6 |
|
|
7 |
|
package jsr166y; |
8 |
|
|
9 |
– |
import java.util.concurrent.*; |
10 |
– |
|
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; |
18 |
– |
import java.util.concurrent.atomic.AtomicInteger; |
19 |
– |
import java.util.concurrent.CountDownLatch; |
25 |
|
|
26 |
|
/** |
27 |
|
* An {@link ExecutorService} for running {@link ForkJoinTask}s. |
74 |
|
* <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td> |
75 |
|
* </tr> |
76 |
|
* <tr> |
77 |
< |
* <td> <b>Arange async execution</td> |
77 |
> |
* <td> <b>Arrange async execution</td> |
78 |
|
* <td> {@link #execute(ForkJoinTask)}</td> |
79 |
|
* <td> {@link ForkJoinTask#fork}</td> |
80 |
|
* </tr> |
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. Becauae we are multiplexing many tasks on to a pool |
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 |
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 or |
165 |
> |
* method helpMaintainParallelism() may create or |
166 |
|
* re-activate a spare thread to compensate for blocked |
167 |
|
* joiners until they unblock. |
168 |
|
* |
169 |
< |
* Because the determining existence of conservatively safe |
170 |
< |
* helping targets, the availability of already-created spares, |
171 |
< |
* and the apparent need to create new spares are all racy and |
172 |
< |
* require heuristic guidance, we rely on multiple retries of |
173 |
< |
* each. Further, because it is impossible to keep exactly the |
174 |
< |
* target (parallelism) number of threads running at any given |
175 |
< |
* time, we allow compensation during joins to fail, and enlist |
176 |
< |
* all other threads to help out whenever they are not otherwise |
177 |
< |
* occupied (i.e., mainly in method preStep). |
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. |
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 artifically suspended) as well as |
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 |
< |
* correspondance of these counts to reality is not guaranteed. In |
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 |
|
* |
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 others (but give up upon |
282 |
< |
* contention to reduce useless flailing). The net effect is a |
283 |
< |
* tree-like diffusion of signals, where released threads (and |
284 |
< |
* possibly others) help with unparks. To further reduce |
285 |
< |
* contention effects a bit, failed CASes to increment field |
279 |
< |
* eventCount are tolerated without retries in signalWork. |
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 is about |
290 |
< |
* to block waiting for a join (or via ManagedBlockers), we may |
291 |
< |
* create a new thread to maintain parallelism level, or at least |
292 |
< |
* avoid starvation. Usually, extra threads are needed for only |
293 |
< |
* very short periods, yet join dependencies are such that we |
294 |
< |
* sometimes need them in bursts. Rather than create new threads |
295 |
< |
* each time this happens, we suspend no-longer-needed extra ones |
296 |
< |
* as "spares". For most purposes, we don't distinguish "extra" |
297 |
< |
* spare threads from normal "core" threads: On each call to |
298 |
< |
* preStep (the only point at which we can do this) a worker |
299 |
< |
* checks to see if there are now too many running workers, and if |
300 |
< |
* so, suspends itself. Method helpMaintainParallelism looks for |
301 |
< |
* suspended threads to resume before considering creating a new |
302 |
< |
* replacement. The spares themselves are encoded on another |
303 |
< |
* variant of a Treiber Stack, headed at field "spareWaiters". |
304 |
< |
* Note that the use of spares is intrinsically racy. One thread |
305 |
< |
* may become a spare at about the same time as another is |
306 |
< |
* needlessly being created. We counteract this and related slop |
307 |
< |
* in part by requiring resumed spares to immediately recheck (in |
308 |
< |
* preStep) to see whether they they should re-suspend. |
309 |
< |
* |
310 |
< |
* 6. Killing off unneeded workers. The Spare and Event queues use |
311 |
< |
* similar mechanisms to shed unused workers: The oldest (first) |
312 |
< |
* waiter uses a timed rather than hard wait. When this wait times |
313 |
< |
* out without a normal wakeup, it tries to shutdown any one (for |
314 |
< |
* convenience the newest) other waiter via tryShutdownSpare or |
315 |
< |
* tryShutdownWaiter, respectively. The wakeup rates for spares |
316 |
< |
* are much shorter than for waiters. Together, they will |
317 |
< |
* eventually reduce the number of worker threads to a minimum of |
318 |
< |
* one after a long enough period without use. |
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 impossble |
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 the |
327 |
< |
* presence of transients phenomena such as GC stalls, dynamic |
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 some slack in |
332 |
< |
* creation thresholds, using rules that reflect the fact that the |
333 |
< |
* more threads we have running, the more likely that we are |
334 |
< |
* underestimating the number running threads. (We also include |
335 |
< |
* some heuristic use of Thread.yield when all workers appear to |
336 |
< |
* be busy, to improve likelihood of counts settling.) The rules |
337 |
< |
* also better cope with the fact that some of the methods in this |
338 |
< |
* class tend to never become compiled (but are interpreted), so |
339 |
< |
* some components of the entire set of controls might execute 100 |
334 |
< |
* times faster than others. And similarly for cases where the |
335 |
< |
* apparent lack of work is just due to GC stalls and other |
336 |
< |
* transient system activity. |
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 |
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 |
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 |
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 |
< |
* worker waiting for an event invokes tryShutdownWaiter to shrink |
439 |
< |
* the number of workers. The exact value does not matter too |
440 |
< |
* much, but should be long enough to slowly release resources |
441 |
< |
* during long periods without use without disrupting normal use. |
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 |
< |
60L * 1000L * 1000L * 1000L; // one minute |
445 |
> |
30L * 1000L * 1000L * 1000L; // 2 per minute |
446 |
|
|
447 |
|
/** |
448 |
|
* Absolute bound for parallelism level. Twice this number plus |
489 |
|
private volatile long stealCount; |
490 |
|
|
491 |
|
/** |
492 |
< |
* Encoded record of top of treiber stack of threads waiting for |
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 long WAITER_ID_MASK = (1L << 16) - 1L; |
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: |
508 |
|
private volatile int eventCount; |
509 |
|
|
510 |
|
/** |
511 |
< |
* Encoded record of top of treiber stack of spare threads waiting |
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. |
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 worker has no tasks and cannot find any. Bits 16-18 |
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. |
582 |
|
*/ |
583 |
|
private final int poolNumber; |
584 |
|
|
570 |
– |
|
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 |
589 |
> |
* Increments running count part of workerCounts. |
590 |
|
*/ |
591 |
|
final void incrementRunningCount() { |
592 |
|
int c; |
596 |
|
} |
597 |
|
|
598 |
|
/** |
599 |
< |
* Tries to decrement running count unless already zero |
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; |
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 |
624 |
> |
* @param dt -- either zero or ONE_TOTAL |
625 |
|
*/ |
626 |
|
private void decrementWorkerCounts(int dr, int dt) { |
627 |
|
for (;;) { |
639 |
|
} |
640 |
|
|
641 |
|
/** |
618 |
– |
* Increments event count |
619 |
– |
*/ |
620 |
– |
private void advanceEventCount() { |
621 |
– |
int c; |
622 |
– |
do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset, |
623 |
– |
c = eventCount, c+1)); |
624 |
– |
} |
625 |
– |
|
626 |
– |
/** |
627 |
– |
* Tries incrementing active count; fails on contention. |
628 |
– |
* Called by workers before executing tasks. |
629 |
– |
* |
630 |
– |
* @return true on success |
631 |
– |
*/ |
632 |
– |
final boolean tryIncrementActiveCount() { |
633 |
– |
int c; |
634 |
– |
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
635 |
– |
c = runState, c + 1); |
636 |
– |
} |
637 |
– |
|
638 |
– |
/** |
642 |
|
* Tries decrementing active count; fails on contention. |
643 |
|
* Called when workers cannot find tasks to run. |
644 |
|
*/ |
679 |
|
for (k = 0; k < n && ws[k] != null; ++k) |
680 |
|
; |
681 |
|
if (k == n) |
682 |
< |
ws = Arrays.copyOf(ws, n << 1); |
682 |
> |
ws = workers = Arrays.copyOf(ws, n << 1); |
683 |
|
} |
684 |
|
ws[k] = w; |
685 |
< |
workers = ws; // volatile array write ensures slot visibility |
685 |
> |
int c = eventCount; // advance event count to ensure visibility |
686 |
> |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1); |
687 |
|
} finally { |
688 |
|
lock.unlock(); |
689 |
|
} |
691 |
|
} |
692 |
|
|
693 |
|
/** |
694 |
< |
* Nulls out record of worker in workers array |
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 unecessary expansion |
698 |
> |
// Locking helps method recordWorker avoid unnecessary expansion |
699 |
|
final ReentrantLock lock = this.workerLock; |
700 |
|
lock.lock(); |
701 |
|
try { |
707 |
|
} |
708 |
|
} |
709 |
|
|
706 |
– |
// adding and removing workers |
707 |
– |
|
708 |
– |
/** |
709 |
– |
* Tries to create and add new worker. Assumes that worker counts |
710 |
– |
* are already updated to accommodate the worker, so adjusts on |
711 |
– |
* failure. |
712 |
– |
* |
713 |
– |
* @return the worker, or null on failure |
714 |
– |
*/ |
715 |
– |
private ForkJoinWorkerThread addWorker() { |
716 |
– |
ForkJoinWorkerThread w = null; |
717 |
– |
try { |
718 |
– |
w = factory.newThread(this); |
719 |
– |
} finally { // Adjust on either null or exceptional factory return |
720 |
– |
if (w == null) { |
721 |
– |
decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL); |
722 |
– |
tryTerminate(false); // in case of failure during shutdown |
723 |
– |
} |
724 |
– |
} |
725 |
– |
if (w != null) { |
726 |
– |
w.start(recordWorker(w), ueh); |
727 |
– |
advanceEventCount(); |
728 |
– |
} |
729 |
– |
return w; |
730 |
– |
} |
731 |
– |
|
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 terminatation. |
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); |
719 |
> |
decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL); |
720 |
|
while (w.stealCount != 0) // collect final count |
721 |
|
tryAccumulateStealCount(w); |
722 |
|
tryTerminate(false); |
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 |
< |
* contention, letting other workers take over. |
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 ((int)(h >>> EVENT_COUNT_SHIFT) != ec && |
741 |
< |
(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 && |
742 |
< |
id < n && (w = ws[id]) != null && |
743 |
< |
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
744 |
< |
h, h = w.nextWaiter)) { |
745 |
< |
LockSupport.unpark(w); |
746 |
< |
if (eventWaiters != h || eventCount != ec) |
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 |
|
|
765 |
|
|
766 |
|
/** |
767 |
|
* Adds the given worker to event queue and blocks until |
768 |
< |
* terminating or event count advances from the workers |
786 |
< |
* lastEventCount value |
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) { |
791 |
< |
int ec = w.lastEventCount; |
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 || |
777 |
> |
(((int)(h = eventWaiters) & WAITER_ID_MASK) == 0 || |
778 |
|
(int)(h >>> EVENT_COUNT_SHIFT) == ec) && |
779 |
|
eventCount == ec) { |
780 |
|
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
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 |
< |
* minumum of one) if the pool has not been used for extended |
793 |
> |
* minimum of one) if the pool has not been used for extended |
794 |
|
* periods. |
795 |
|
* |
796 |
|
* @param w the calling worker thread |
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(); |
804 |
> |
long startTime = untimed ? 0 : System.nanoTime(); |
805 |
|
Thread.interrupted(); // clear/ignore interrupt |
806 |
< |
if (eventCount != ec || !w.isRunning() || |
807 |
< |
runState >= TERMINATING) // recheck after clear |
826 |
< |
break; |
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.isRunning() || |
832 |
< |
runState >= TERMINATING) |
812 |
> |
if (eventCount != ec || w.isTerminating()) |
813 |
|
break; |
814 |
|
if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS) |
815 |
< |
tryShutdownWaiter(ec); |
815 |
> |
tryShutdownUnusedWorker(ec); |
816 |
|
} |
817 |
|
} |
818 |
|
} |
819 |
|
} |
820 |
|
|
821 |
< |
/** |
842 |
< |
* Callback from the oldest waiter in awaitEvent waking up after a |
843 |
< |
* period of non-use. Tries (once) to shutdown an event waiter (or |
844 |
< |
* a spare, if one exists). Note that we don't need CAS or locks |
845 |
< |
* here because the method is called only from one thread |
846 |
< |
* occasionally waking (and even misfires are OK). Note that |
847 |
< |
* until the shutdown worker fully terminates, workerCounts |
848 |
< |
* will overestimate total count, which is tolerable. |
849 |
< |
* |
850 |
< |
* @param ec the event count waited on by caller (to abort |
851 |
< |
* attempt if count has since changed). |
852 |
< |
*/ |
853 |
< |
private void tryShutdownWaiter(int ec) { |
854 |
< |
if (spareWaiters != 0) { // prefer killing spares |
855 |
< |
tryShutdownSpare(); |
856 |
< |
return; |
857 |
< |
} |
858 |
< |
ForkJoinWorkerThread[] ws = workers; |
859 |
< |
int n = ws.length; |
860 |
< |
long h = eventWaiters; |
861 |
< |
ForkJoinWorkerThread w; int id; long nh; |
862 |
< |
if (runState == 0 && |
863 |
< |
submissionQueue.isEmpty() && |
864 |
< |
eventCount == ec && |
865 |
< |
(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 && |
866 |
< |
id < n && (w = ws[id]) != null && |
867 |
< |
(nh = w.nextWaiter) != 0L && // keep at least one worker |
868 |
< |
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) { |
869 |
< |
w.shutdown(); |
870 |
< |
LockSupport.unpark(w); |
871 |
< |
} |
872 |
< |
releaseEventWaiters(); |
873 |
< |
} |
874 |
< |
|
875 |
< |
// Maintaining spares |
821 |
> |
// Maintaining parallelism |
822 |
|
|
823 |
|
/** |
824 |
< |
* Pushes worker onto the spare stack |
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); |
830 |
|
} |
831 |
|
|
832 |
|
/** |
833 |
< |
* Callback from oldest spare occasionally waking up. Tries |
834 |
< |
* (once) to shutdown a spare. Same idea as tryShutdownWaiter. |
833 |
> |
* Tries (once) to resume a spare if the number of running |
834 |
> |
* threads is less than target. |
835 |
|
*/ |
836 |
< |
final void tryShutdownSpare() { |
836 |
> |
private void tryResumeSpare() { |
837 |
|
int sw, id; |
892 |
– |
ForkJoinWorkerThread w; |
893 |
– |
ForkJoinWorkerThread[] ws; |
894 |
– |
if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 && |
895 |
– |
id < (ws = workers).length && (w = ws[id]) != null && |
896 |
– |
(workerCounts & RUNNING_COUNT_MASK) >= parallelism && |
897 |
– |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
898 |
– |
sw, w.nextSpare)) { |
899 |
– |
w.shutdown(); |
900 |
– |
LockSupport.unpark(w); |
901 |
– |
advanceEventCount(); |
902 |
– |
} |
903 |
– |
} |
904 |
– |
|
905 |
– |
/** |
906 |
– |
* Tries (once) to resume a spare if worker counts match |
907 |
– |
* the given count. |
908 |
– |
* |
909 |
– |
* @param wc workerCounts value on invocation of this method |
910 |
– |
*/ |
911 |
– |
private void tryResumeSpare(int wc) { |
838 |
|
ForkJoinWorkerThread[] ws = workers; |
839 |
|
int n = ws.length; |
840 |
< |
int sw, id, rs; ForkJoinWorkerThread w; |
841 |
< |
if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 && |
840 |
> |
ForkJoinWorkerThread w; |
841 |
> |
if ((sw = spareWaiters) != 0 && |
842 |
> |
(id = (sw & SPARE_ID_MASK) - 1) >= 0 && |
843 |
|
id < n && (w = ws[id]) != null && |
844 |
< |
(rs = runState) < TERMINATING && |
845 |
< |
eventWaiters == 0L && workerCounts == wc) { |
846 |
< |
// In case all workers busy, heuristically back off to let settle |
847 |
< |
Thread.yield(); |
848 |
< |
if (eventWaiters == 0L && runState == rs && // recheck |
849 |
< |
workerCounts == wc && spareWaiters == sw && |
850 |
< |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
851 |
< |
sw, w.nextSpare)) { |
852 |
< |
int c; // increment running count before resume |
853 |
< |
do {} while(!UNSAFE.compareAndSwapInt |
854 |
< |
(this, workerCountsOffset, |
855 |
< |
c = workerCounts, c + ONE_RUNNING)); |
856 |
< |
if (w.tryUnsuspend()) |
930 |
< |
LockSupport.unpark(w); |
931 |
< |
else // back out if w was shutdown |
932 |
< |
decrementWorkerCounts(ONE_RUNNING, 0); |
933 |
< |
} |
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 |
|
|
937 |
– |
// adding workers on demand |
938 |
– |
|
860 |
|
/** |
861 |
< |
* Adds one or more workers if needed to establish target parallelism. |
862 |
< |
* Retries upon contention. |
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 addWorkerIfBelowTarget() { |
867 |
> |
private void helpMaintainParallelism() { |
868 |
|
int pc = parallelism; |
869 |
< |
int wc; |
870 |
< |
while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < pc && |
871 |
< |
runState < TERMINATING) { |
872 |
< |
if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
873 |
< |
wc + (ONE_RUNNING|ONE_TOTAL))) { |
874 |
< |
if (addWorker() == null) |
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 |
< |
* Tries (once) to add a new worker if all existing workers are |
909 |
< |
* busy, and there are either no running workers or the deficit is |
910 |
< |
* at least twice the surplus. |
911 |
< |
* |
912 |
< |
* @param wc workerCounts value on invocation of this method |
913 |
< |
*/ |
914 |
< |
private void tryAddWorkerIfBusy(int wc) { |
964 |
< |
int tc, rc, rs; |
965 |
< |
int pc = parallelism; |
966 |
< |
if ((tc = wc >>> TOTAL_COUNT_SHIFT) < MAX_WORKERS && |
967 |
< |
((rc = wc & RUNNING_COUNT_MASK) == 0 || |
968 |
< |
rc < pc - ((tc - pc) << 1)) && |
969 |
< |
(rs = runState) < TERMINATING && |
970 |
< |
(rs & ACTIVE_COUNT_MASK) == tc) { |
971 |
< |
// Since all workers busy, heuristically back off to let settle |
972 |
< |
Thread.yield(); |
973 |
< |
if (eventWaiters == 0L && spareWaiters == 0 && // recheck |
974 |
< |
runState == rs && workerCounts == wc && |
975 |
< |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
976 |
< |
wc + (ONE_RUNNING|ONE_TOTAL))) |
977 |
< |
addWorker(); |
978 |
< |
} |
979 |
< |
} |
980 |
< |
|
981 |
< |
/** |
982 |
< |
* Does at most one of: |
983 |
< |
* |
984 |
< |
* 1. Help wake up existing workers waiting for work via |
985 |
< |
* releaseEventWaiters. (If any exist, then it doesn't |
986 |
< |
* matter right now if under target parallelism level.) |
987 |
< |
* |
988 |
< |
* 2. If a spare exists, try (once) to resume it via tryResumeSpare. |
989 |
< |
* |
990 |
< |
* 3. If there are not enough total workers, add some |
991 |
< |
* via addWorkerIfBelowTarget; |
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 |
< |
* 4. Try (once) to add a new worker if all existing workers |
917 |
< |
* are busy, via tryAddWorkerIfBusy |
916 |
> |
* @param ec the event count waited on by caller (to abort |
917 |
> |
* attempt if count has since changed). |
918 |
|
*/ |
919 |
< |
private void helpMaintainParallelism() { |
920 |
< |
long h; int pc, wc; |
921 |
< |
if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0) { |
922 |
< |
if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
923 |
< |
releaseEventWaiters(); // avoid useless call |
924 |
< |
} |
925 |
< |
else if ((pc = parallelism) > |
926 |
< |
((wc = workerCounts) & RUNNING_COUNT_MASK)) { |
927 |
< |
if (spareWaiters != 0) |
928 |
< |
tryResumeSpare(wc); |
929 |
< |
else if ((wc >>> TOTAL_COUNT_SHIFT) < pc) |
930 |
< |
addWorkerIfBelowTarget(); |
931 |
< |
else |
932 |
< |
tryAddWorkerIfBusy(wc); |
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 |
|
/** |
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, try to set its active status to |
956 |
< |
* inactive and update activeCount. On contention, we may try |
957 |
< |
* again on this or subsequent call. |
958 |
< |
* |
959 |
< |
* 2. Release any existing event waiters that are now relesable |
960 |
< |
* |
961 |
< |
* 3. If there are too many running threads, suspend this worker |
962 |
< |
* (first forcing inactive if necessary). If it is not |
963 |
< |
* needed, it may be killed while suspended via |
964 |
< |
* tryShutdownSpare. Otherwise, upon resume it rechecks to make |
965 |
< |
* sure that it is still needed. |
966 |
< |
* |
967 |
< |
* 4. If more than 1 miss, await the next task event via |
968 |
< |
* eventSync (first forcing inactivation if necessary), upon |
969 |
< |
* which worker may also be killed, via tryShutdownWaiter. |
970 |
< |
* |
971 |
< |
* 5. Help reactivate other workers via helpMaintainParallelism |
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 misses the number of scans by caller failing to find work |
1038 |
< |
* (saturating at 2 to avoid wraparound) |
975 |
> |
* @param ran true if worker ran a task since last call to this method |
976 |
|
*/ |
977 |
< |
final void preStep(ForkJoinWorkerThread w, int misses) { |
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 |
< |
for (;;) { |
983 |
< |
int rs, wc, rc, ec; long h; |
984 |
< |
if (active && UNSAFE.compareAndSwapInt(this, runStateOffset, |
985 |
< |
rs = runState, rs - 1)) |
986 |
< |
active = w.active = false; |
987 |
< |
if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0 && |
988 |
< |
(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) { |
989 |
< |
releaseEventWaiters(); |
990 |
< |
if (misses > 1) |
991 |
< |
continue; // clear before sync below |
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 |
< |
if ((rc = ((wc = workerCounts) & RUNNING_COUNT_MASK)) > pc) { |
997 |
< |
if (!active && // must inactivate to suspend |
998 |
< |
workerCounts == wc && // try to suspend as spare |
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)) { |
1001 |
> |
wc, wc - ONE_RUNNING)) |
1002 |
|
w.suspendAsSpare(); |
1060 |
– |
if (!w.isRunning()) |
1061 |
– |
break; // was killed while spare |
1062 |
– |
} |
1063 |
– |
continue; |
1003 |
|
} |
1004 |
< |
if (misses > 0) { |
1005 |
< |
if ((ec = eventCount) == w.lastEventCount && misses > 1) { |
1006 |
< |
if (!active) { // must inactivate to sync |
1007 |
< |
eventSync(w); |
1008 |
< |
if (w.isRunning()) |
1009 |
< |
misses = 1; // don't re-sync |
1010 |
< |
else |
1011 |
< |
break; // was killed while waiting |
1012 |
< |
} |
1013 |
< |
continue; |
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 |
< |
w.lastEventCount = ec; |
1017 |
> |
else if (!(inactivate |= active)) |
1018 |
> |
eventSync(w, wec); // must inactivate before sync |
1019 |
|
} |
1078 |
– |
if (rc < pc) |
1079 |
– |
helpMaintainParallelism(); |
1080 |
– |
break; |
1020 |
|
} |
1021 |
|
} |
1022 |
|
|
1023 |
|
/** |
1024 |
|
* Helps and/or blocks awaiting join of the given task. |
1025 |
< |
* Alternates between helpJoinTask() and helpMaintainParallelism() |
1087 |
< |
* as many times as there is a deficit in running count (or longer |
1088 |
< |
* if running count would become zero), then blocks if task still |
1089 |
< |
* not done. |
1025 |
> |
* See above for explanation. |
1026 |
|
* |
1027 |
|
* @param joinMe the task to join |
1028 |
< |
*/ |
1029 |
< |
final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) { |
1030 |
< |
int threshold = parallelism; // descend blocking thresholds |
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 |
< |
boolean block; int wc; |
1039 |
< |
worker.helpJoinTask(joinMe); |
1098 |
< |
if (joinMe.status < 0) |
1038 |
> |
if (runState >= TERMINATING) { |
1039 |
> |
joinMe.cancelIgnoringExceptions(); |
1040 |
|
break; |
1100 |
– |
if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) { |
1101 |
– |
if (threshold > 0) |
1102 |
– |
--threshold; |
1103 |
– |
else |
1104 |
– |
advanceEventCount(); // force release |
1105 |
– |
block = false; |
1041 |
|
} |
1042 |
< |
else |
1043 |
< |
block = UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1109 |
< |
wc, wc - ONE_RUNNING); |
1110 |
< |
helpMaintainParallelism(); |
1111 |
< |
if (block) { |
1112 |
< |
int c; |
1113 |
< |
joinMe.internalAwaitDone(); |
1114 |
< |
do {} while (!UNSAFE.compareAndSwapInt |
1115 |
< |
(this, workerCountsOffset, |
1116 |
< |
c = workerCounts, c + ONE_RUNNING)); |
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 |
1094 |
> |
* Same idea as awaitJoin, but no helping, retries, or timeouts. |
1095 |
|
*/ |
1096 |
|
final void awaitBlocker(ManagedBlocker blocker) |
1097 |
|
throws InterruptedException { |
1127 |
– |
int threshold = parallelism; |
1098 |
|
while (!blocker.isReleasable()) { |
1099 |
< |
boolean block; int wc; |
1100 |
< |
if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) { |
1101 |
< |
if (threshold > 0) |
1102 |
< |
--threshold; |
1103 |
< |
else |
1134 |
< |
advanceEventCount(); |
1135 |
< |
block = false; |
1136 |
< |
} |
1137 |
< |
else |
1138 |
< |
block = UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1139 |
< |
wc, wc - ONE_RUNNING); |
1140 |
< |
helpMaintainParallelism(); |
1141 |
< |
if (block) { |
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 |
< |
do {} while (!blocker.isReleasable() && !blocker.block()); |
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 |
1145 |
|
// Finish now if all threads terminated; else in some subsequent call |
1146 |
|
if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) { |
1147 |
|
advanceRunLevel(TERMINATED); |
1148 |
< |
termination.arrive(); |
1148 |
> |
termination.forceTermination(); |
1149 |
|
} |
1150 |
|
return true; |
1151 |
|
} |
1162 |
|
private void startTerminating() { |
1163 |
|
cancelSubmissions(); |
1164 |
|
for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) { |
1165 |
< |
advanceEventCount(); |
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 |
< |
ForkJoinWorkerThread[] ws = workers; |
1197 |
< |
int n = ws.length; |
1198 |
< |
for (int i = 0; i < n; ++i) { |
1199 |
< |
ForkJoinWorkerThread w = ws[i]; |
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) { |
1176 |
> |
if (passes > 1 && !w.isInterrupted()) { |
1177 |
|
try { |
1178 |
|
w.interrupt(); |
1179 |
|
} catch (SecurityException ignore) { |
1186 |
|
} |
1187 |
|
|
1188 |
|
/** |
1189 |
< |
* Clear out and cancel submissions, ignoring exceptions |
1189 |
> |
* Clears out and cancels submissions, ignoring exceptions. |
1190 |
|
*/ |
1191 |
|
private void cancelSubmissions() { |
1192 |
|
ForkJoinTask<?> task; |
1201 |
|
// misc support for ForkJoinWorkerThread |
1202 |
|
|
1203 |
|
/** |
1204 |
< |
* Returns pool number |
1204 |
> |
* Returns pool number. |
1205 |
|
*/ |
1206 |
|
final int getPoolNumber() { |
1207 |
|
return poolNumber; |
1208 |
|
} |
1209 |
|
|
1210 |
|
/** |
1211 |
< |
* Tries to accumulates steal count from a worker, clearing |
1212 |
< |
* the worker's value. |
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 |
|
*/ |
1231 |
|
*/ |
1232 |
|
final int idlePerActive() { |
1233 |
|
int pc = parallelism; // use parallelism, not rc |
1234 |
< |
int ac = runState; // no mask -- artifically boosts during shutdown |
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 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3; |
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 |
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</code>. |
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</code>. |
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 |
1327 |
|
* @param pc the initial parallelism level |
1328 |
|
*/ |
1329 |
|
private static int initialArraySizeFor(int pc) { |
1330 |
< |
// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16) |
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; |
1340 |
|
// Execution methods |
1341 |
|
|
1342 |
|
/** |
1343 |
< |
* Common code for execute, invoke and submit |
1343 |
> |
* Submits task and creates, starts, or resumes some workers if necessary |
1344 |
|
*/ |
1345 |
|
private <T> void doSubmit(ForkJoinTask<T> task) { |
1371 |
– |
if (task == null) |
1372 |
– |
throw new NullPointerException(); |
1373 |
– |
if (runState >= SHUTDOWN) |
1374 |
– |
throw new RejectedExecutionException(); |
1346 |
|
submissionQueue.offer(task); |
1347 |
< |
advanceEventCount(); |
1348 |
< |
if (eventWaiters != 0L) |
1349 |
< |
releaseEventWaiters(); |
1379 |
< |
if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism) |
1380 |
< |
addWorkerIfBelowTarget(); |
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 |
|
/** |
1359 |
|
* scheduled for execution |
1360 |
|
*/ |
1361 |
|
public <T> T invoke(ForkJoinTask<T> task) { |
1362 |
< |
doSubmit(task); |
1363 |
< |
return task.join(); |
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 |
|
/** |
1397 |
|
* scheduled for execution |
1398 |
|
*/ |
1399 |
|
public void execute(ForkJoinTask<?> task) { |
1400 |
< |
doSubmit(task); |
1400 |
> |
if (task == null) |
1401 |
> |
throw new NullPointerException(); |
1402 |
> |
forkOrSubmit(task); |
1403 |
|
} |
1404 |
|
|
1405 |
|
// AbstractExecutorService methods |
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 |
< |
doSubmit(job); |
1420 |
> |
forkOrSubmit(job); |
1421 |
|
} |
1422 |
|
|
1423 |
|
/** |
1430 |
|
* scheduled for execution |
1431 |
|
*/ |
1432 |
|
public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) { |
1433 |
< |
doSubmit(task); |
1433 |
> |
if (task == null) |
1434 |
> |
throw new NullPointerException(); |
1435 |
> |
forkOrSubmit(task); |
1436 |
|
return task; |
1437 |
|
} |
1438 |
|
|
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 |
< |
doSubmit(job); |
1448 |
> |
forkOrSubmit(job); |
1449 |
|
return job; |
1450 |
|
} |
1451 |
|
|
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 |
< |
doSubmit(job); |
1461 |
> |
forkOrSubmit(job); |
1462 |
|
return job; |
1463 |
|
} |
1464 |
|
|
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 |
< |
doSubmit(job); |
1478 |
> |
forkOrSubmit(job); |
1479 |
|
return job; |
1480 |
|
} |
1481 |
|
|
1535 |
|
|
1536 |
|
/** |
1537 |
|
* Returns the number of worker threads that have started but not |
1538 |
< |
* yet terminated. This result returned by this method may differ |
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 |
|
* |
1620 |
|
*/ |
1621 |
|
public long getQueuedTaskCount() { |
1622 |
|
long count = 0; |
1623 |
< |
ForkJoinWorkerThread[] ws = workers; |
1618 |
< |
int n = ws.length; |
1619 |
< |
for (int i = 0; i < n; ++i) { |
1620 |
< |
ForkJoinWorkerThread w = ws[i]; |
1623 |
> |
for (ForkJoinWorkerThread w : workers) |
1624 |
|
if (w != null) |
1625 |
|
count += w.getQueueSize(); |
1623 |
– |
} |
1626 |
|
return count; |
1627 |
|
} |
1628 |
|
|
1677 |
|
*/ |
1678 |
|
protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) { |
1679 |
|
int count = submissionQueue.drainTo(c); |
1680 |
< |
ForkJoinWorkerThread[] ws = workers; |
1679 |
< |
int n = ws.length; |
1680 |
< |
for (int i = 0; i < n; ++i) { |
1681 |
< |
ForkJoinWorkerThread w = ws[i]; |
1680 |
> |
for (ForkJoinWorkerThread w : workers) |
1681 |
|
if (w != null) |
1682 |
|
count += w.drainTasksTo(c); |
1684 |
– |
} |
1683 |
|
return count; |
1684 |
|
} |
1685 |
|
|
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, causing this executor not |
1777 |
< |
* to properly terminate. |
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 |
|
*/ |
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 |
1815 |
|
public boolean awaitTermination(long timeout, TimeUnit unit) |
1816 |
|
throws InterruptedException { |
1817 |
|
try { |
1818 |
< |
return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0; |
1819 |
< |
} catch(TimeoutException ex) { |
1818 |
> |
termination.awaitAdvanceInterruptibly(0, timeout, unit); |
1819 |
> |
} catch (TimeoutException ex) { |
1820 |
|
return false; |
1821 |
|
} |
1822 |
+ |
return true; |
1823 |
|
} |
1824 |
|
|
1825 |
|
/** |
1864 |
|
* QueueTaker(BlockingQueue<E> q) { this.queue = q; } |
1865 |
|
* public boolean block() throws InterruptedException { |
1866 |
|
* if (item == null) |
1867 |
< |
* item = queue.take |
1867 |
> |
* item = queue.take(); |
1868 |
|
* return true; |
1869 |
|
* } |
1870 |
|
* public boolean isReleasable() { |
1871 |
< |
* return item != null || (item = queue.poll) != null; |
1871 |
> |
* return item != null || (item = queue.poll()) != null; |
1872 |
|
* } |
1873 |
|
* public E getItem() { // call after pool.managedBlock completes |
1874 |
|
* return item; |
1947 |
|
private static final long eventCountOffset = |
1948 |
|
objectFieldOffset("eventCount", ForkJoinPool.class); |
1949 |
|
private static final long eventWaitersOffset = |
1950 |
< |
objectFieldOffset("eventWaiters",ForkJoinPool.class); |
1950 |
> |
objectFieldOffset("eventWaiters", ForkJoinPool.class); |
1951 |
|
private static final long stealCountOffset = |
1952 |
< |
objectFieldOffset("stealCount",ForkJoinPool.class); |
1952 |
> |
objectFieldOffset("stealCount", ForkJoinPool.class); |
1953 |
|
private static final long spareWaitersOffset = |
1954 |
< |
objectFieldOffset("spareWaiters",ForkJoinPool.class); |
1954 |
> |
objectFieldOffset("spareWaiters", ForkJoinPool.class); |
1955 |
|
|
1956 |
|
private static long objectFieldOffset(String field, Class<?> klazz) { |
1957 |
|
try { |