activities as well. However, |
66 |
* tasks that are already executing in a pool should normally |
67 |
* <em>NOT</em> use these pool execution methods, but instead use the |
68 |
* within-computation forms listed in the table. |
69 |
* |
70 |
* <table BORDER CELLPADDING=3 CELLSPACING=1> |
71 |
* <tr> |
72 |
* <td></td> |
73 |
* <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td> |
74 |
* <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td> |
75 |
* </tr> |
76 |
* <tr> |
77 |
* <td> <b>Arrange async execution</td> |
78 |
* <td> {@link #execute(ForkJoinTask)}</td> |
79 |
* <td> {@link ForkJoinTask#fork}</td> |
80 |
* </tr> |
81 |
* <tr> |
82 |
* <td> <b>Await and obtain result</td> |
83 |
* <td> {@link #invoke(ForkJoinTask)}</td> |
84 |
* <td> {@link ForkJoinTask#invoke}</td> |
85 |
* </tr> |
86 |
* <tr> |
87 |
* <td> <b>Arrange exec and obtain Future</td> |
88 |
* <td> {@link #submit(ForkJoinTask)}</td> |
89 |
* <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td> |
90 |
* </tr> |
91 |
* </table> |
92 |
* |
93 |
* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is |
94 |
* used for all parallel task execution in a program or subsystem. |
95 |
* Otherwise, use would not usually outweigh the construction and |
96 |
* bookkeeping overhead of creating a large set of threads. For |
97 |
* example, a common pool could be used for the {@code SortTasks} |
98 |
* illustrated in {@link RecursiveAction}. Because {@code |
99 |
* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon |
100 |
* daemon} mode, there is typically no need to explicitly {@link |
101 |
* #shutdown} such a pool upon program exit. |
102 |
* |
103 |
* <pre> |
104 |
* static final ForkJoinPool mainPool = new ForkJoinPool(); |
105 |
* ... |
106 |
* public void sort(long[] array) { |
107 |
* mainPool.invoke(new SortTask(array, 0, array.length)); |
108 |
* } |
109 |
* </pre> |
110 |
* |
111 |
* <p><b>Implementation notes</b>: This implementation restricts the |
112 |
* maximum number of running threads to 32767. Attempts to create |
113 |
* pools with greater than the maximum number result in |
114 |
* {@code IllegalArgumentException}. |
115 |
* |
116 |
* <p>This implementation rejects submitted tasks (that is, by throwing |
117 |
* {@link RejectedExecutionException}) only when the pool is shut down |
118 |
* or internal resources have been exhausted. |
119 |
* |
120 |
* @since 1.7 |
121 |
* @author Doug Lea |
122 |
*/ |
123 |
public class ForkJoinPool extends AbstractExecutorService { |
124 |
|
125 |
/* |
126 |
* Implementation Overview |
127 |
* |
128 |
* This class provides the central bookkeeping and control for a |
129 |
* set of worker threads: Submissions from non-FJ threads enter |
130 |
* into a submission queue. Workers take these tasks and typically |
131 |
* split them into subtasks that may be stolen by other workers. |
132 |
* The main work-stealing mechanics implemented in class |
133 |
* ForkJoinWorkerThread give first priority to processing tasks |
134 |
* from their own queues (LIFO or FIFO, depending on mode), then |
135 |
* to randomized FIFO steals of tasks in other worker queues, and |
136 |
* lastly to new submissions. These mechanics do not consider |
137 |
* affinities, loads, cache localities, etc, so rarely provide the |
138 |
* best possible performance on a given machine, but portably |
139 |
* provide good throughput by averaging over these factors. |
140 |
* (Further, even if we did try to use such information, we do not |
141 |
* usually have a basis for exploiting it. For example, some sets |
142 |
* of tasks profit from cache affinities, but others are harmed by |
143 |
* cache pollution effects.) |
144 |
* |
145 |
* Beyond work-stealing support and essential bookkeeping, the |
146 |
* main responsibility of this framework is to take actions when |
147 |
* one worker is waiting to join a task stolen (or always held by) |
148 |
* another. Because we are multiplexing many tasks on to a pool |
149 |
* of workers, we can't just let them block (as in Thread.join). |
150 |
* We also cannot just reassign the joiner's run-time stack with |
151 |
* another and replace it later, which would be a form of |
152 |
* "continuation", that even if possible is not necessarily a good |
153 |
* idea. Given that the creation costs of most threads on most |
154 |
* systems mainly surrounds setting up runtime stacks, thread |
155 |
* creation and switching is usually not much more expensive than |
156 |
* stack creation and switching, and is more flexible). Instead we |
157 |
* combine two tactics: |
158 |
* |
159 |
* Helping: Arranging for the joiner to execute some task that it |
160 |
* would be running if the steal had not occurred. Method |
161 |
* ForkJoinWorkerThread.helpJoinTask tracks joining->stealing |
162 |
* links to try to find such a task. |
163 |
* |
164 |
* Compensating: Unless there are already enough live threads, |
165 |
* method helpMaintainParallelism() may create or |
166 |
* re-activate a spare thread to compensate for blocked |
167 |
* joiners until they unblock. |
168 |
* |
169 |
* It is impossible to keep exactly the target (parallelism) |
170 |
* number of threads running at any given time. Determining |
171 |
* existence of conservatively safe helping targets, the |
172 |
* availability of already-created spares, and the apparent need |
173 |
* to create new spares are all racy and require heuristic |
174 |
* guidance, so we rely on multiple retries of each. Compensation |
175 |
* occurs in slow-motion. It is triggered only upon timeouts of |
176 |
* Object.wait used for joins. This reduces poor decisions that |
177 |
* would otherwise be made when threads are waiting for others |
178 |
* that are stalled because of unrelated activities such as |
179 |
* garbage collection. |
180 |
* |
181 |
* The ManagedBlocker extension API can't use helping so relies |
182 |
* only on compensation in method awaitBlocker. |
183 |
* |
184 |
* The main throughput advantages of work-stealing stem from |
185 |
* decentralized control -- workers mostly steal tasks from each |
186 |
* other. We do not want to negate this by creating bottlenecks |
187 |
* implementing other management responsibilities. So we use a |
188 |
* collection of techniques that avoid, reduce, or cope well with |
189 |
* contention. These entail several instances of bit-packing into |
190 |
* CASable fields to maintain only the minimally required |
191 |
* atomicity. To enable such packing, we restrict maximum |
192 |
* parallelism to (1<<15)-1 (enabling twice this (to accommodate |
193 |
* unbalanced increments and decrements) to fit into a 16 bit |
194 |
* field, which is far in excess of normal operating range. Even |
195 |
* though updates to some of these bookkeeping fields do sometimes |
196 |
* contend with each other, they don't normally cache-contend with |
197 |
* updates to others enough to warrant memory padding or |
198 |
* isolation. So they are all held as fields of ForkJoinPool |
199 |
* objects. The main capabilities are as follows: |
200 |
* |
201 |
* 1. Creating and removing workers. Workers are recorded in the |
202 |
* "workers" array. This is an array as opposed to some other data |
203 |
* structure to support index-based random steals by workers. |
204 |
* Updates to the array recording new workers and unrecording |
205 |
* terminated ones are protected from each other by a lock |
206 |
* (workerLock) but the array is otherwise concurrently readable, |
207 |
* and accessed directly by workers. To simplify index-based |
208 |
* operations, the array size is always a power of two, and all |
209 |
* readers must tolerate null slots. Currently, all worker thread |
210 |
* creation is on-demand, triggered by task submissions, |
211 |
* replacement of terminated workers, and/or compensation for |
212 |
* blocked workers. However, all other support code is set up to |
213 |
* work with other policies. |
214 |
* |
215 |
* To ensure that we do not hold on to worker references that |
216 |
* would prevent GC, ALL accesses to workers are via indices into |
217 |
* the workers array (which is one source of some of the unusual |
218 |
* code constructions here). In essence, the workers array serves |
219 |
* as a WeakReference mechanism. Thus for example the event queue |
220 |
* stores worker indices, not worker references. Access to the |
221 |
* workers in associated methods (for example releaseEventWaiters) |
222 |
* must both index-check and null-check the IDs. All such accesses |
223 |
* ignore bad IDs by returning out early from what they are doing, |
224 |
* since this can only be associated with shutdown, in which case |
225 |
* it is OK to give up. On termination, we just clobber these |
226 |
* data structures without trying to use them. |
227 |
* |
228 |
* 2. Bookkeeping for dynamically adding and removing workers. We |
229 |
* aim to approximately maintain the given level of parallelism. |
230 |
* When some workers are known to be blocked (on joins or via |
231 |
* ManagedBlocker), we may create or resume others to take their |
232 |
* place until they unblock (see below). Implementing this |
233 |
* requires counts of the number of "running" threads (i.e., those |
234 |
* that are neither blocked nor artificially suspended) as well as |
235 |
* the total number. These two values are packed into one field, |
236 |
* "workerCounts" because we need accurate snapshots when deciding |
237 |
* to create, resume or suspend. Note however that the |
238 |
* correspondence of these counts to reality is not guaranteed. In |
239 |
* particular updates for unblocked threads may lag until they |
240 |
* actually wake up. |
241 |
* |
242 |
* 3. Maintaining global run state. The run state of the pool |
243 |
* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to |
244 |
* those in other Executor implementations, as well as a count of |
245 |
* "active" workers -- those that are, or soon will be, or |
246 |
* recently were executing tasks. The runLevel and active count |
247 |
* are packed together in order to correctly trigger shutdown and |
248 |
* termination. Without care, active counts can be subject to very |
249 |
* high contention. We substantially reduce this contention by |
250 |
* relaxing update rules. A worker must claim active status |
251 |
* prospectively, by activating if it sees that a submitted or |
252 |
* stealable task exists (it may find after activating that the |
253 |
* task no longer exists). It stays active while processing this |
254 |
* task (if it exists) and any other local subtasks it produces, |
255 |
* until it cannot find any other tasks. It then tries |
256 |
* inactivating (see method preStep), but upon update contention |
257 |
* instead scans for more tasks, later retrying inactivation if it |
258 |
* doesn't find any. |
259 |
* |
260 |
* 4. Managing idle workers waiting for tasks. We cannot let |
261 |
* workers spin indefinitely scanning for tasks when none are |
262 |
* available. On the other hand, we must quickly prod them into |
263 |
* action when new tasks are submitted or generated. We |
264 |
* park/unpark these idle workers using an event-count scheme. |
265 |
* Field eventCount is incremented upon events that may enable |
266 |
* workers that previously could not find a task to now find one: |
267 |
* Submission of a new task to the pool, or another worker pushing |
268 |
* a task onto a previously empty queue. (We also use this |
269 |
* mechanism for configuration and termination actions that |
270 |
* require wakeups of idle workers). Each worker maintains its |
271 |
* last known event count, and blocks when a scan for work did not |
272 |
* find a task AND its lastEventCount matches the current |
273 |
* eventCount. Waiting idle workers are recorded in a variant of |
274 |
* Treiber stack headed by field eventWaiters which, when nonzero, |
275 |
* encodes the thread index and count awaited for by the worker |
276 |
* thread most recently calling eventSync. This thread in turn has |
277 |
* a record (field nextEventWaiter) for the next waiting worker. |
278 |
* In addition to allowing simpler decisions about need for |
279 |
* wakeup, the event count bits in eventWaiters serve the role of |
280 |
* tags to avoid ABA errors in Treiber stacks. Upon any wakeup, |
281 |
* released threads also try to release at most two others. The |
282 |
* net effect is a tree-like diffusion of signals, where released |
283 |
* threads (and possibly others) help with unparks. To further |
284 |
* reduce contention effects a bit, failed CASes to increment |
285 |
* field eventCount are tolerated without retries in signalWork. |
286 |
* Conceptually they are merged into the same event, which is OK |
287 |
* when their only purpose is to enable workers to scan for work. |
288 |
* |
289 |
* 5. Managing suspension of extra workers. When a worker notices |
290 |
* (usually upon timeout of a wait()) that there are too few |
291 |
* running threads, we may create a new thread to maintain |
292 |
* parallelism level, or at least avoid starvation. Usually, extra |
293 |
* threads are needed for only very short periods, yet join |
294 |
* dependencies are such that we sometimes need them in |
295 |
* bursts. Rather than create new threads each time this happens, |
296 |
* we suspend no-longer-needed extra ones as "spares". For most |
297 |
* purposes, we don't distinguish "extra" spare threads from |
298 |
* normal "core" threads: On each call to preStep (the only point |
299 |
* at which we can do this) a worker checks to see if there are |
300 |
* now too many running workers, and if so, suspends itself. |
301 |
* Method helpMaintainParallelism looks for suspended threads to |
302 |
* resume before considering creating a new replacement. The |
303 |
* spares themselves are encoded on another variant of a Treiber |
304 |
* Stack, headed at field "spareWaiters". Note that the use of |
305 |
* spares is intrinsically racy. One thread may become a spare at |
306 |
* about the same time as another is needlessly being created. We |
307 |
* counteract this and related slop in part by requiring resumed |
308 |
* spares to immediately recheck (in preStep) to see whether they |
309 |
* should re-suspend. |
310 |
* |
311 |
* 6. Killing off unneeded workers. A timeout mechanism is used to |
312 |
* shed unused workers: The oldest (first) event queue waiter uses |
313 |
* a timed rather than hard wait. When this wait times out without |
314 |
* a normal wakeup, it tries to shutdown any one (for convenience |
315 |
* the newest) other spare or event waiter via |
316 |
* tryShutdownUnusedWorker. This eventually reduces the number of |
317 |
* worker threads to a minimum of one after a long enough period |
318 |
* without use. |
319 |
* |
320 |
* 7. Deciding when to create new workers. The main dynamic |
321 |
* control in this class is deciding when to create extra threads |
322 |
* in method helpMaintainParallelism. We would like to keep |
323 |
* exactly #parallelism threads running, which is an impossible |
324 |
* task. We always need to create one when the number of running |
325 |
* threads would become zero and all workers are busy. Beyond |
326 |
* this, we must rely on heuristics that work well in the |
327 |
* presence of transient phenomena such as GC stalls, dynamic |
328 |
* compilation, and wake-up lags. These transients are extremely |
329 |
* common -- we are normally trying to fully saturate the CPUs on |
330 |
* a machine, so almost any activity other than running tasks |
331 |
* impedes accuracy. Our main defense is to allow parallelism to |
332 |
* lapse for a while during joins, and use a timeout to see if, |
333 |
* after the resulting settling, there is still a need for |
334 |
* additional workers. This also better copes with the fact that |
335 |
* some of the methods in this class tend to never become compiled |
336 |
* (but are interpreted), so some components of the entire set of |
337 |
* controls might execute 100 times faster than others. And |
338 |
* similarly for cases where the apparent lack of work is just due |
339 |
* to GC stalls and other transient system activity. |
340 |
* |
341 |
* Beware that there is a lot of representation-level coupling |
342 |
* among classes ForkJoinPool, ForkJoinWorkerThread, and |
343 |
* ForkJoinTask. For example, direct access to "workers" array by |
344 |
* workers, and direct access to ForkJoinTask.status by both |
345 |
* ForkJoinPool and ForkJoinWorkerThread. There is little point |
346 |
* trying to reduce this, since any associated future changes in |
347 |
* representations will need to be accompanied by algorithmic |
348 |
* changes anyway. |
349 |
* |
350 |
* Style notes: There are lots of inline assignments (of form |
351 |
* "while ((local = field) != 0)") which are usually the simplest |
352 |
* way to ensure the required read orderings (which are sometimes |
353 |
* critical). Also several occurrences of the unusual "do {} |
354 |
* while (!cas...)" which is the simplest way to force an update of |
355 |
* a CAS'ed variable. There are also other coding oddities that |
356 |
* help some methods perform reasonably even when interpreted (not |
357 |
* compiled), at the expense of some messy constructions that |
358 |
* reduce byte code counts. |
359 |
* |
360 |
* The order of declarations in this file is: (1) statics (2) |
361 |
* fields (along with constants used when unpacking some of them) |
362 |
* (3) internal control methods (4) callbacks and other support |
363 |
* for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported |
364 |
* methods (plus a few little helpers). |
365 |
*/ |
366 |
|
367 |
/** |
368 |
* Factory for creating new {@link ForkJoinWorkerThread}s. |
369 |
* A {@code ForkJoinWorkerThreadFactory} must be defined and used |
370 |
* for {@code ForkJoinWorkerThread} subclasses that extend base |
371 |
* functionality or initialize threads with different contexts. |
372 |
*/ |
373 |
public static interface ForkJoinWorkerThreadFactory { |
374 |
/** |
375 |
* Returns a new worker thread operating in the given pool. |
376 |
* |
377 |
* @param pool the pool this thread works in |
378 |
* @throws NullPointerException if the pool is null |
379 |
*/ |
380 |
public ForkJoinWorkerThread newThread(ForkJoinPool pool); |
381 |
} |
382 |
|
383 |
/** |
384 |
* Default ForkJoinWorkerThreadFactory implementation; creates a |
385 |
* new ForkJoinWorkerThread. |
386 |
*/ |
387 |
static class DefaultForkJoinWorkerThreadFactory |
388 |
implements ForkJoinWorkerThreadFactory { |
389 |
public ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
390 |
return new ForkJoinWorkerThread(pool); |
391 |
} |
392 |
} |
393 |
|
394 |
/** |
395 |
* Creates a new ForkJoinWorkerThread. This factory is used unless |
396 |
* overridden in ForkJoinPool constructors. |
397 |
*/ |
398 |
public static final ForkJoinWorkerThreadFactory |
399 |
defaultForkJoinWorkerThreadFactory = |
400 |
new DefaultForkJoinWorkerThreadFactory(); |
401 |
|
402 |
/** |
403 |
* Permission required for callers of methods that may start or |
404 |
* kill threads. |
405 |
*/ |
406 |
private static final RuntimePermission modifyThreadPermission = |
407 |
new RuntimePermission("modifyThread"); |
408 |
|
409 |
/** |
410 |
* If there is a security manager, makes sure caller has |
411 |
* permission to modify threads. |
412 |
*/ |
413 |
private static void checkPermission() { |
414 |
SecurityManager security = System.getSecurityManager(); |
415 |
if (security != null) |
416 |
security.checkPermission(modifyThreadPermission); |
417 |
} |
418 |
|
419 |
/** |
420 |
* Generator for assigning sequence numbers as pool names. |
421 |
*/ |
422 |
private static final AtomicInteger poolNumberGenerator = |
423 |
new AtomicInteger(); |
424 |
|
425 |
/** |
426 |
* The time to block in a join (see awaitJoin) before checking if |
427 |
* a new worker should be (re)started to maintain parallelism |
428 |
* level. The value should be short enough to maintain global |
429 |
* responsiveness and progress but long enough to avoid |
430 |
* counterproductive firings during GC stalls or unrelated system |
431 |
* activity, and to not bog down systems with continual re-firings |
432 |
* on GCs or legitimately long waits. |
433 |
*/ |
434 |
private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second |
435 |
|
436 |
/** |
437 |
* The wakeup interval (in nanoseconds) for the oldest worker |
438 |
* waiting for an event to invoke tryShutdownUnusedWorker to |
439 |
* shrink the number of workers. The exact value does not matter |
440 |
* too much. It must be short enough to release resources during |
441 |
* sustained periods of idleness, but not so short that threads |
442 |
* are continually re-created. |
443 |
*/ |
444 |
private static final long SHRINK_RATE_NANOS = |
445 |
30L * 1000L * 1000L * 1000L; // 2 per minute |
446 |
|
447 |
/** |
448 |
* Absolute bound for parallelism level. Twice this number plus |
449 |
* one (i.e., 0xfff) must fit into a 16bit field to enable |
450 |
* word-packing for some counts and indices. |
451 |
*/ |
452 |
private static final int MAX_WORKERS = 0x7fff; |
453 |
|
454 |
/** |
455 |
* Array holding all worker threads in the pool. Array size must |
456 |
* be a power of two. Updates and replacements are protected by |
457 |
* workerLock, but the array is always kept in a consistent enough |
458 |
* state to be randomly accessed without locking by workers |
459 |
* performing work-stealing, as well as other traversal-based |
460 |
* methods in this class. All readers must tolerate that some |
461 |
* array slots may be null. |
462 |
*/ |
463 |
volatile ForkJoinWorkerThread[] workers; |
464 |
|
465 |
/** |
466 |
* Queue for external submissions. |
467 |
*/ |
468 |
private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue; |
469 |
|
470 |
/** |
471 |
* Lock protecting updates to workers array. |
472 |
*/ |
473 |
private final ReentrantLock workerLock; |
474 |
|
475 |
/** |
476 |
* Latch released upon termination. |
477 |
*/ |
478 |
private final Phaser termination; |
479 |
|
480 |
/** |
481 |
* Creation factory for worker threads. |
482 |
*/ |
483 |
private final ForkJoinWorkerThreadFactory factory; |
484 |
|
485 |
/** |
486 |
* Sum of per-thread steal counts, updated only when threads are |
487 |
* idle or terminating. |
488 |
*/ |
489 |
private volatile long stealCount; |
490 |
|
491 |
/** |
492 |
* Encoded record of top of Treiber stack of threads waiting for |
493 |
* events. The top 32 bits contain the count being waited for. The |
494 |
* bottom 16 bits contains one plus the pool index of waiting |
495 |
* worker thread. (Bits 16-31 are unused.) |
496 |
*/ |
497 |
private volatile long eventWaiters; |
498 |
|
499 |
private static final int EVENT_COUNT_SHIFT = 32; |
500 |
private static final int WAITER_ID_MASK = (1 << 16) - 1; |
501 |
|
502 |
/** |
503 |
* A counter for events that may wake up worker threads: |
504 |
* - Submission of a new task to the pool |
505 |
* - A worker pushing a task on an empty queue |
506 |
* - termination |
507 |
*/ |
508 |
private volatile int eventCount; |
509 |
|
510 |
/** |
511 |
* Encoded record of top of Treiber stack of spare threads waiting |
512 |
* for resumption. The top 16 bits contain an arbitrary count to |
513 |
* avoid ABA effects. The bottom 16bits contains one plus the pool |
514 |
* index of waiting worker thread. |
515 |
*/ |
516 |
private volatile int spareWaiters; |
517 |
|
518 |
private static final int SPARE_COUNT_SHIFT = 16; |
519 |
private static final int SPARE_ID_MASK = (1 << 16) - 1; |
520 |
|
521 |
/** |
522 |
* Lifecycle control. The low word contains the number of workers |
523 |
* that are (probably) executing tasks. This value is atomically |
524 |
* incremented before a worker gets a task to run, and decremented |
525 |
* when a worker has no tasks and cannot find any. Bits 16-18 |
526 |
* contain runLevel value. When all are zero, the pool is |
527 |
* running. Level transitions are monotonic (running -> shutdown |
528 |
* -> terminating -> terminated) so each transition adds a bit. |
529 |
* These are bundled together to ensure consistent read for |
530 |
* termination checks (i.e., that runLevel is at least SHUTDOWN |
531 |
* and active threads is zero). |
532 |
* |
533 |
* Notes: Most direct CASes are dependent on these bitfield |
534 |
* positions. Also, this field is non-private to enable direct |
535 |
* performance-sensitive CASes in ForkJoinWorkerThread. |
536 |
*/ |
537 |
volatile int runState; |
538 |
|
539 |
// Note: The order among run level values matters. |
540 |
private static final int RUNLEVEL_SHIFT = 16; |
541 |
private static final int SHUTDOWN = 1 << RUNLEVEL_SHIFT; |
542 |
private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1); |
543 |
private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2); |
544 |
private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1; |
545 |
|
546 |
/** |
547 |
* Holds number of total (i.e., created and not yet terminated) |
548 |
* and running (i.e., not blocked on joins or other managed sync) |
549 |
* threads, packed together to ensure consistent snapshot when |
550 |
* making decisions about creating and suspending spare |
551 |
* threads. Updated only by CAS. Note that adding a new worker |
552 |
* requires incrementing both counts, since workers start off in |
553 |
* running state. |
554 |
*/ |
555 |
private volatile int workerCounts; |
556 |
|
557 |
private static final int TOTAL_COUNT_SHIFT = 16; |
558 |
private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1; |
559 |
private static final int ONE_RUNNING = 1; |
560 |
private static final int ONE_TOTAL = 1 << TOTAL_COUNT_SHIFT; |
561 |
|
562 |
/** |
563 |
* The target parallelism level. |
564 |
* Accessed directly by ForkJoinWorkerThreads. |
565 |
*/ |
566 |
final int parallelism; |
567 |
|
568 |
/** |
569 |
* True if use local fifo, not default lifo, for local polling |
570 |
* Read by, and replicated by ForkJoinWorkerThreads |
571 |
*/ |
572 |
final boolean locallyFifo; |
573 |
|
574 |
/** |
575 |
* The uncaught exception handler used when any worker abruptly |
576 |
* terminates. |
577 |
*/ |
578 |
private final Thread.UncaughtExceptionHandler ueh; |
579 |
|
580 |
/** |
581 |
* Pool number, just for assigning useful names to worker threads |
582 |
*/ |
583 |
private final int poolNumber; |
584 |
|
585 |
// Utilities for CASing fields. Note that most of these |
586 |
// are usually manually inlined by callers |
587 |
|
588 |
/** |
589 |
* Increments running count part of workerCounts. |
590 |
*/ |
591 |
final void incrementRunningCount() { |
592 |
int c; |
593 |
do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
594 |
c = workerCounts, |
595 |
c + ONE_RUNNING)); |
596 |
} |
597 |
|
598 |
/** |
599 |
* Tries to increment running count part of workerCounts. |
600 |
*/ |
601 |
final boolean tryIncrementRunningCount() { |
602 |
int c; |
603 |
return UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
604 |
c = workerCounts, |
605 |
c + ONE_RUNNING); |
606 |
} |
607 |
|
608 |
/** |
609 |
* Tries to decrement running count unless already zero. |
610 |
*/ |
611 |
final boolean tryDecrementRunningCount() { |
612 |
int wc = workerCounts; |
613 |
if ((wc & RUNNING_COUNT_MASK) == 0) |
614 |
return false; |
615 |
return UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
616 |
wc, wc - ONE_RUNNING); |
617 |
} |
618 |
|
619 |
/** |
620 |
* Forces decrement of encoded workerCounts, awaiting nonzero if |
621 |
* (rarely) necessary when other count updates lag. |
622 |
* |
623 |
* @param dr -- either zero or ONE_RUNNING |
624 |
* @param dt -- either zero or ONE_TOTAL |
625 |
*/ |
626 |
private void decrementWorkerCounts(int dr, int dt) { |
627 |
for (;;) { |
628 |
int wc = workerCounts; |
629 |
if ((wc & RUNNING_COUNT_MASK) - dr < 0 || |
630 |
(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) { |
631 |
if ((runState & TERMINATED) != 0) |
632 |
return; // lagging termination on a backout |
633 |
Thread.yield(); |
634 |
} |
635 |
if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
636 |
wc, wc - (dr + dt))) |
637 |
return; |
638 |
} |
639 |
} |
640 |
|
641 |
/** |
642 |
* Tries decrementing active count; fails on contention. |
643 |
* Called when workers cannot find tasks to run. |
644 |
*/ |
645 |
final boolean tryDecrementActiveCount() { |
646 |
int c; |
647 |
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
648 |
c = runState, c - 1); |
649 |
} |
650 |
|
651 |
/** |
652 |
* Advances to at least the given level. Returns true if not |
653 |
* already in at least the given level. |
654 |
*/ |
655 |
private boolean advanceRunLevel(int level) { |
656 |
for (;;) { |
657 |
int s = runState; |
658 |
if ((s & level) != 0) |
659 |
return false; |
660 |
if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level)) |
661 |
return true; |
662 |
} |
663 |
} |
664 |
|
665 |
// workers array maintenance |
666 |
|
667 |
/** |
668 |
* Records and returns a workers array index for new worker. |
669 |
*/ |
670 |
private int recordWorker(ForkJoinWorkerThread w) { |
671 |
// Try using slot totalCount-1. If not available, scan and/or resize |
672 |
int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1; |
673 |
final ReentrantLock lock = this.workerLock; |
674 |
lock.lock(); |
675 |
try { |
676 |
ForkJoinWorkerThread[] ws = workers; |
677 |
int n = ws.length; |
678 |
if (k < 0 || k >= n || ws[k] != null) { |
679 |
for (k = 0; k < n && ws[k] != null; ++k) |
680 |
; |
681 |
if (k == n) |
682 |
ws = workers = Arrays.copyOf(ws, n << 1); |
683 |
} |
684 |
ws[k] = w; |
685 |
int c = eventCount; // advance event count to ensure visibility |
686 |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1); |
687 |
} finally { |
688 |
lock.unlock(); |
689 |
} |
690 |
return k; |
691 |
} |
692 |
|
693 |
/** |
694 |
* Nulls out record of worker in workers array. |
695 |
*/ |
696 |
private void forgetWorker(ForkJoinWorkerThread w) { |
697 |
int idx = w.poolIndex; |
698 |
// Locking helps method recordWorker avoid unnecessary expansion |
699 |
final ReentrantLock lock = this.workerLock; |
700 |
lock.lock(); |
701 |
try { |
702 |
ForkJoinWorkerThread[] ws = workers; |
703 |
if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify |
704 |
ws[idx] = null; |
705 |
} finally { |
706 |
lock.unlock(); |
707 |
} |
708 |
} |
709 |
|
710 |
/** |
711 |
* Final callback from terminating worker. Removes record of |
712 |
* worker from array, and adjusts counts. If pool is shutting |
713 |
* down, tries to complete termination. |
714 |
* |
715 |
* @param w the worker |
716 |
*/ |
717 |
final void workerTerminated(ForkJoinWorkerThread w) { |
718 |
forgetWorker(w); |
719 |
decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL); |
720 |
while (w.stealCount != 0) // collect final count |
721 |
tryAccumulateStealCount(w); |
722 |
tryTerminate(false); |
723 |
} |
724 |
|
725 |
// Waiting for and signalling events |
726 |
|
727 |
/** |
728 |
* Releases workers blocked on a count not equal to current count. |
729 |
* Normally called after precheck that eventWaiters isn't zero to |
730 |
* avoid wasted array checks. Gives up upon a change in count or |
731 |
* upon releasing four workers, letting others take over. |
732 |
*/ |
733 |
private void releaseEventWaiters() { |
734 |
ForkJoinWorkerThread[] ws = workers; |
735 |
int n = ws.length; |
736 |
long h = eventWaiters; |
737 |
int ec = eventCount; |
738 |
int releases = 4; |
739 |
ForkJoinWorkerThread w; int id; |
740 |
while ((id = (((int)h) & WAITER_ID_MASK) - 1) >= 0 && |
741 |
(int)(h >>> EVENT_COUNT_SHIFT) != ec && |
742 |
id < n && (w = ws[id]) != null) { |
743 |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
744 |
h, w.nextWaiter)) { |
745 |
LockSupport.unpark(w); |
746 |
if (--releases == 0) |
747 |
break; |
748 |
} |
749 |
if (eventCount != ec) |
750 |
break; |
751 |
h = eventWaiters; |
752 |
} |
753 |
} |
754 |
|
755 |
/** |
756 |
* Tries to advance eventCount and releases waiters. Called only |
757 |
* from workers. |
758 |
*/ |
759 |
final void signalWork() { |
760 |
int c; // try to increment event count -- CAS failure OK |
761 |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); |
762 |
if (eventWaiters != 0L) |
763 |
releaseEventWaiters(); |
764 |
} |
765 |
|
766 |
/** |
767 |
* Adds the given worker to event queue and blocks until |
768 |
* terminating or event count advances from the given value |
769 |
* |
770 |
* @param w the calling worker thread |
771 |
* @param ec the count |
772 |
*/ |
773 |
private void eventSync(ForkJoinWorkerThread w, int ec) { |
774 |
long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1)); |
775 |
long h; |
776 |
while ((runState < SHUTDOWN || !tryTerminate(false)) && |
777 |
(((int)(h = eventWaiters) & WAITER_ID_MASK) == 0 || |
778 |
(int)(h >>> EVENT_COUNT_SHIFT) == ec) && |
779 |
eventCount == ec) { |
780 |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
781 |
w.nextWaiter = h, nh)) { |
782 |
awaitEvent(w, ec); |
783 |
break; |
784 |
} |
785 |
} |
786 |
} |
787 |
|
788 |
/** |
789 |
* Blocks the given worker (that has already been entered as an |
790 |
* event waiter) until terminating or event count advances from |
791 |
* the given value. The oldest (first) waiter uses a timed wait to |
792 |
* occasionally one-by-one shrink the number of workers (to a |
793 |
* minimum of one) if the pool has not been used for extended |
794 |
* periods. |
795 |
* |
796 |
* @param w the calling worker thread |
797 |
* @param ec the count |
798 |
*/ |
799 |
private void awaitEvent(ForkJoinWorkerThread w, int ec) { |
800 |
while (eventCount == ec) { |
801 |
if (tryAccumulateStealCount(w)) { // transfer while idle |
802 |
boolean untimed = (w.nextWaiter != 0L || |
803 |
(workerCounts & RUNNING_COUNT_MASK) <= 1); |
804 |
long startTime = untimed ? 0 : System.nanoTime(); |
805 |
Thread.interrupted(); // clear/ignore interrupt |
806 |
if (w.isTerminating() || eventCount != ec) |
807 |
break; // recheck after clear |
808 |
if (untimed) |
809 |
LockSupport.park(w); |
810 |
else { |
811 |
LockSupport.parkNanos(w, SHRINK_RATE_NANOS); |
812 |
if (eventCount != ec || w.isTerminating()) |
813 |
break; |
814 |
if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS) |
815 |
tryShutdownUnusedWorker(ec); |
816 |
} |
817 |
} |
818 |
} |
819 |
} |
820 |
|
821 |
// Maintaining parallelism |
822 |
|
823 |
/** |
824 |
* Pushes worker onto the spare stack. |
825 |
*/ |
826 |
final void pushSpare(ForkJoinWorkerThread w) { |
827 |
int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1); |
828 |
do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
829 |
w.nextSpare = spareWaiters,ns)); |
830 |
} |
831 |
|
832 |
/** |
833 |
* Tries (once) to resume a spare if the number of running |
834 |
* threads is less than target. |
835 |
*/ |
836 |
private void tryResumeSpare() { |
837 |
int sw, id; |
838 |
ForkJoinWorkerThread[] ws = workers; |
839 |
int n = ws.length; |
840 |
ForkJoinWorkerThread w; |
841 |
if ((sw = spareWaiters) != 0 && |
842 |
(id = (sw & SPARE_ID_MASK) - 1) >= 0 && |
843 |
id < n && (w = ws[id]) != null && |
844 |
(runState >= TERMINATING || |
845 |
(workerCounts & RUNNING_COUNT_MASK) < parallelism) && |
846 |
spareWaiters == sw && |
847 |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
848 |
sw, w.nextSpare)) { |
849 |
int c; // increment running count before resume |
850 |
do {} while (!UNSAFE.compareAndSwapInt |
851 |
(this, workerCountsOffset, |
852 |
c = workerCounts, c + ONE_RUNNING)); |
853 |
if (w.tryUnsuspend()) |
854 |
LockSupport.unpark(w); |
855 |
else // back out if w was shutdown |
856 |
decrementWorkerCounts(ONE_RUNNING, 0); |
857 |
} |
858 |
} |
859 |
|
860 |
/** |
861 |
* Tries to increase the number of running workers if below target |
862 |
* parallelism: If a spare exists tries to resume it via |
863 |
* tryResumeSpare. Otherwise, if not enough total workers or all |
864 |
* existing workers are busy, adds a new worker. In all cases also |
865 |
* helps wake up releasable workers waiting for work. |
866 |
*/ |
867 |
private void helpMaintainParallelism() { |
868 |
int pc = parallelism; |
869 |
int wc, rs, tc; |
870 |
while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc && |
871 |
(rs = runState) < TERMINATING) { |
872 |
if (spareWaiters != 0) |
873 |
tryResumeSpare(); |
874 |
else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS || |
875 |
(tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc)) |
876 |
break; // enough total |
877 |
else if (runState == rs && workerCounts == wc && |
878 |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
879 |
wc + (ONE_RUNNING|ONE_TOTAL))) { |
880 |
ForkJoinWorkerThread w = null; |
881 |
Throwable fail = null; |
882 |
try { |
883 |
w = factory.newThread(this); |
884 |
} catch (Throwable ex) { |
885 |
fail = ex; |
886 |
} |
887 |
if (w == null) { // null or exceptional factory return |
888 |
decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL); |
889 |
tryTerminate(false); // handle failure during shutdown |
890 |
// If originating from an external caller, |
891 |
// propagate exception, else ignore |
892 |
if (fail != null && runState < TERMINATING && |
893 |
!(Thread.currentThread() instanceof |
894 |
ForkJoinWorkerThread)) |
895 |
UNSAFE.throwException(fail); |
896 |
break; |
897 |
} |
898 |
w.start(recordWorker(w), ueh); |
899 |
if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) |
900 |
break; // add at most one unless total below target |
901 |
} |
902 |
} |
903 |
if (eventWaiters != 0L) |
904 |
releaseEventWaiters(); |
905 |
} |
906 |
|
907 |
/** |
908 |
* Callback from the oldest waiter in awaitEvent waking up after a |
909 |
* period of non-use. If all workers are idle, tries (once) to |
910 |
* shutdown an event waiter or a spare, if one exists. Note that |
911 |
* we don't need CAS or locks here because the method is called |
912 |
* only from one thread occasionally waking (and even misfires are |
913 |
* OK). Note that until the shutdown worker fully terminates, |
914 |
* workerCounts will overestimate total count, which is tolerable. |
915 |
* |
916 |
* @param ec the event count waited on by caller (to abort |
917 |
* attempt if count has since changed). |
918 |
*/ |
919 |
private void tryShutdownUnusedWorker(int ec) { |
920 |
if (runState == 0 && eventCount == ec) { // only trigger if all idle |
921 |
ForkJoinWorkerThread[] ws = workers; |
922 |
int n = ws.length; |
923 |
ForkJoinWorkerThread w = null; |
924 |
boolean shutdown = false; |
925 |
int sw; |
926 |
long h; |
927 |
if ((sw = spareWaiters) != 0) { // prefer killing spares |
928 |
int id = (sw & SPARE_ID_MASK) - 1; |
929 |
if (id >= 0 && id < n && (w = ws[id]) != null && |
930 |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
931 |
sw, w.nextSpare)) |
932 |
shutdown = true; |
933 |
} |
934 |
else if ((h = eventWaiters) != 0L) { |
935 |
long nh; |
936 |
int id = (((int)h) & WAITER_ID_MASK) - 1; |
937 |
if (id >= 0 && id < n && (w = ws[id]) != null && |
938 |
(nh = w.nextWaiter) != 0L && // keep at least one worker |
939 |
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) |
940 |
shutdown = true; |
941 |
} |
942 |
if (w != null && shutdown) { |
943 |
w.shutdown(); |
944 |
LockSupport.unpark(w); |
945 |
} |
946 |
} |
947 |
releaseEventWaiters(); // in case of interference |
948 |
} |
949 |
|
950 |
/** |
951 |
* Callback from workers invoked upon each top-level action (i.e., |
952 |
* stealing a task or taking a submission and running it). |
953 |
* Performs one or more of the following: |
954 |
* |
955 |
* 1. If the worker is active and either did not run a task |
956 |
* or there are too many workers, try to set its active status |
957 |
* to inactive and update activeCount. On contention, we may |
958 |
* try again in this or a subsequent call. |
959 |
* |
960 |
* 2. If not enough total workers, help create some. |
961 |
* |
962 |
* 3. If there are too many running workers, suspend this worker |
963 |
* (first forcing inactive if necessary). If it is not needed, |
964 |
* it may be shutdown while suspended (via |
965 |
* tryShutdownUnusedWorker). Otherwise, upon resume it |
966 |
* rechecks running thread count and need for event sync. |
967 |
* |
968 |
* 4. If worker did not run a task, await the next task event via |
969 |
* eventSync if necessary (first forcing inactivation), upon |
970 |
* which the worker may be shutdown via |
971 |
* tryShutdownUnusedWorker. Otherwise, help release any |
972 |
* existing event waiters that are now releasable, |
973 |
* |
974 |
* @param w the worker |
975 |
* @param ran true if worker ran a task since last call to this method |
976 |
*/ |
977 |
final void preStep(ForkJoinWorkerThread w, boolean ran) { |
978 |
int wec = w.lastEventCount; |
979 |
boolean active = w.active; |
980 |
boolean inactivate = false; |
981 |
int pc = parallelism; |
982 |
while (w.runState == 0) { |
983 |
int rs = runState; |
984 |
if (rs >= TERMINATING) { // propagate shutdown |
985 |
w.shutdown(); |
986 |
break; |
987 |
} |
988 |
if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) && |
989 |
UNSAFE.compareAndSwapInt(this, runStateOffset, rs, --rs)) { |
990 |
inactivate = active = w.active = false; |
991 |
if (rs == SHUTDOWN) { // all inactive and shut down |
992 |
tryTerminate(false); |
993 |
continue; |
994 |
} |
995 |
} |
996 |
int wc = workerCounts; // try to suspend as spare |
997 |
if ((wc & RUNNING_COUNT_MASK) > pc) { |
998 |
if (!(inactivate |= active) && // must inactivate to suspend |
999 |
workerCounts == wc && |
1000 |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1001 |
wc, wc - ONE_RUNNING)) |
1002 |
w.suspendAsSpare(); |
1003 |
} |
1004 |
else if ((wc >>> TOTAL_COUNT_SHIFT) < pc) |
1005 |
helpMaintainParallelism(); // not enough workers |
1006 |
else if (ran) |
1007 |
break; |
1008 |
else { |
1009 |
long h = eventWaiters; |
1010 |
int ec = eventCount; |
1011 |
if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec) |
1012 |
releaseEventWaiters(); // release others before waiting |
1013 |
else if (ec != wec) { |
1014 |
w.lastEventCount = ec; // no need to wait |
1015 |
break; |
1016 |
} |
1017 |
else if (!(inactivate |= active)) |
1018 |
eventSync(w, wec); // must inactivate before sync |
1019 |
} |
1020 |
} |
1021 |
} |
1022 |
|
1023 |
/** |
1024 |
* Helps and/or blocks awaiting join of the given task. |
1025 |
* See above for explanation. |
1026 |
* |
1027 |
* @param joinMe the task to join |
1028 |
* @param worker the current worker thread |
1029 |
* @param timed true if wait should time out |
1030 |
* @param nanos timeout value if timed |
1031 |
*/ |
1032 |
final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker, |
1033 |
boolean timed, long nanos) { |
1034 |
long startTime = timed ? System.nanoTime() : 0L; |
1035 |
int retries = 2 + (parallelism >> 2); // #helpJoins before blocking |
1036 |
boolean running = true; // false when count decremented |
1037 |
while (joinMe.status >= 0) { |
1038 |
if (runState >= TERMINATING) { |
1039 |
joinMe.cancelIgnoringExceptions(); |
1040 |
break; |
1041 |
} |
1042 |
running = worker.helpJoinTask(joinMe, running); |
1043 |
if (joinMe.status < 0) |
1044 |
break; |
1045 |
if (retries > 0) { |
1046 |
--retries; |
1047 |
continue; |
1048 |
} |
1049 |
int wc = workerCounts; |
1050 |
if ((wc & RUNNING_COUNT_MASK) != 0) { |
1051 |
if (running) { |
1052 |
if (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1053 |
wc, wc - ONE_RUNNING)) |
1054 |
continue; |
1055 |
running = false; |
1056 |
} |
1057 |
long h = eventWaiters; |
1058 |
if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
1059 |
releaseEventWaiters(); |
1060 |
if ((workerCounts & RUNNING_COUNT_MASK) != 0) { |
1061 |
long ms; int ns; |
1062 |
if (!timed) { |
1063 |
ms = JOIN_TIMEOUT_MILLIS; |
1064 |
ns = 0; |
1065 |
} |
1066 |
else { // at most JOIN_TIMEOUT_MILLIS per wait |
1067 |
long nt = nanos - (System.nanoTime() - startTime); |
1068 |
if (nt <= 0L) |
1069 |
break; |
1070 |
ms = nt / 1000000; |
1071 |
if (ms > JOIN_TIMEOUT_MILLIS) { |
1072 |
ms = JOIN_TIMEOUT_MILLIS; |
1073 |
ns = 0; |
1074 |
} |
1075 |
else |
1076 |
ns = (int) (nt % 1000000); |
1077 |
} |
1078 |
joinMe.internalAwaitDone(ms, ns); |
1079 |
} |
1080 |
if (joinMe.status < 0) |
1081 |
break; |
1082 |
} |
1083 |
helpMaintainParallelism(); |
1084 |
} |
1085 |
if (!running) { |
1086 |
int c; |
1087 |
do {} while (!UNSAFE.compareAndSwapInt |
1088 |
(this, workerCountsOffset, |
1089 |
c = workerCounts, c + ONE_RUNNING)); |
1090 |
} |
1091 |
} |
1092 |
|
1093 |
/** |
1094 |
* Same idea as awaitJoin, but no helping, retries, or timeouts. |
1095 |
*/ |
1096 |
final void awaitBlocker(ManagedBlocker blocker) |
1097 |
throws InterruptedException { |
1098 |
while (!blocker.isReleasable()) { |
1099 |
int wc = workerCounts; |
1100 |
if ((wc & RUNNING_COUNT_MASK) == 0) |
1101 |
helpMaintainParallelism(); |
1102 |
else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1103 |
wc, wc - ONE_RUNNING)) { |
1104 |
try { |
1105 |
while (!blocker.isReleasable()) { |
1106 |
long h = eventWaiters; |
1107 |
if (h != 0L && |
1108 |
(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
1109 |
releaseEventWaiters(); |
1110 |
else if ((workerCounts & RUNNING_COUNT_MASK) == 0 && |
1111 |
runState < TERMINATING) |
1112 |
helpMaintainParallelism(); |
1113 |
else if (blocker.block()) |
1114 |
break; |
1115 |
} |
1116 |
} finally { |
1117 |
int c; |
1118 |
do {} while (!UNSAFE.compareAndSwapInt |
1119 |
(this, workerCountsOffset, |
1120 |
c = workerCounts, c + ONE_RUNNING)); |
1121 |
} |
1122 |
break; |
1123 |
} |
1124 |
} |
1125 |
} |
1126 |
|
1127 |
/** |
1128 |
* Possibly initiates and/or completes termination. |
1129 |
* |
1130 |
* @param now if true, unconditionally terminate, else only |
1131 |
* if shutdown and empty queue and no active workers |
1132 |
* @return true if now terminating or terminated |
1133 |
*/ |
1134 |
private boolean tryTerminate(boolean now) { |
1135 |
if (now) |
1136 |
advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN |
1137 |
else if (runState < SHUTDOWN || |
1138 |
!submissionQueue.isEmpty() || |
1139 |
(runState & ACTIVE_COUNT_MASK) != 0) |
1140 |
return false; |
1141 |
|
1142 |
if (advanceRunLevel(TERMINATING)) |
1143 |
startTerminating(); |
1144 |
|
1145 |
// Finish now if all threads terminated; else in some subsequent call |
1146 |
if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) { |
1147 |
advanceRunLevel(TERMINATED); |
1148 |
termination.forceTermination(); |
1149 |
} |
1150 |
return true; |
1151 |
} |
1152 |
|
1153 |
/** |
1154 |
* Actions on transition to TERMINATING |
1155 |
* |
1156 |
* Runs up to four passes through workers: (0) shutting down each |
1157 |
* (without waking up if parked) to quickly spread notifications |
1158 |
* without unnecessary bouncing around event queues etc (1) wake |
1159 |
* up and help cancel tasks (2) interrupt (3) mop up races with |
1160 |
* interrupted workers |
1161 |
*/ |
1162 |
private void startTerminating() { |
1163 |
cancelSubmissions(); |
1164 |
for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) { |
1165 |
int c; // advance event count |
1166 |
UNSAFE.compareAndSwapInt(this, eventCountOffset, |
1167 |
c = eventCount, c+1); |
1168 |
eventWaiters = 0L; // clobber lists |
1169 |
spareWaiters = 0; |
1170 |
for (ForkJoinWorkerThread w : workers) { |
1171 |
if (w != null) { |
1172 |
w.shutdown(); |
1173 |
if (passes > 0 && !w.isTerminated()) { |
1174 |
w.cancelTasks(); |
1175 |
LockSupport.unpark(w); |
1176 |
if (passes > 1 && !w.isInterrupted()) { |
1177 |
try { |
1178 |
w.interrupt(); |
1179 |
} catch (SecurityException ignore) { |
1180 |
} |
1181 |
} |
1182 |
} |
1183 |
} |
1184 |
} |
1185 |
} |
1186 |
} |
1187 |
|
1188 |
/** |
1189 |
* Clears out and cancels submissions, ignoring exceptions. |
1190 |
*/ |
1191 |
private void cancelSubmissions() { |
1192 |
ForkJoinTask<?> task; |
1193 |
while ((task = submissionQueue.poll()) != null) { |
1194 |
try { |
1195 |
task.cancel(false); |
1196 |
} catch (Throwable ignore) { |
1197 |
} |
1198 |
} |
1199 |
} |
1200 |
|
1201 |
// misc support for ForkJoinWorkerThread |
1202 |
|
1203 |
/** |
1204 |
* Returns pool number. |
1205 |
*/ |
1206 |
final int getPoolNumber() { |
1207 |
return poolNumber; |
1208 |
} |
1209 |
|
1210 |
/** |
1211 |
* Tries to accumulate steal count from a worker, clearing |
1212 |
* the worker's value if successful. |
1213 |
* |
1214 |
* @return true if worker steal count now zero |
1215 |
*/ |
1216 |
final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) { |
1217 |
int sc = w.stealCount; |
1218 |
long c = stealCount; |
1219 |
// CAS even if zero, for fence effects |
1220 |
if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) { |
1221 |
if (sc != 0) |
1222 |
w.stealCount = 0; |
1223 |
return true; |
1224 |
} |
1225 |
return sc == 0; |
1226 |
} |
1227 |
|
1228 |
/** |
1229 |
* Returns the approximate (non-atomic) number of idle threads per |
1230 |
* active thread. |
1231 |
*/ |
1232 |
final int idlePerActive() { |
1233 |
int pc = parallelism; // use parallelism, not rc |
1234 |
int ac = runState; // no mask -- artificially boosts during shutdown |
1235 |
// Use exact results for small values, saturate past 4 |
1236 |
return ((pc <= ac) ? 0 : |
1237 |
(pc >>> 1 <= ac) ? 1 : |
1238 |
(pc >>> 2 <= ac) ? 3 : |
1239 |
pc >>> 3); |
1240 |
} |
1241 |
|
1242 |
// Public and protected methods |
1243 |
|
1244 |
// Constructors |
1245 |
|
1246 |
/** |
1247 |
* Creates a {@code ForkJoinPool} with parallelism equal to {@link |
1248 |
* java.lang.Runtime#availableProcessors}, using the {@linkplain |
1249 |
* #defaultForkJoinWorkerThreadFactory default thread factory}, |
1250 |
* no UncaughtExceptionHandler, and non-async LIFO processing mode. |
1251 |
* |
1252 |
* @throws SecurityException if a security manager exists and |
1253 |
* the caller is not permitted to modify threads |
1254 |
* because it does not hold {@link |
1255 |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1256 |
*/ |
1257 |
public ForkJoinPool() { |
1258 |
this(Runtime.getRuntime().availableProcessors(), |
1259 |
defaultForkJoinWorkerThreadFactory, null, false); |
1260 |
} |
1261 |
|
1262 |
/** |
1263 |
* Creates a {@code ForkJoinPool} with the indicated parallelism |
1264 |
* level, the {@linkplain |
1265 |
* #defaultForkJoinWorkerThreadFactory default thread factory}, |
1266 |
* no UncaughtExceptionHandler, and non-async LIFO processing mode. |
1267 |
* |
1268 |
* @param parallelism the parallelism level |
1269 |
* @throws IllegalArgumentException if parallelism less than or |
1270 |
* equal to zero, or greater than implementation limit |
1271 |
* @throws SecurityException if a security manager exists and |
1272 |
* the caller is not permitted to modify threads |
1273 |
* because it does not hold {@link |
1274 |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1275 |
*/ |
1276 |
public ForkJoinPool(int parallelism) { |
1277 |
this(parallelism, defaultForkJoinWorkerThreadFactory, null, false); |
1278 |
} |
1279 |
|
1280 |
/** |
1281 |
* Creates a {@code ForkJoinPool} with the given parameters. |
1282 |
* |
1283 |
* @param parallelism the parallelism level. For default value, |
1284 |
* use {@link java.lang.Runtime#availableProcessors}. |
1285 |
* @param factory the factory for creating new threads. For default value, |
1286 |
* use {@link #defaultForkJoinWorkerThreadFactory}. |
1287 |
* @param handler the handler for internal worker threads that |
1288 |
* terminate due to unrecoverable errors encountered while executing |
1289 |
* tasks. For default value, use {@code null}. |
1290 |
* @param asyncMode if true, |
1291 |
* establishes local first-in-first-out scheduling mode for forked |
1292 |
* tasks that are never joined. This mode may be more appropriate |
1293 |
* than default locally stack-based mode in applications in which |
1294 |
* worker threads only process event-style asynchronous tasks. |
1295 |
* For default value, use {@code false}. |
1296 |
* @throws IllegalArgumentException if parallelism less than or |
1297 |
* equal to zero, or greater than implementation limit |
1298 |
* @throws NullPointerException if the factory is null |
1299 |
* @throws SecurityException if a security manager exists and |
1300 |
* the caller is not permitted to modify threads |
1301 |
* because it does not hold {@link |
1302 |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1303 |
*/ |
1304 |
public ForkJoinPool(int parallelism, |
1305 |
ForkJoinWorkerThreadFactory factory, |
1306 |
Thread.UncaughtExceptionHandler handler, |
1307 |
boolean asyncMode) { |
1308 |
checkPermission(); |
1309 |
if (factory == null) |
1310 |
throw new NullPointerException(); |
1311 |
if (parallelism <= 0 || parallelism > MAX_WORKERS) |
1312 |
throw new IllegalArgumentException(); |
1313 |
this.parallelism = parallelism; |
1314 |
this.factory = factory; |
1315 |
this.ueh = handler; |
1316 |
this.locallyFifo = asyncMode; |
1317 |
int arraySize = initialArraySizeFor(parallelism); |
1318 |
this.workers = new ForkJoinWorkerThread[arraySize]; |
1319 |
this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>(); |
1320 |
this.workerLock = new ReentrantLock(); |
1321 |
this.termination = new Phaser(1); |
1322 |
this.poolNumber = poolNumberGenerator.incrementAndGet(); |
1323 |
} |
1324 |
|
1325 |
/** |
1326 |
* Returns initial power of two size for workers array. |
1327 |
* @param pc the initial parallelism level |
1328 |
*/ |
1329 |
private static int initialArraySizeFor(int pc) { |
1330 |
// If possible, initially allocate enough space for one spare |
1331 |
int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS; |
1332 |
// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16) |
1333 |
size |= size >>> 1; |
1334 |
size |= size >>> 2; |
1335 |
size |= size >>> 4; |
1336 |
size |= size >>> 8; |
1337 |
return size + 1; |
1338 |
} |
1339 |
|
1340 |
// Execution methods |
1341 |
|
1342 |
/** |
1343 |
* Submits task and creates, starts, or resumes some workers if necessary |
1344 |
*/ |
1345 |
private <T> void doSubmit(ForkJoinTask<T> task) { |
1346 |
submissionQueue.offer(task); |
1347 |
int c; // try to increment event count -- CAS failure OK |
1348 |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); |
1349 |
helpMaintainParallelism(); |
1350 |
} |
1351 |
|
1352 |
/** |
1353 |
* Performs the given task, returning its result upon completion. |
1354 |
* |
1355 |
* @param task the task |
1356 |
* @return the task's result |
1357 |
* @throws NullPointerException if the task is null |
1358 |
* @throws RejectedExecutionException if the task cannot be |
1359 |
* scheduled for execution |
1360 |
*/ |
1361 |
public <T> T invoke(ForkJoinTask<T> task) { |
1362 |
if (task == null) |
1363 |
throw new NullPointerException(); |
1364 |
if (runState >= SHUTDOWN) |
1365 |
throw new RejectedExecutionException(); |
1366 |
Thread t = Thread.currentThread(); |
1367 |
if ((t instanceof ForkJoinWorkerThread) && |
1368 |
((ForkJoinWorkerThread)t).pool == this) |
1369 |
return task.invoke(); // bypass submit if in same pool |
1370 |
else { |
1371 |
doSubmit(task); |
1372 |
return task.join(); |
1373 |
} |
1374 |
} |
1375 |
|
1376 |
/** |
1377 |
* Unless terminating, forks task if within an ongoing FJ |
1378 |
* computation in the current pool, else submits as external task. |
1379 |
*/ |
1380 |
private <T> void forkOrSubmit(ForkJoinTask<T> task) { |
1381 |
if (runState >= SHUTDOWN) |
1382 |
throw new RejectedExecutionException(); |
1383 |
Thread t = Thread.currentThread(); |
1384 |
if ((t instanceof ForkJoinWorkerThread) && |
1385 |
((ForkJoinWorkerThread)t).pool == this) |
1386 |
task.fork(); |
1387 |
else |
1388 |
doSubmit(task); |
1389 |
} |
1390 |
|
1391 |
/** |
1392 |
* Arranges for (asynchronous) execution of the given task. |
1393 |
* |
1394 |
* @param task the task |
1395 |
* @throws NullPointerException if the task is null |
1396 |
* @throws RejectedExecutionException if the task cannot be |
1397 |
* scheduled for execution |
1398 |
*/ |
1399 |
public void execute(ForkJoinTask<?> task) { |
1400 |
if (task == null) |
1401 |
throw new NullPointerException(); |
1402 |
forkOrSubmit(task); |
1403 |
} |
1404 |
|
1405 |
// AbstractExecutorService methods |
1406 |
|
1407 |
/** |
1408 |
* @throws NullPointerException if the task is null |
1409 |
* @throws RejectedExecutionException if the task cannot be |
1410 |
* scheduled for execution |
1411 |
*/ |
1412 |
public void execute(Runnable task) { |
1413 |
if (task == null) |
1414 |
throw new NullPointerException(); |
1415 |
ForkJoinTask<?> job; |
1416 |
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1417 |
job = (ForkJoinTask<?>) task; |
1418 |
else |
1419 |
job = ForkJoinTask.adapt(task, null); |
1420 |
forkOrSubmit(job); |
1421 |
} |
1422 |
|
1423 |
/** |
1424 |
* Submits a ForkJoinTask for execution. |
1425 |
* |
1426 |
* @param task the task to submit |
1427 |
* @return the task |
1428 |
* @throws NullPointerException if the task is null |
1429 |
* @throws RejectedExecutionException if the task cannot be |
1430 |
* scheduled for execution |
1431 |
*/ |
1432 |
public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) { |
1433 |
if (task == null) |
1434 |
throw new NullPointerException(); |
1435 |
forkOrSubmit(task); |
1436 |
return task; |
1437 |
} |
1438 |
|
1439 |
/** |
1440 |
* @throws NullPointerException if the task is null |
1441 |
* @throws RejectedExecutionException if the task cannot be |
1442 |
* scheduled for execution |
1443 |
*/ |
1444 |
public <T> ForkJoinTask<T> submit(Callable<T> task) { |
1445 |
if (task == null) |
1446 |
throw new NullPointerException(); |
1447 |
ForkJoinTask<T> job = ForkJoinTask.adapt(task); |
1448 |
forkOrSubmit(job); |
1449 |
return job; |
1450 |
} |
1451 |
|
1452 |
/** |
1453 |
* @throws NullPointerException if the task is null |
1454 |
* @throws RejectedExecutionException if the task cannot be |
1455 |
* scheduled for execution |
1456 |
*/ |
1457 |
public <T> ForkJoinTask<T> submit(Runnable task, T result) { |
1458 |
if (task == null) |
1459 |
throw new NullPointerException(); |
1460 |
ForkJoinTask<T> job = ForkJoinTask.adapt(task, result); |
1461 |
forkOrSubmit(job); |
1462 |
return job; |
1463 |
} |
1464 |
|
1465 |
/** |
1466 |
* @throws NullPointerException if the task is null |
1467 |
* @throws RejectedExecutionException if the task cannot be |
1468 |
* scheduled for execution |
1469 |
*/ |
1470 |
public ForkJoinTask<?> submit(Runnable task) { |
1471 |
if (task == null) |
1472 |
throw new NullPointerException(); |
1473 |
ForkJoinTask<?> job; |
1474 |
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1475 |
job = (ForkJoinTask<?>) task; |
1476 |
else |
1477 |
job = ForkJoinTask.adapt(task, null); |
1478 |
forkOrSubmit(job); |
1479 |
return job; |
1480 |
} |
1481 |
|
1482 |
/** |
1483 |
* @throws NullPointerException {@inheritDoc} |
1484 |
* @throws RejectedExecutionException {@inheritDoc} |
1485 |
*/ |
1486 |
public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) { |
1487 |
ArrayList<ForkJoinTask<T>> forkJoinTasks = |
1488 |
new ArrayList<ForkJoinTask<T>>(tasks.size()); |
1489 |
for (Callable<T> task : tasks) |
1490 |
forkJoinTasks.add(ForkJoinTask.adapt(task)); |
1491 |
invoke(new InvokeAll<T>(forkJoinTasks)); |
1492 |
|
1493 |
@SuppressWarnings({"unchecked", "rawtypes"}) |
1494 |
List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks; |
1495 |
return futures; |
1496 |
} |
1497 |
|
1498 |
static final class InvokeAll<T> extends RecursiveAction { |
1499 |
final ArrayList<ForkJoinTask<T>> tasks; |
1500 |
InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; } |
1501 |
public void compute() { |
1502 |
try { invokeAll(tasks); } |
1503 |
catch (Exception ignore) {} |
1504 |
} |
1505 |
private static final long serialVersionUID = -7914297376763021607L; |
1506 |
} |
1507 |
|
1508 |
/** |
1509 |
* Returns the factory used for constructing new workers. |
1510 |
* |
1511 |
* @return the factory used for constructing new workers |
1512 |
*/ |
1513 |
public ForkJoinWorkerThreadFactory getFactory() { |
1514 |
return factory; |
1515 |
} |
1516 |
|
1517 |
/** |
1518 |
* Returns the handler for internal worker threads that terminate |
1519 |
* due to unrecoverable errors encountered while executing tasks. |
1520 |
* |
1521 |
* @return the handler, or {@code null} if none |
1522 |
*/ |
1523 |
public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() { |
1524 |
return ueh; |
1525 |
} |
1526 |
|
1527 |
/** |
1528 |
* Returns the targeted parallelism level of this pool. |
1529 |
* |
1530 |
* @return the targeted parallelism level of this pool |
1531 |
*/ |
1532 |
public int getParallelism() { |
1533 |
return parallelism; |
1534 |
} |
1535 |
|
1536 |
/** |
1537 |
* Returns the number of worker threads that have started but not |
1538 |
* yet terminated. The result returned by this method may differ |
1539 |
* from {@link #getParallelism} when threads are created to |
1540 |
* maintain parallelism when others are cooperatively blocked. |
1541 |
* |
1542 |
* @return the number of worker threads |
1543 |
*/ |
1544 |
public int getPoolSize() { |
1545 |
return workerCounts >>> TOTAL_COUNT_SHIFT; |
1546 |
} |
1547 |
|
1548 |
/** |
1549 |
* Returns {@code true} if this pool uses local first-in-first-out |
1550 |
* scheduling mode for forked tasks that are never joined. |
1551 |
* |
1552 |
* @return {@code true} if this pool uses async mode |
1553 |
*/ |
1554 |
public boolean getAsyncMode() { |
1555 |
return locallyFifo; |
1556 |
} |
1557 |
|
1558 |
/** |
1559 |
* Returns an estimate of the number of worker threads that are |
1560 |
* not blocked waiting to join tasks or for other managed |
1561 |
* synchronization. This method may overestimate the |
1562 |
* number of running threads. |
1563 |
* |
1564 |
* @return the number of worker threads |
1565 |
*/ |
1566 |
public int getRunningThreadCount() { |
1567 |
return workerCounts & RUNNING_COUNT_MASK; |
1568 |
} |
1569 |
|
1570 |
/** |
1571 |
* Returns an estimate of the number of threads that are currently |
1572 |
* stealing or executing tasks. This method may overestimate the |
1573 |
* number of active threads. |
1574 |
* |
1575 |
* @return the number of active threads |
1576 |
*/ |
1577 |
public int getActiveThreadCount() { |
1578 |
return runState & ACTIVE_COUNT_MASK; |
1579 |
} |
1580 |
|
1581 |
/** |
1582 |
* Returns {@code true} if all worker threads are currently idle. |
1583 |
* An idle worker is one that cannot obtain a task to execute |
1584 |
* because none are available to steal from other threads, and |
1585 |
* there are no pending submissions to the pool. This method is |
1586 |
* conservative; it might not return {@code true} immediately upon |
1587 |
* idleness of all threads, but will eventually become true if |
1588 |
* threads remain inactive. |
1589 |
* |
1590 |
* @return {@code true} if all threads are currently idle |
1591 |
*/ |
1592 |
public boolean isQuiescent() { |
1593 |
return (runState & ACTIVE_COUNT_MASK) == 0; |
1594 |
} |
1595 |
|
1596 |
/** |
1597 |
* Returns an estimate of the total number of tasks stolen from |
1598 |
* one thread's work queue by another. The reported value |
1599 |
* underestimates the actual total number of steals when the pool |
1600 |
* is not quiescent. This value may be useful for monitoring and |
1601 |
* tuning fork/join programs: in general, steal counts should be |
1602 |
* high enough to keep threads busy, but low enough to avoid |
1603 |
* overhead and contention across threads. |
1604 |
* |
1605 |
* @return the number of steals |
1606 |
*/ |
1607 |
public long getStealCount() { |
1608 |
return stealCount; |
1609 |
} |
1610 |
|
1611 |
/** |
1612 |
* Returns an estimate of the total number of tasks currently held |
1613 |
* in queues by worker threads (but not including tasks submitted |
1614 |
* to the pool that have not begun executing). This value is only |
1615 |
* an approximation, obtained by iterating across all threads in |
1616 |
* the pool. This method may be useful for tuning task |
1617 |
* granularities. |
1618 |
* |
1619 |
* @return the number of queued tasks |
1620 |
*/ |
1621 |
public long getQueuedTaskCount() { |
1622 |
long count = 0; |
1623 |
for (ForkJoinWorkerThread w : workers) |
1624 |
if (w != null) |
1625 |
count += w.getQueueSize(); |
1626 |
return count; |
1627 |
} |
1628 |
|
1629 |
/** |
1630 |
* Returns an estimate of the number of tasks submitted to this |
1631 |
* pool that have not yet begun executing. This method takes time |
1632 |
* proportional to the number of submissions. |
1633 |
* |
1634 |
* @return the number of queued submissions |
1635 |
*/ |
1636 |
public int getQueuedSubmissionCount() { |
1637 |
return submissionQueue.size(); |
1638 |
} |
1639 |
|
1640 |
/** |
1641 |
* Returns {@code true} if there are any tasks submitted to this |
1642 |
* pool that have not yet begun executing. |
1643 |
* |
1644 |
* @return {@code true} if there are any queued submissions |
1645 |
*/ |
1646 |
public boolean hasQueuedSubmissions() { |
1647 |
return !submissionQueue.isEmpty(); |
1648 |
} |
1649 |
|
1650 |
/** |
1651 |
* Removes and returns the next unexecuted submission if one is |
1652 |
* available. This method may be useful in extensions to this |
1653 |
* class that re-assign work in systems with multiple pools. |
1654 |
* |
1655 |
* @return the next submission, or {@code null} if none |
1656 |
*/ |
1657 |
protected ForkJoinTask<?> pollSubmission() { |
1658 |
return submissionQueue.poll(); |
1659 |
} |
1660 |
|
1661 |
/** |
1662 |
* Removes all available unexecuted submitted and forked tasks |
1663 |
* from scheduling queues and adds them to the given collection, |
1664 |
* without altering their execution status. These may include |
1665 |
* artificially generated or wrapped tasks. This method is |
1666 |
* designed to be invoked only when the pool is known to be |
1667 |
* quiescent. Invocations at other times may not remove all |
1668 |
* tasks. A failure encountered while attempting to add elements |
1669 |
* to collection {@code c} may result in elements being in |
1670 |
* neither, either or both collections when the associated |
1671 |
* exception is thrown. The behavior of this operation is |
1672 |
* undefined if the specified collection is modified while the |
1673 |
* operation is in progress. |
1674 |
* |
1675 |
* @param c the collection to transfer elements into |
1676 |
* @return the number of elements transferred |
1677 |
*/ |
1678 |
protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) { |
1679 |
int count = submissionQueue.drainTo(c); |
1680 |
for (ForkJoinWorkerThread w : workers) |
1681 |
if (w != null) |
1682 |
count += w.drainTasksTo(c); |
1683 |
return count; |
1684 |
} |
1685 |
|
1686 |
/** |
1687 |
* Returns a string identifying this pool, as well as its state, |
1688 |
* including indications of run state, parallelism level, and |
1689 |
* worker and task counts. |
1690 |
* |
1691 |
* @return a string identifying this pool, as well as its state |
1692 |
*/ |
1693 |
public String toString() { |
1694 |
long st = getStealCount(); |
1695 |
long qt = getQueuedTaskCount(); |
1696 |
long qs = getQueuedSubmissionCount(); |
1697 |
int wc = workerCounts; |
1698 |
int tc = wc >>> TOTAL_COUNT_SHIFT; |
1699 |
int rc = wc & RUNNING_COUNT_MASK; |
1700 |
int pc = parallelism; |
1701 |
int rs = runState; |
1702 |
int ac = rs & ACTIVE_COUNT_MASK; |
1703 |
return super.toString() + |
1704 |
"[" + runLevelToString(rs) + |
1705 |
", parallelism = " + pc + |
1706 |
", size = " + tc + |
1707 |
", active = " + ac + |
1708 |
", running = " + rc + |
1709 |
", steals = " + st + |
1710 |
", tasks = " + qt + |
1711 |
", submissions = " + qs + |
1712 |
"]"; |
1713 |
} |
1714 |
|
1715 |
private static String runLevelToString(int s) { |
1716 |
return ((s & TERMINATED) != 0 ? "Terminated" : |
1717 |
((s & TERMINATING) != 0 ? "Terminating" : |
1718 |
((s & SHUTDOWN) != 0 ? "Shutting down" : |
1719 |
"Running"))); |
1720 |
} |
1721 |
|
1722 |
/** |
1723 |
* Initiates an orderly shutdown in which previously submitted |
1724 |
* tasks are executed, but no new tasks will be accepted. |
1725 |
* Invocation has no additional effect if already shut down. |
1726 |
* Tasks that are in the process of being submitted concurrently |
1727 |
* during the course of this method may or may not be rejected. |
1728 |
* |
1729 |
* @throws SecurityException if a security manager exists and |
1730 |
* the caller is not permitted to modify threads |
1731 |
* because it does not hold {@link |
1732 |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1733 |
*/ |
1734 |
public void shutdown() { |
1735 |
checkPermission(); |
1736 |
advanceRunLevel(SHUTDOWN); |
1737 |
tryTerminate(false); |
1738 |
} |
1739 |
|
1740 |
/** |
1741 |
* Attempts to cancel and/or stop all tasks, and reject all |
1742 |
* subsequently submitted tasks. Tasks that are in the process of |
1743 |
* being submitted or executed concurrently during the course of |
1744 |
* this method may or may not be rejected. This method cancels |
1745 |
* both existing and unexecuted tasks, in order to permit |
1746 |
* termination in the presence of task dependencies. So the method |
1747 |
* always returns an empty list (unlike the case for some other |
1748 |
* Executors). |
1749 |
* |
1750 |
* @return an empty list |
1751 |
* @throws SecurityException if a security manager exists and |
1752 |
* the caller is not permitted to modify threads |
1753 |
* because it does not hold {@link |
1754 |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1755 |
*/ |
1756 |
public List<Runnable> shutdownNow() { |
1757 |
checkPermission(); |
1758 |
tryTerminate(true); |
1759 |
return Collections.emptyList(); |
1760 |
} |
1761 |
|
1762 |
/** |
1763 |
* Returns {@code true} if all tasks have completed following shut down. |
1764 |
* |
1765 |
* @return {@code true} if all tasks have completed following shut down |
1766 |
*/ |
1767 |
public boolean isTerminated() { |
1768 |
return runState >= TERMINATED; |
1769 |
} |
1770 |
|
1771 |
/** |
1772 |
* Returns {@code true} if the process of termination has |
1773 |
* commenced but not yet completed. This method may be useful for |
1774 |
* debugging. A return of {@code true} reported a sufficient |
1775 |
* period after shutdown may indicate that submitted tasks have |
1776 |
* ignored or suppressed interruption, or are waiting for IO, |
1777 |
* causing this executor not to properly terminate. (See the |
1778 |
* advisory notes for class {@link ForkJoinTask} stating that |
1779 |
* tasks should not normally entail blocking operations. But if |
1780 |
* they do, they must abort them on interrupt.) |
1781 |
* |
1782 |
* @return {@code true} if terminating but not yet terminated |
1783 |
*/ |
1784 |
public boolean isTerminating() { |
1785 |
return (runState & (TERMINATING|TERMINATED)) == TERMINATING; |
1786 |
} |
1787 |
|
1788 |
/** |
1789 |
* Returns true if terminating or terminated. Used by ForkJoinWorkerThread. |
1790 |
*/ |
1791 |
final boolean isAtLeastTerminating() { |
1792 |
return runState >= TERMINATING; |
1793 |
} |
1794 |
|
1795 |
/** |
1796 |
* Returns {@code true} if this pool has been shut down. |
1797 |
* |
1798 |
* @return {@code true} if this pool has been shut down |
1799 |
*/ |
1800 |
public boolean isShutdown() { |
1801 |
return runState >= SHUTDOWN; |
1802 |
} |
1803 |
|
1804 |
/** |
1805 |
* Blocks until all tasks have completed execution after a shutdown |
1806 |
* request, or the timeout occurs, or the current thread is |
1807 |
* interrupted, whichever happens first. |
1808 |
* |
1809 |
* @param timeout the maximum time to wait |
1810 |
* @param unit the time unit of the timeout argument |
1811 |
* @return {@code true} if this executor terminated and |
1812 |
* {@code false} if the timeout elapsed before termination |
1813 |
* @throws InterruptedException if interrupted while waiting |
1814 |
*/ |
1815 |
public boolean awaitTermination(long timeout, TimeUnit unit) |
1816 |
throws InterruptedException { |
1817 |
try { |
1818 |
termination.awaitAdvanceInterruptibly(0, timeout, unit); |
1819 |
} catch (TimeoutException ex) { |
1820 |
return false; |
1821 |
} |
1822 |
return true; |
1823 |
} |
1824 |
|
1825 |
/** |
1826 |
* Interface for extending managed parallelism for tasks running |
1827 |
* in {@link ForkJoinPool}s. |
1828 |
* |
1829 |
* <p>A {@code ManagedBlocker} provides two methods. Method |
1830 |
* {@code isReleasable} must return {@code true} if blocking is |
1831 |
* not necessary. Method {@code block} blocks the current thread |
1832 |
* if necessary (perhaps internally invoking {@code isReleasable} |
1833 |
* before actually blocking). The unusual methods in this API |
1834 |
* accommodate synchronizers that may, but don't usually, block |
1835 |
* for long periods. Similarly, they allow more efficient internal |
1836 |
* handling of cases in which additional workers may be, but |
1837 |
* usually are not, needed to ensure sufficient parallelism. |
1838 |
* Toward this end, implementations of method {@code isReleasable} |
1839 |
* must be amenable to repeated invocation. |
1840 |
* |
1841 |
* <p>For example, here is a ManagedBlocker based on a |
1842 |
* ReentrantLock: |
1843 |
* <pre> {@code |
1844 |
* class ManagedLocker implements ManagedBlocker { |
1845 |
* final ReentrantLock lock; |
1846 |
* boolean hasLock = false; |
1847 |
* ManagedLocker(ReentrantLock lock) { this.lock = lock; } |
1848 |
* public boolean block() { |
1849 |
* if (!hasLock) |
1850 |
* lock.lock(); |
1851 |
* return true; |
1852 |
* } |
1853 |
* public boolean isReleasable() { |
1854 |
* return hasLock || (hasLock = lock.tryLock()); |
1855 |
* } |
1856 |
* }}</pre> |
1857 |
* |
1858 |
* <p>Here is a class that possibly blocks waiting for an |
1859 |
* item on a given queue: |
1860 |
* <pre> {@code |
1861 |
* class QueueTaker<E> implements ManagedBlocker { |
1862 |
* final BlockingQueue<E> queue; |
1863 |
* volatile E item = null; |
1864 |
* QueueTaker(BlockingQueue<E> q) { this.queue = q; } |
1865 |
* public boolean block() throws InterruptedException { |
1866 |
* if (item == null) |
1867 |
* item = queue.take(); |
1868 |
* return true; |
1869 |
* } |
1870 |
* public boolean isReleasable() { |
1871 |
* return item != null || (item = queue.poll()) != null; |
1872 |
* } |
1873 |
* public E getItem() { // call after pool.managedBlock completes |
1874 |
* return item; |
1875 |
* } |
1876 |
* }}</pre> |
1877 |
*/ |
1878 |
public static interface ManagedBlocker { |
1879 |
/** |
1880 |
* Possibly blocks the current thread, for example waiting for |
1881 |
* a lock or condition. |
1882 |
* |
1883 |
* @return {@code true} if no additional blocking is necessary |
1884 |
* (i.e., if isReleasable would return true) |
1885 |
* @throws InterruptedException if interrupted while waiting |
1886 |
* (the method is not required to do so, but is allowed to) |
1887 |
*/ |
1888 |
boolean block() throws InterruptedException; |
1889 |
|
1890 |
/** |
1891 |
* Returns {@code true} if blocking is unnecessary. |
1892 |
*/ |
1893 |
boolean isReleasable(); |
1894 |
} |
1895 |
|
1896 |
/** |
1897 |
* Blocks in accord with the given blocker. If the current thread |
1898 |
* is a {@link ForkJoinWorkerThread}, this method possibly |
1899 |
* arranges for a spare thread to be activated if necessary to |
1900 |
* ensure sufficient parallelism while the current thread is blocked. |
1901 |
* |
1902 |
* <p>If the caller is not a {@link ForkJoinTask}, this method is |
1903 |
* behaviorally equivalent to |
1904 |
* <pre> {@code |
1905 |
* while (!blocker.isReleasable()) |
1906 |
* if (blocker.block()) |
1907 |
* return; |
1908 |
* }</pre> |
1909 |
* |
1910 |
* If the caller is a {@code ForkJoinTask}, then the pool may |
1911 |
* first be expanded to ensure parallelism, and later adjusted. |
1912 |
* |
1913 |
* @param blocker the blocker |
1914 |
* @throws InterruptedException if blocker.block did so |
1915 |
*/ |
1916 |
public static void managedBlock(ManagedBlocker blocker) |
1917 |
throws InterruptedException { |
1918 |
Thread t = Thread.currentThread(); |
1919 |
if (t instanceof ForkJoinWorkerThread) { |
1920 |
ForkJoinWorkerThread w = (ForkJoinWorkerThread) t; |
1921 |
w.pool.awaitBlocker(blocker); |
1922 |
} |
1923 |
else { |
1924 |
do {} while (!blocker.isReleasable() && !blocker.block()); |
1925 |
} |
1926 |
} |
1927 |
|
1928 |
// AbstractExecutorService overrides. These rely on undocumented |
1929 |
// fact that ForkJoinTask.adapt returns ForkJoinTasks that also |
1930 |
// implement RunnableFuture. |
1931 |
|
1932 |
protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) { |
1933 |
return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value); |
1934 |
} |
1935 |
|
1936 |
protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) { |
1937 |
return (RunnableFuture<T>) ForkJoinTask.adapt(callable); |
1938 |
} |
1939 |
|
1940 |
// Unsafe mechanics |
1941 |
|
1942 |
private static final sun.misc.Unsafe UNSAFE = getUnsafe(); |
1943 |
private static final long workerCountsOffset = |
1944 |
objectFieldOffset("workerCounts", ForkJoinPool.class); |
1945 |
private static final long runStateOffset = |
1946 |
objectFieldOffset("runState", ForkJoinPool.class); |
1947 |
private static final long eventCountOffset = |
1948 |
objectFieldOffset("eventCount", ForkJoinPool.class); |
1949 |
private static final long eventWaitersOffset = |
1950 |
objectFieldOffset("eventWaiters", ForkJoinPool.class); |
1951 |
private static final long stealCountOffset = |
1952 |
objectFieldOffset("stealCount", ForkJoinPool.class); |
1953 |
private static final long spareWaitersOffset = |
1954 |
objectFieldOffset("spareWaiters", ForkJoinPool.class); |
1955 |
|
1956 |
private static long objectFieldOffset(String field, Class<?> klazz) { |
1957 |
try { |
1958 |
return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field)); |
1959 |
} catch (NoSuchFieldException e) { |
1960 |
// Convert Exception to corresponding Error |
1961 |
NoSuchFieldError error = new NoSuchFieldError(field); |
1962 |
error.initCause(e); |
1963 |
throw error; |
1964 |
} |
1965 |
} |
1966 |
|
1967 |
/** |
1968 |
* Returns a sun.misc.Unsafe. Suitable for use in a 3rd party package. |
1969 |
* Replace with a simple call to Unsafe.getUnsafe when integrating |
1970 |
* into a jdk. |
1971 |
* |
1972 |
* @return a sun.misc.Unsafe |
1973 |
*/ |
1974 |
private static sun.misc.Unsafe getUnsafe() { |
1975 |
try { |
1976 |
return sun.misc.Unsafe.getUnsafe(); |
1977 |
} catch (SecurityException se) { |
1978 |
try { |
1979 |
return java.security.AccessController.doPrivileged |
1980 |
(new java.security |
1981 |
.PrivilegedExceptionAction<sun.misc.Unsafe>() { |
1982 |
public sun.misc.Unsafe run() throws Exception { |
1983 |
java.lang.reflect.Field f = sun.misc |
1984 |
.Unsafe.class.getDeclaredField("theUnsafe"); |
1985 |
f.setAccessible(true); |
1986 |
return (sun.misc.Unsafe) f.get(null); |
1987 |
}}); |
1988 |
} catch (java.security.PrivilegedActionException e) { |
1989 |
throw new RuntimeException("Could not initialize intrinsics", |
1990 |
e.getCause()); |
1991 |
} |
1992 |
} |
1993 |
} |
1994 |
} |