1 |
jsr166 |
1.1 |
/* |
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 java.util.concurrent; |
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.locks.LockSupport; |
15 |
|
|
import java.util.concurrent.locks.ReentrantLock; |
16 |
|
|
import java.util.concurrent.atomic.AtomicInteger; |
17 |
dl |
1.14 |
import java.util.concurrent.CountDownLatch; |
18 |
jsr166 |
1.1 |
|
19 |
|
|
/** |
20 |
jsr166 |
1.4 |
* An {@link ExecutorService} for running {@link ForkJoinTask}s. |
21 |
jsr166 |
1.8 |
* A {@code ForkJoinPool} provides the entry point for submissions |
22 |
dl |
1.18 |
* from non-{@code ForkJoinTask} clients, as well as management and |
23 |
jsr166 |
1.11 |
* monitoring operations. |
24 |
jsr166 |
1.1 |
* |
25 |
jsr166 |
1.9 |
* <p>A {@code ForkJoinPool} differs from other kinds of {@link |
26 |
|
|
* ExecutorService} mainly by virtue of employing |
27 |
|
|
* <em>work-stealing</em>: all threads in the pool attempt to find and |
28 |
|
|
* execute subtasks created by other active tasks (eventually blocking |
29 |
|
|
* waiting for work if none exist). This enables efficient processing |
30 |
|
|
* when most tasks spawn other subtasks (as do most {@code |
31 |
dl |
1.18 |
* ForkJoinTask}s). When setting <em>asyncMode</em> to true in |
32 |
|
|
* constructors, {@code ForkJoinPool}s may also be appropriate for use |
33 |
|
|
* with event-style tasks that are never joined. |
34 |
jsr166 |
1.1 |
* |
35 |
jsr166 |
1.9 |
* <p>A {@code ForkJoinPool} is constructed with a given target |
36 |
|
|
* parallelism level; by default, equal to the number of available |
37 |
dl |
1.18 |
* processors. The pool attempts to maintain enough active (or |
38 |
|
|
* available) threads by dynamically adding, suspending, or resuming |
39 |
|
|
* internal worker threads, even if some tasks are stalled waiting to |
40 |
|
|
* join others. However, no such adjustments are guaranteed in the |
41 |
|
|
* face of blocked IO or other unmanaged synchronization. The nested |
42 |
|
|
* {@link ManagedBlocker} interface enables extension of the kinds of |
43 |
|
|
* synchronization accommodated. |
44 |
jsr166 |
1.1 |
* |
45 |
|
|
* <p>In addition to execution and lifecycle control methods, this |
46 |
|
|
* class provides status check methods (for example |
47 |
jsr166 |
1.4 |
* {@link #getStealCount}) that are intended to aid in developing, |
48 |
jsr166 |
1.1 |
* tuning, and monitoring fork/join applications. Also, method |
49 |
jsr166 |
1.4 |
* {@link #toString} returns indications of pool state in a |
50 |
jsr166 |
1.1 |
* convenient form for informal monitoring. |
51 |
|
|
* |
52 |
dl |
1.18 |
* <p> As is the case with other ExecutorServices, there are three |
53 |
dl |
1.19 |
* main task execution methods summarized in the following |
54 |
dl |
1.18 |
* table. These are designed to be used by clients not already engaged |
55 |
|
|
* in fork/join computations in the current pool. The main forms of |
56 |
|
|
* these methods accept instances of {@code ForkJoinTask}, but |
57 |
|
|
* overloaded forms also allow mixed execution of plain {@code |
58 |
|
|
* Runnable}- or {@code Callable}- based activities as well. However, |
59 |
|
|
* tasks that are already executing in a pool should normally |
60 |
|
|
* <em>NOT</em> use these pool execution methods, but instead use the |
61 |
dl |
1.19 |
* within-computation forms listed in the table. |
62 |
dl |
1.18 |
* |
63 |
|
|
* <table BORDER CELLPADDING=3 CELLSPACING=1> |
64 |
|
|
* <tr> |
65 |
|
|
* <td></td> |
66 |
|
|
* <td ALIGN=CENTER> <b>Call from non-fork/join clients</b></td> |
67 |
|
|
* <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td> |
68 |
|
|
* </tr> |
69 |
|
|
* <tr> |
70 |
|
|
* <td> <b>Arange async execution</td> |
71 |
|
|
* <td> {@link #execute(ForkJoinTask)}</td> |
72 |
|
|
* <td> {@link ForkJoinTask#fork}</td> |
73 |
|
|
* </tr> |
74 |
|
|
* <tr> |
75 |
|
|
* <td> <b>Await and obtain result</td> |
76 |
|
|
* <td> {@link #invoke(ForkJoinTask)}</td> |
77 |
|
|
* <td> {@link ForkJoinTask#invoke}</td> |
78 |
|
|
* </tr> |
79 |
|
|
* <tr> |
80 |
|
|
* <td> <b>Arrange exec and obtain Future</td> |
81 |
|
|
* <td> {@link #submit(ForkJoinTask)}</td> |
82 |
|
|
* <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td> |
83 |
|
|
* </tr> |
84 |
|
|
* </table> |
85 |
dl |
1.19 |
* |
86 |
jsr166 |
1.9 |
* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is |
87 |
|
|
* used for all parallel task execution in a program or subsystem. |
88 |
|
|
* Otherwise, use would not usually outweigh the construction and |
89 |
|
|
* bookkeeping overhead of creating a large set of threads. For |
90 |
|
|
* example, a common pool could be used for the {@code SortTasks} |
91 |
|
|
* illustrated in {@link RecursiveAction}. Because {@code |
92 |
|
|
* ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon |
93 |
|
|
* daemon} mode, there is typically no need to explicitly {@link |
94 |
|
|
* #shutdown} such a pool upon program exit. |
95 |
|
|
* |
96 |
|
|
* <pre> |
97 |
|
|
* static final ForkJoinPool mainPool = new ForkJoinPool(); |
98 |
|
|
* ... |
99 |
|
|
* public void sort(long[] array) { |
100 |
|
|
* mainPool.invoke(new SortTask(array, 0, array.length)); |
101 |
|
|
* } |
102 |
|
|
* </pre> |
103 |
|
|
* |
104 |
jsr166 |
1.1 |
* <p><b>Implementation notes</b>: This implementation restricts the |
105 |
|
|
* maximum number of running threads to 32767. Attempts to create |
106 |
jsr166 |
1.11 |
* pools with greater than the maximum number result in |
107 |
jsr166 |
1.8 |
* {@code IllegalArgumentException}. |
108 |
jsr166 |
1.1 |
* |
109 |
jsr166 |
1.11 |
* <p>This implementation rejects submitted tasks (that is, by throwing |
110 |
dl |
1.19 |
* {@link RejectedExecutionException}) only when the pool is shut down |
111 |
dl |
1.20 |
* or internal resources have been exhausted. |
112 |
jsr166 |
1.11 |
* |
113 |
jsr166 |
1.1 |
* @since 1.7 |
114 |
|
|
* @author Doug Lea |
115 |
|
|
*/ |
116 |
|
|
public class ForkJoinPool extends AbstractExecutorService { |
117 |
|
|
|
118 |
|
|
/* |
119 |
dl |
1.14 |
* Implementation Overview |
120 |
|
|
* |
121 |
|
|
* This class provides the central bookkeeping and control for a |
122 |
|
|
* set of worker threads: Submissions from non-FJ threads enter |
123 |
|
|
* into a submission queue. Workers take these tasks and typically |
124 |
|
|
* split them into subtasks that may be stolen by other workers. |
125 |
|
|
* The main work-stealing mechanics implemented in class |
126 |
|
|
* ForkJoinWorkerThread give first priority to processing tasks |
127 |
|
|
* from their own queues (LIFO or FIFO, depending on mode), then |
128 |
|
|
* to randomized FIFO steals of tasks in other worker queues, and |
129 |
|
|
* lastly to new submissions. These mechanics do not consider |
130 |
|
|
* affinities, loads, cache localities, etc, so rarely provide the |
131 |
|
|
* best possible performance on a given machine, but portably |
132 |
|
|
* provide good throughput by averaging over these factors. |
133 |
|
|
* (Further, even if we did try to use such information, we do not |
134 |
|
|
* usually have a basis for exploiting it. For example, some sets |
135 |
|
|
* of tasks profit from cache affinities, but others are harmed by |
136 |
|
|
* cache pollution effects.) |
137 |
|
|
* |
138 |
dl |
1.19 |
* Beyond work-stealing support and essential bookkeeping, the |
139 |
|
|
* main responsibility of this framework is to take actions when |
140 |
|
|
* one worker is waiting to join a task stolen (or always held by) |
141 |
|
|
* another. Becauae we are multiplexing many tasks on to a pool |
142 |
|
|
* of workers, we can't just let them block (as in Thread.join). |
143 |
|
|
* We also cannot just reassign the joiner's run-time stack with |
144 |
|
|
* another and replace it later, which would be a form of |
145 |
|
|
* "continuation", that even if possible is not necessarily a good |
146 |
|
|
* idea. Given that the creation costs of most threads on most |
147 |
|
|
* systems mainly surrounds setting up runtime stacks, thread |
148 |
|
|
* creation and switching is usually not much more expensive than |
149 |
|
|
* stack creation and switching, and is more flexible). Instead we |
150 |
|
|
* combine two tactics: |
151 |
|
|
* |
152 |
|
|
* Helping: Arranging for the joiner to execute some task that it |
153 |
|
|
* would be running if the steal had not occurred. Method |
154 |
|
|
* ForkJoinWorkerThread.helpJoinTask tracks joining->stealing |
155 |
|
|
* links to try to find such a task. |
156 |
|
|
* |
157 |
|
|
* Compensating: Unless there are already enough live threads, |
158 |
|
|
* method helpMaintainParallelism() may create or or |
159 |
|
|
* re-activate a spare thread to compensate for blocked |
160 |
|
|
* joiners until they unblock. |
161 |
|
|
* |
162 |
|
|
* Because the determining existence of conservatively safe |
163 |
|
|
* helping targets, the availability of already-created spares, |
164 |
|
|
* and the apparent need to create new spares are all racy and |
165 |
|
|
* require heuristic guidance, we rely on multiple retries of |
166 |
|
|
* each. Further, because it is impossible to keep exactly the |
167 |
|
|
* target (parallelism) number of threads running at any given |
168 |
|
|
* time, we allow compensation during joins to fail, and enlist |
169 |
|
|
* all other threads to help out whenever they are not otherwise |
170 |
|
|
* occupied (i.e., mainly in method preStep). |
171 |
|
|
* |
172 |
|
|
* The ManagedBlocker extension API can't use helping so relies |
173 |
|
|
* only on compensation in method awaitBlocker. |
174 |
|
|
* |
175 |
dl |
1.14 |
* The main throughput advantages of work-stealing stem from |
176 |
|
|
* decentralized control -- workers mostly steal tasks from each |
177 |
|
|
* other. We do not want to negate this by creating bottlenecks |
178 |
dl |
1.19 |
* implementing other management responsibilities. So we use a |
179 |
|
|
* collection of techniques that avoid, reduce, or cope well with |
180 |
|
|
* contention. These entail several instances of bit-packing into |
181 |
|
|
* CASable fields to maintain only the minimally required |
182 |
|
|
* atomicity. To enable such packing, we restrict maximum |
183 |
|
|
* parallelism to (1<<15)-1 (enabling twice this (to accommodate |
184 |
|
|
* unbalanced increments and decrements) to fit into a 16 bit |
185 |
|
|
* field, which is far in excess of normal operating range. Even |
186 |
|
|
* though updates to some of these bookkeeping fields do sometimes |
187 |
|
|
* contend with each other, they don't normally cache-contend with |
188 |
|
|
* updates to others enough to warrant memory padding or |
189 |
|
|
* isolation. So they are all held as fields of ForkJoinPool |
190 |
|
|
* objects. The main capabilities are as follows: |
191 |
dl |
1.14 |
* |
192 |
|
|
* 1. Creating and removing workers. Workers are recorded in the |
193 |
|
|
* "workers" array. This is an array as opposed to some other data |
194 |
|
|
* structure to support index-based random steals by workers. |
195 |
|
|
* Updates to the array recording new workers and unrecording |
196 |
|
|
* terminated ones are protected from each other by a lock |
197 |
|
|
* (workerLock) but the array is otherwise concurrently readable, |
198 |
|
|
* and accessed directly by workers. To simplify index-based |
199 |
|
|
* operations, the array size is always a power of two, and all |
200 |
dl |
1.17 |
* readers must tolerate null slots. Currently, all worker thread |
201 |
|
|
* creation is on-demand, triggered by task submissions, |
202 |
|
|
* replacement of terminated workers, and/or compensation for |
203 |
|
|
* blocked workers. However, all other support code is set up to |
204 |
|
|
* work with other policies. |
205 |
dl |
1.14 |
* |
206 |
dl |
1.19 |
* To ensure that we do not hold on to worker references that |
207 |
|
|
* would prevent GC, ALL accesses to workers are via indices into |
208 |
|
|
* the workers array (which is one source of some of the unusual |
209 |
|
|
* code constructions here). In essence, the workers array serves |
210 |
|
|
* as a WeakReference mechanism. Thus for example the event queue |
211 |
|
|
* stores worker indices, not worker references. Access to the |
212 |
|
|
* workers in associated methods (for example releaseEventWaiters) |
213 |
|
|
* must both index-check and null-check the IDs. All such accesses |
214 |
|
|
* ignore bad IDs by returning out early from what they are doing, |
215 |
|
|
* since this can only be associated with shutdown, in which case |
216 |
|
|
* it is OK to give up. On termination, we just clobber these |
217 |
|
|
* data structures without trying to use them. |
218 |
|
|
* |
219 |
dl |
1.14 |
* 2. Bookkeeping for dynamically adding and removing workers. We |
220 |
dl |
1.18 |
* aim to approximately maintain the given level of parallelism. |
221 |
|
|
* When some workers are known to be blocked (on joins or via |
222 |
dl |
1.14 |
* ManagedBlocker), we may create or resume others to take their |
223 |
|
|
* place until they unblock (see below). Implementing this |
224 |
|
|
* requires counts of the number of "running" threads (i.e., those |
225 |
|
|
* that are neither blocked nor artifically suspended) as well as |
226 |
|
|
* the total number. These two values are packed into one field, |
227 |
|
|
* "workerCounts" because we need accurate snapshots when deciding |
228 |
dl |
1.19 |
* to create, resume or suspend. Note however that the |
229 |
|
|
* correspondance of these counts to reality is not guaranteed. In |
230 |
|
|
* particular updates for unblocked threads may lag until they |
231 |
|
|
* actually wake up. |
232 |
dl |
1.14 |
* |
233 |
|
|
* 3. Maintaining global run state. The run state of the pool |
234 |
|
|
* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to |
235 |
|
|
* those in other Executor implementations, as well as a count of |
236 |
|
|
* "active" workers -- those that are, or soon will be, or |
237 |
|
|
* recently were executing tasks. The runLevel and active count |
238 |
|
|
* are packed together in order to correctly trigger shutdown and |
239 |
|
|
* termination. Without care, active counts can be subject to very |
240 |
|
|
* high contention. We substantially reduce this contention by |
241 |
|
|
* relaxing update rules. A worker must claim active status |
242 |
|
|
* prospectively, by activating if it sees that a submitted or |
243 |
|
|
* stealable task exists (it may find after activating that the |
244 |
|
|
* task no longer exists). It stays active while processing this |
245 |
|
|
* task (if it exists) and any other local subtasks it produces, |
246 |
|
|
* until it cannot find any other tasks. It then tries |
247 |
|
|
* inactivating (see method preStep), but upon update contention |
248 |
|
|
* instead scans for more tasks, later retrying inactivation if it |
249 |
|
|
* doesn't find any. |
250 |
|
|
* |
251 |
|
|
* 4. Managing idle workers waiting for tasks. We cannot let |
252 |
|
|
* workers spin indefinitely scanning for tasks when none are |
253 |
|
|
* available. On the other hand, we must quickly prod them into |
254 |
|
|
* action when new tasks are submitted or generated. We |
255 |
|
|
* park/unpark these idle workers using an event-count scheme. |
256 |
|
|
* Field eventCount is incremented upon events that may enable |
257 |
|
|
* workers that previously could not find a task to now find one: |
258 |
|
|
* Submission of a new task to the pool, or another worker pushing |
259 |
|
|
* a task onto a previously empty queue. (We also use this |
260 |
dl |
1.22 |
* mechanism for configuration and termination actions that |
261 |
|
|
* require wakeups of idle workers). Each worker maintains its |
262 |
|
|
* last known event count, and blocks when a scan for work did not |
263 |
|
|
* find a task AND its lastEventCount matches the current |
264 |
|
|
* eventCount. Waiting idle workers are recorded in a variant of |
265 |
|
|
* Treiber stack headed by field eventWaiters which, when nonzero, |
266 |
|
|
* encodes the thread index and count awaited for by the worker |
267 |
|
|
* thread most recently calling eventSync. This thread in turn has |
268 |
|
|
* a record (field nextEventWaiter) for the next waiting worker. |
269 |
|
|
* In addition to allowing simpler decisions about need for |
270 |
|
|
* wakeup, the event count bits in eventWaiters serve the role of |
271 |
|
|
* tags to avoid ABA errors in Treiber stacks. Upon any wakeup, |
272 |
|
|
* released threads also try to release others (but give up upon |
273 |
|
|
* contention to reduce useless flailing). The net effect is a |
274 |
|
|
* tree-like diffusion of signals, where released threads (and |
275 |
|
|
* possibly others) help with unparks. To further reduce |
276 |
|
|
* contention effects a bit, failed CASes to increment field |
277 |
|
|
* eventCount are tolerated without retries in signalWork. |
278 |
dl |
1.14 |
* Conceptually they are merged into the same event, which is OK |
279 |
|
|
* when their only purpose is to enable workers to scan for work. |
280 |
|
|
* |
281 |
|
|
* 5. Managing suspension of extra workers. When a worker is about |
282 |
|
|
* to block waiting for a join (or via ManagedBlockers), we may |
283 |
|
|
* create a new thread to maintain parallelism level, or at least |
284 |
dl |
1.19 |
* avoid starvation. Usually, extra threads are needed for only |
285 |
|
|
* very short periods, yet join dependencies are such that we |
286 |
|
|
* sometimes need them in bursts. Rather than create new threads |
287 |
|
|
* each time this happens, we suspend no-longer-needed extra ones |
288 |
|
|
* as "spares". For most purposes, we don't distinguish "extra" |
289 |
|
|
* spare threads from normal "core" threads: On each call to |
290 |
|
|
* preStep (the only point at which we can do this) a worker |
291 |
dl |
1.14 |
* checks to see if there are now too many running workers, and if |
292 |
dl |
1.19 |
* so, suspends itself. Method helpMaintainParallelism looks for |
293 |
|
|
* suspended threads to resume before considering creating a new |
294 |
|
|
* replacement. The spares themselves are encoded on another |
295 |
|
|
* variant of a Treiber Stack, headed at field "spareWaiters". |
296 |
|
|
* Note that the use of spares is intrinsically racy. One thread |
297 |
dl |
1.14 |
* may become a spare at about the same time as another is |
298 |
|
|
* needlessly being created. We counteract this and related slop |
299 |
|
|
* in part by requiring resumed spares to immediately recheck (in |
300 |
dl |
1.22 |
* preStep) to see whether they they should re-suspend. |
301 |
dl |
1.14 |
* |
302 |
dl |
1.22 |
* 6. Killing off unneeded workers. The Spare and Event queues use |
303 |
|
|
* similar mechanisms to shed unused workers: The oldest (first) |
304 |
|
|
* waiter uses a timed rather than hard wait. When this wait times |
305 |
|
|
* out without a normal wakeup, it tries to shutdown any one (for |
306 |
|
|
* convenience the newest) other waiter via tryShutdownSpare or |
307 |
|
|
* tryShutdownWaiter, respectively. The wakeup rates for spares |
308 |
|
|
* are much shorter than for waiters. Together, they will |
309 |
|
|
* eventually reduce the number of worker threads to a minimum of |
310 |
|
|
* one after a long enough period without use. |
311 |
|
|
* |
312 |
|
|
* 7. Deciding when to create new workers. The main dynamic |
313 |
dl |
1.19 |
* control in this class is deciding when to create extra threads |
314 |
|
|
* in method helpMaintainParallelism. We would like to keep |
315 |
|
|
* exactly #parallelism threads running, which is an impossble |
316 |
|
|
* task. We always need to create one when the number of running |
317 |
|
|
* threads would become zero and all workers are busy. Beyond |
318 |
|
|
* this, we must rely on heuristics that work well in the the |
319 |
|
|
* presence of transients phenomena such as GC stalls, dynamic |
320 |
|
|
* compilation, and wake-up lags. These transients are extremely |
321 |
|
|
* common -- we are normally trying to fully saturate the CPUs on |
322 |
|
|
* a machine, so almost any activity other than running tasks |
323 |
|
|
* impedes accuracy. Our main defense is to allow some slack in |
324 |
|
|
* creation thresholds, using rules that reflect the fact that the |
325 |
|
|
* more threads we have running, the more likely that we are |
326 |
dl |
1.22 |
* underestimating the number running threads. (We also include |
327 |
|
|
* some heuristic use of Thread.yield when all workers appear to |
328 |
|
|
* be busy, to improve likelihood of counts settling.) The rules |
329 |
|
|
* also better cope with the fact that some of the methods in this |
330 |
dl |
1.19 |
* class tend to never become compiled (but are interpreted), so |
331 |
|
|
* some components of the entire set of controls might execute 100 |
332 |
|
|
* times faster than others. And similarly for cases where the |
333 |
|
|
* apparent lack of work is just due to GC stalls and other |
334 |
|
|
* transient system activity. |
335 |
dl |
1.14 |
* |
336 |
|
|
* Beware that there is a lot of representation-level coupling |
337 |
|
|
* among classes ForkJoinPool, ForkJoinWorkerThread, and |
338 |
|
|
* ForkJoinTask. For example, direct access to "workers" array by |
339 |
|
|
* workers, and direct access to ForkJoinTask.status by both |
340 |
|
|
* ForkJoinPool and ForkJoinWorkerThread. There is little point |
341 |
|
|
* trying to reduce this, since any associated future changes in |
342 |
|
|
* representations will need to be accompanied by algorithmic |
343 |
|
|
* changes anyway. |
344 |
|
|
* |
345 |
|
|
* Style notes: There are lots of inline assignments (of form |
346 |
|
|
* "while ((local = field) != 0)") which are usually the simplest |
347 |
dl |
1.19 |
* way to ensure the required read orderings (which are sometimes |
348 |
|
|
* critical). Also several occurrences of the unusual "do {} |
349 |
|
|
* while(!cas...)" which is the simplest way to force an update of |
350 |
|
|
* a CAS'ed variable. There are also other coding oddities that |
351 |
|
|
* help some methods perform reasonably even when interpreted (not |
352 |
|
|
* compiled), at the expense of some messy constructions that |
353 |
|
|
* reduce byte code counts. |
354 |
dl |
1.14 |
* |
355 |
|
|
* The order of declarations in this file is: (1) statics (2) |
356 |
|
|
* fields (along with constants used when unpacking some of them) |
357 |
|
|
* (3) internal control methods (4) callbacks and other support |
358 |
|
|
* for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported |
359 |
|
|
* methods (plus a few little helpers). |
360 |
jsr166 |
1.1 |
*/ |
361 |
|
|
|
362 |
|
|
/** |
363 |
jsr166 |
1.8 |
* Factory for creating new {@link ForkJoinWorkerThread}s. |
364 |
|
|
* A {@code ForkJoinWorkerThreadFactory} must be defined and used |
365 |
|
|
* for {@code ForkJoinWorkerThread} subclasses that extend base |
366 |
|
|
* functionality or initialize threads with different contexts. |
367 |
jsr166 |
1.1 |
*/ |
368 |
|
|
public static interface ForkJoinWorkerThreadFactory { |
369 |
|
|
/** |
370 |
|
|
* Returns a new worker thread operating in the given pool. |
371 |
|
|
* |
372 |
|
|
* @param pool the pool this thread works in |
373 |
jsr166 |
1.11 |
* @throws NullPointerException if the pool is null |
374 |
jsr166 |
1.1 |
*/ |
375 |
|
|
public ForkJoinWorkerThread newThread(ForkJoinPool pool); |
376 |
|
|
} |
377 |
|
|
|
378 |
|
|
/** |
379 |
|
|
* Default ForkJoinWorkerThreadFactory implementation; creates a |
380 |
|
|
* new ForkJoinWorkerThread. |
381 |
|
|
*/ |
382 |
dl |
1.18 |
static class DefaultForkJoinWorkerThreadFactory |
383 |
jsr166 |
1.1 |
implements ForkJoinWorkerThreadFactory { |
384 |
|
|
public ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
385 |
dl |
1.14 |
return new ForkJoinWorkerThread(pool); |
386 |
jsr166 |
1.1 |
} |
387 |
|
|
} |
388 |
|
|
|
389 |
|
|
/** |
390 |
|
|
* Creates a new ForkJoinWorkerThread. This factory is used unless |
391 |
|
|
* overridden in ForkJoinPool constructors. |
392 |
|
|
*/ |
393 |
|
|
public static final ForkJoinWorkerThreadFactory |
394 |
|
|
defaultForkJoinWorkerThreadFactory = |
395 |
|
|
new DefaultForkJoinWorkerThreadFactory(); |
396 |
|
|
|
397 |
|
|
/** |
398 |
|
|
* Permission required for callers of methods that may start or |
399 |
|
|
* kill threads. |
400 |
|
|
*/ |
401 |
|
|
private static final RuntimePermission modifyThreadPermission = |
402 |
|
|
new RuntimePermission("modifyThread"); |
403 |
|
|
|
404 |
|
|
/** |
405 |
|
|
* If there is a security manager, makes sure caller has |
406 |
|
|
* permission to modify threads. |
407 |
|
|
*/ |
408 |
|
|
private static void checkPermission() { |
409 |
|
|
SecurityManager security = System.getSecurityManager(); |
410 |
|
|
if (security != null) |
411 |
|
|
security.checkPermission(modifyThreadPermission); |
412 |
|
|
} |
413 |
|
|
|
414 |
|
|
/** |
415 |
|
|
* Generator for assigning sequence numbers as pool names. |
416 |
|
|
*/ |
417 |
|
|
private static final AtomicInteger poolNumberGenerator = |
418 |
|
|
new AtomicInteger(); |
419 |
|
|
|
420 |
|
|
/** |
421 |
dl |
1.22 |
* The wakeup interval (in nanoseconds) for the oldest worker |
422 |
|
|
* worker waiting for an event invokes tryShutdownWaiter to shrink |
423 |
|
|
* the number of workers. The exact value does not matter too |
424 |
|
|
* much, but should be long enough to slowly release resources |
425 |
|
|
* during long periods without use without disrupting normal use. |
426 |
|
|
*/ |
427 |
|
|
private static final long SHRINK_RATE_NANOS = |
428 |
|
|
60L * 1000L * 1000L * 1000L; // one minute |
429 |
|
|
|
430 |
|
|
/** |
431 |
dl |
1.19 |
* Absolute bound for parallelism level. Twice this number plus |
432 |
|
|
* one (i.e., 0xfff) must fit into a 16bit field to enable |
433 |
|
|
* word-packing for some counts and indices. |
434 |
dl |
1.14 |
*/ |
435 |
dl |
1.19 |
private static final int MAX_WORKERS = 0x7fff; |
436 |
dl |
1.14 |
|
437 |
|
|
/** |
438 |
|
|
* Array holding all worker threads in the pool. Array size must |
439 |
|
|
* be a power of two. Updates and replacements are protected by |
440 |
|
|
* workerLock, but the array is always kept in a consistent enough |
441 |
|
|
* state to be randomly accessed without locking by workers |
442 |
|
|
* performing work-stealing, as well as other traversal-based |
443 |
|
|
* methods in this class. All readers must tolerate that some |
444 |
|
|
* array slots may be null. |
445 |
jsr166 |
1.1 |
*/ |
446 |
|
|
volatile ForkJoinWorkerThread[] workers; |
447 |
|
|
|
448 |
|
|
/** |
449 |
dl |
1.14 |
* Queue for external submissions. |
450 |
jsr166 |
1.1 |
*/ |
451 |
dl |
1.14 |
private final LinkedTransferQueue<ForkJoinTask<?>> submissionQueue; |
452 |
jsr166 |
1.1 |
|
453 |
|
|
/** |
454 |
dl |
1.14 |
* Lock protecting updates to workers array. |
455 |
jsr166 |
1.1 |
*/ |
456 |
dl |
1.14 |
private final ReentrantLock workerLock; |
457 |
jsr166 |
1.1 |
|
458 |
|
|
/** |
459 |
dl |
1.14 |
* Latch released upon termination. |
460 |
jsr166 |
1.1 |
*/ |
461 |
dl |
1.18 |
private final Phaser termination; |
462 |
jsr166 |
1.1 |
|
463 |
|
|
/** |
464 |
|
|
* Creation factory for worker threads. |
465 |
|
|
*/ |
466 |
|
|
private final ForkJoinWorkerThreadFactory factory; |
467 |
|
|
|
468 |
|
|
/** |
469 |
dl |
1.14 |
* Sum of per-thread steal counts, updated only when threads are |
470 |
|
|
* idle or terminating. |
471 |
jsr166 |
1.1 |
*/ |
472 |
dl |
1.14 |
private volatile long stealCount; |
473 |
jsr166 |
1.1 |
|
474 |
|
|
/** |
475 |
dl |
1.14 |
* Encoded record of top of treiber stack of threads waiting for |
476 |
|
|
* events. The top 32 bits contain the count being waited for. The |
477 |
dl |
1.19 |
* bottom 16 bits contains one plus the pool index of waiting |
478 |
|
|
* worker thread. (Bits 16-31 are unused.) |
479 |
jsr166 |
1.1 |
*/ |
480 |
dl |
1.14 |
private volatile long eventWaiters; |
481 |
|
|
|
482 |
|
|
private static final int EVENT_COUNT_SHIFT = 32; |
483 |
dl |
1.19 |
private static final long WAITER_ID_MASK = (1L << 16) - 1L; |
484 |
jsr166 |
1.1 |
|
485 |
|
|
/** |
486 |
dl |
1.14 |
* A counter for events that may wake up worker threads: |
487 |
|
|
* - Submission of a new task to the pool |
488 |
|
|
* - A worker pushing a task on an empty queue |
489 |
dl |
1.19 |
* - termination |
490 |
jsr166 |
1.1 |
*/ |
491 |
dl |
1.14 |
private volatile int eventCount; |
492 |
|
|
|
493 |
|
|
/** |
494 |
dl |
1.19 |
* Encoded record of top of treiber stack of spare threads waiting |
495 |
|
|
* for resumption. The top 16 bits contain an arbitrary count to |
496 |
|
|
* avoid ABA effects. The bottom 16bits contains one plus the pool |
497 |
|
|
* index of waiting worker thread. |
498 |
|
|
*/ |
499 |
|
|
private volatile int spareWaiters; |
500 |
|
|
|
501 |
|
|
private static final int SPARE_COUNT_SHIFT = 16; |
502 |
|
|
private static final int SPARE_ID_MASK = (1 << 16) - 1; |
503 |
|
|
|
504 |
|
|
/** |
505 |
dl |
1.14 |
* Lifecycle control. The low word contains the number of workers |
506 |
|
|
* that are (probably) executing tasks. This value is atomically |
507 |
|
|
* incremented before a worker gets a task to run, and decremented |
508 |
|
|
* when worker has no tasks and cannot find any. Bits 16-18 |
509 |
|
|
* contain runLevel value. When all are zero, the pool is |
510 |
|
|
* running. Level transitions are monotonic (running -> shutdown |
511 |
|
|
* -> terminating -> terminated) so each transition adds a bit. |
512 |
|
|
* These are bundled together to ensure consistent read for |
513 |
|
|
* termination checks (i.e., that runLevel is at least SHUTDOWN |
514 |
|
|
* and active threads is zero). |
515 |
dl |
1.22 |
* |
516 |
|
|
* Notes: Most direct CASes are dependent on these bitfield |
517 |
|
|
* positions. Also, this field is non-private to enable direct |
518 |
|
|
* performance-sensitive CASes in ForkJoinWorkerThread. |
519 |
dl |
1.14 |
*/ |
520 |
dl |
1.22 |
volatile int runState; |
521 |
dl |
1.14 |
|
522 |
|
|
// Note: The order among run level values matters. |
523 |
|
|
private static final int RUNLEVEL_SHIFT = 16; |
524 |
|
|
private static final int SHUTDOWN = 1 << RUNLEVEL_SHIFT; |
525 |
|
|
private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1); |
526 |
|
|
private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2); |
527 |
|
|
private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1; |
528 |
jsr166 |
1.1 |
|
529 |
|
|
/** |
530 |
dl |
1.14 |
* Holds number of total (i.e., created and not yet terminated) |
531 |
|
|
* and running (i.e., not blocked on joins or other managed sync) |
532 |
|
|
* threads, packed together to ensure consistent snapshot when |
533 |
|
|
* making decisions about creating and suspending spare |
534 |
|
|
* threads. Updated only by CAS. Note that adding a new worker |
535 |
|
|
* requires incrementing both counts, since workers start off in |
536 |
dl |
1.19 |
* running state. |
537 |
dl |
1.14 |
*/ |
538 |
|
|
private volatile int workerCounts; |
539 |
|
|
|
540 |
|
|
private static final int TOTAL_COUNT_SHIFT = 16; |
541 |
|
|
private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1; |
542 |
|
|
private static final int ONE_RUNNING = 1; |
543 |
|
|
private static final int ONE_TOTAL = 1 << TOTAL_COUNT_SHIFT; |
544 |
|
|
|
545 |
jsr166 |
1.1 |
/** |
546 |
dl |
1.14 |
* The target parallelism level. |
547 |
dl |
1.18 |
* Accessed directly by ForkJoinWorkerThreads. |
548 |
jsr166 |
1.1 |
*/ |
549 |
dl |
1.18 |
final int parallelism; |
550 |
jsr166 |
1.1 |
|
551 |
|
|
/** |
552 |
dl |
1.14 |
* True if use local fifo, not default lifo, for local polling |
553 |
dl |
1.18 |
* Read by, and replicated by ForkJoinWorkerThreads |
554 |
jsr166 |
1.1 |
*/ |
555 |
dl |
1.18 |
final boolean locallyFifo; |
556 |
jsr166 |
1.1 |
|
557 |
|
|
/** |
558 |
dl |
1.18 |
* The uncaught exception handler used when any worker abruptly |
559 |
|
|
* terminates. |
560 |
jsr166 |
1.1 |
*/ |
561 |
dl |
1.18 |
private final Thread.UncaughtExceptionHandler ueh; |
562 |
jsr166 |
1.1 |
|
563 |
|
|
/** |
564 |
dl |
1.14 |
* Pool number, just for assigning useful names to worker threads |
565 |
jsr166 |
1.1 |
*/ |
566 |
dl |
1.14 |
private final int poolNumber; |
567 |
jsr166 |
1.1 |
|
568 |
dl |
1.19 |
|
569 |
dl |
1.22 |
// Utilities for CASing fields. Note that most of these |
570 |
|
|
// are usually manually inlined by callers |
571 |
jsr166 |
1.1 |
|
572 |
|
|
/** |
573 |
dl |
1.19 |
* Increments running count part of workerCounts |
574 |
jsr166 |
1.1 |
*/ |
575 |
dl |
1.18 |
final void incrementRunningCount() { |
576 |
|
|
int c; |
577 |
dl |
1.14 |
do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
578 |
dl |
1.19 |
c = workerCounts, |
579 |
dl |
1.18 |
c + ONE_RUNNING)); |
580 |
jsr166 |
1.1 |
} |
581 |
dl |
1.19 |
|
582 |
jsr166 |
1.1 |
/** |
583 |
dl |
1.18 |
* Tries to decrement running count unless already zero |
584 |
dl |
1.17 |
*/ |
585 |
|
|
final boolean tryDecrementRunningCount() { |
586 |
|
|
int wc = workerCounts; |
587 |
|
|
if ((wc & RUNNING_COUNT_MASK) == 0) |
588 |
|
|
return false; |
589 |
|
|
return UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
590 |
|
|
wc, wc - ONE_RUNNING); |
591 |
|
|
} |
592 |
|
|
|
593 |
|
|
/** |
594 |
dl |
1.19 |
* Forces decrement of encoded workerCounts, awaiting nonzero if |
595 |
|
|
* (rarely) necessary when other count updates lag. |
596 |
|
|
* |
597 |
|
|
* @param dr -- either zero or ONE_RUNNING |
598 |
|
|
* @param dt == either zero or ONE_TOTAL |
599 |
|
|
*/ |
600 |
|
|
private void decrementWorkerCounts(int dr, int dt) { |
601 |
|
|
for (;;) { |
602 |
|
|
int wc = workerCounts; |
603 |
|
|
if ((wc & RUNNING_COUNT_MASK) - dr < 0 || |
604 |
dl |
1.22 |
(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) { |
605 |
|
|
if ((runState & TERMINATED) != 0) |
606 |
|
|
return; // lagging termination on a backout |
607 |
dl |
1.19 |
Thread.yield(); |
608 |
dl |
1.22 |
} |
609 |
dl |
1.19 |
if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
610 |
|
|
wc, wc - (dr + dt))) |
611 |
|
|
return; |
612 |
|
|
} |
613 |
|
|
} |
614 |
|
|
|
615 |
|
|
/** |
616 |
|
|
* Increments event count |
617 |
|
|
*/ |
618 |
|
|
private void advanceEventCount() { |
619 |
|
|
int c; |
620 |
|
|
do {} while(!UNSAFE.compareAndSwapInt(this, eventCountOffset, |
621 |
|
|
c = eventCount, c+1)); |
622 |
|
|
} |
623 |
|
|
|
624 |
|
|
/** |
625 |
jsr166 |
1.1 |
* Tries incrementing active count; fails on contention. |
626 |
dl |
1.14 |
* Called by workers before executing tasks. |
627 |
jsr166 |
1.1 |
* |
628 |
|
|
* @return true on success |
629 |
|
|
*/ |
630 |
|
|
final boolean tryIncrementActiveCount() { |
631 |
dl |
1.14 |
int c; |
632 |
|
|
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
633 |
dl |
1.22 |
c = runState, c + 1); |
634 |
jsr166 |
1.1 |
} |
635 |
|
|
|
636 |
|
|
/** |
637 |
|
|
* Tries decrementing active count; fails on contention. |
638 |
dl |
1.14 |
* Called when workers cannot find tasks to run. |
639 |
|
|
*/ |
640 |
|
|
final boolean tryDecrementActiveCount() { |
641 |
|
|
int c; |
642 |
|
|
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
643 |
dl |
1.22 |
c = runState, c - 1); |
644 |
dl |
1.14 |
} |
645 |
|
|
|
646 |
|
|
/** |
647 |
|
|
* Advances to at least the given level. Returns true if not |
648 |
|
|
* already in at least the given level. |
649 |
|
|
*/ |
650 |
|
|
private boolean advanceRunLevel(int level) { |
651 |
|
|
for (;;) { |
652 |
|
|
int s = runState; |
653 |
|
|
if ((s & level) != 0) |
654 |
|
|
return false; |
655 |
|
|
if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level)) |
656 |
|
|
return true; |
657 |
|
|
} |
658 |
|
|
} |
659 |
|
|
|
660 |
|
|
// workers array maintenance |
661 |
|
|
|
662 |
|
|
/** |
663 |
|
|
* Records and returns a workers array index for new worker. |
664 |
|
|
*/ |
665 |
|
|
private int recordWorker(ForkJoinWorkerThread w) { |
666 |
|
|
// Try using slot totalCount-1. If not available, scan and/or resize |
667 |
|
|
int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1; |
668 |
|
|
final ReentrantLock lock = this.workerLock; |
669 |
|
|
lock.lock(); |
670 |
|
|
try { |
671 |
|
|
ForkJoinWorkerThread[] ws = workers; |
672 |
dl |
1.19 |
int n = ws.length; |
673 |
|
|
if (k < 0 || k >= n || ws[k] != null) { |
674 |
|
|
for (k = 0; k < n && ws[k] != null; ++k) |
675 |
dl |
1.14 |
; |
676 |
dl |
1.19 |
if (k == n) |
677 |
|
|
ws = Arrays.copyOf(ws, n << 1); |
678 |
dl |
1.14 |
} |
679 |
|
|
ws[k] = w; |
680 |
|
|
workers = ws; // volatile array write ensures slot visibility |
681 |
|
|
} finally { |
682 |
|
|
lock.unlock(); |
683 |
|
|
} |
684 |
|
|
return k; |
685 |
|
|
} |
686 |
|
|
|
687 |
|
|
/** |
688 |
|
|
* Nulls out record of worker in workers array |
689 |
|
|
*/ |
690 |
|
|
private void forgetWorker(ForkJoinWorkerThread w) { |
691 |
|
|
int idx = w.poolIndex; |
692 |
|
|
// Locking helps method recordWorker avoid unecessary expansion |
693 |
|
|
final ReentrantLock lock = this.workerLock; |
694 |
|
|
lock.lock(); |
695 |
|
|
try { |
696 |
|
|
ForkJoinWorkerThread[] ws = workers; |
697 |
|
|
if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify |
698 |
|
|
ws[idx] = null; |
699 |
|
|
} finally { |
700 |
|
|
lock.unlock(); |
701 |
|
|
} |
702 |
|
|
} |
703 |
|
|
|
704 |
|
|
// adding and removing workers |
705 |
|
|
|
706 |
|
|
/** |
707 |
|
|
* Tries to create and add new worker. Assumes that worker counts |
708 |
|
|
* are already updated to accommodate the worker, so adjusts on |
709 |
|
|
* failure. |
710 |
dl |
1.22 |
* |
711 |
|
|
* @return the worker, or null on failure |
712 |
dl |
1.14 |
*/ |
713 |
dl |
1.22 |
private ForkJoinWorkerThread addWorker() { |
714 |
dl |
1.14 |
ForkJoinWorkerThread w = null; |
715 |
|
|
try { |
716 |
|
|
w = factory.newThread(this); |
717 |
|
|
} finally { // Adjust on either null or exceptional factory return |
718 |
|
|
if (w == null) { |
719 |
dl |
1.19 |
decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL); |
720 |
|
|
tryTerminate(false); // in case of failure during shutdown |
721 |
dl |
1.14 |
} |
722 |
|
|
} |
723 |
dl |
1.22 |
if (w != null) { |
724 |
dl |
1.19 |
w.start(recordWorker(w), ueh); |
725 |
dl |
1.22 |
advanceEventCount(); |
726 |
|
|
} |
727 |
|
|
return w; |
728 |
dl |
1.14 |
} |
729 |
|
|
|
730 |
|
|
/** |
731 |
|
|
* Final callback from terminating worker. Removes record of |
732 |
|
|
* worker from array, and adjusts counts. If pool is shutting |
733 |
dl |
1.19 |
* down, tries to complete terminatation. |
734 |
dl |
1.14 |
* |
735 |
|
|
* @param w the worker |
736 |
|
|
*/ |
737 |
|
|
final void workerTerminated(ForkJoinWorkerThread w) { |
738 |
|
|
forgetWorker(w); |
739 |
dl |
1.19 |
decrementWorkerCounts(w.isTrimmed()? 0 : ONE_RUNNING, ONE_TOTAL); |
740 |
|
|
while (w.stealCount != 0) // collect final count |
741 |
|
|
tryAccumulateStealCount(w); |
742 |
|
|
tryTerminate(false); |
743 |
dl |
1.14 |
} |
744 |
|
|
|
745 |
|
|
// Waiting for and signalling events |
746 |
|
|
|
747 |
|
|
/** |
748 |
|
|
* Releases workers blocked on a count not equal to current count. |
749 |
dl |
1.19 |
* Normally called after precheck that eventWaiters isn't zero to |
750 |
dl |
1.22 |
* avoid wasted array checks. Gives up upon a change in count or |
751 |
|
|
* contention, letting other workers take over. |
752 |
dl |
1.14 |
*/ |
753 |
dl |
1.22 |
private void releaseEventWaiters() { |
754 |
dl |
1.19 |
ForkJoinWorkerThread[] ws = workers; |
755 |
|
|
int n = ws.length; |
756 |
dl |
1.22 |
long h = eventWaiters; |
757 |
|
|
int ec = eventCount; |
758 |
|
|
ForkJoinWorkerThread w; int id; |
759 |
|
|
while ((int)(h >>> EVENT_COUNT_SHIFT) != ec && |
760 |
|
|
(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 && |
761 |
|
|
id < n && (w = ws[id]) != null && |
762 |
|
|
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
763 |
|
|
h, h = w.nextWaiter)) { |
764 |
|
|
LockSupport.unpark(w); |
765 |
|
|
if (eventWaiters != h || eventCount != ec) |
766 |
dl |
1.19 |
break; |
767 |
dl |
1.14 |
} |
768 |
|
|
} |
769 |
|
|
|
770 |
|
|
/** |
771 |
dl |
1.19 |
* Tries to advance eventCount and releases waiters. Called only |
772 |
|
|
* from workers. |
773 |
dl |
1.18 |
*/ |
774 |
dl |
1.19 |
final void signalWork() { |
775 |
|
|
int c; // try to increment event count -- CAS failure OK |
776 |
|
|
UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); |
777 |
|
|
if (eventWaiters != 0L) |
778 |
dl |
1.22 |
releaseEventWaiters(); |
779 |
dl |
1.18 |
} |
780 |
|
|
|
781 |
|
|
/** |
782 |
dl |
1.22 |
* Adds the given worker to event queue and blocks until |
783 |
|
|
* terminating or event count advances from the workers |
784 |
|
|
* lastEventCount value |
785 |
dl |
1.19 |
* |
786 |
|
|
* @param w the calling worker thread |
787 |
dl |
1.14 |
*/ |
788 |
dl |
1.19 |
private void eventSync(ForkJoinWorkerThread w) { |
789 |
dl |
1.22 |
int ec = w.lastEventCount; |
790 |
|
|
long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1)); |
791 |
dl |
1.19 |
long h; |
792 |
|
|
while ((runState < SHUTDOWN || !tryTerminate(false)) && |
793 |
dl |
1.22 |
(((int)((h = eventWaiters) & WAITER_ID_MASK)) == 0 || |
794 |
|
|
(int)(h >>> EVENT_COUNT_SHIFT) == ec) && |
795 |
|
|
eventCount == ec) { |
796 |
dl |
1.19 |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
797 |
|
|
w.nextWaiter = h, nh)) { |
798 |
dl |
1.22 |
awaitEvent(w, ec); |
799 |
|
|
break; |
800 |
|
|
} |
801 |
|
|
} |
802 |
|
|
} |
803 |
|
|
|
804 |
|
|
/** |
805 |
|
|
* Blocks the given worker (that has already been entered as an |
806 |
|
|
* event waiter) until terminating or event count advances from |
807 |
|
|
* the given value. The oldest (first) waiter uses a timed wait to |
808 |
|
|
* occasionally one-by-one shrink the number of workers (to a |
809 |
|
|
* minumum of one) if the pool has not been used for extended |
810 |
|
|
* periods. |
811 |
|
|
* |
812 |
|
|
* @param w the calling worker thread |
813 |
|
|
* @param ec the count |
814 |
|
|
*/ |
815 |
|
|
private void awaitEvent(ForkJoinWorkerThread w, int ec) { |
816 |
|
|
while (eventCount == ec) { |
817 |
|
|
if (tryAccumulateStealCount(w)) { // transfer while idle |
818 |
|
|
boolean untimed = (w.nextWaiter != 0L || |
819 |
|
|
(workerCounts & RUNNING_COUNT_MASK) <= 1); |
820 |
|
|
long startTime = untimed? 0 : System.nanoTime(); |
821 |
|
|
Thread.interrupted(); // clear/ignore interrupt |
822 |
|
|
if (eventCount != ec || !w.isRunning() || |
823 |
|
|
runState >= TERMINATING) // recheck after clear |
824 |
|
|
break; |
825 |
|
|
if (untimed) |
826 |
|
|
LockSupport.park(w); |
827 |
|
|
else { |
828 |
|
|
LockSupport.parkNanos(w, SHRINK_RATE_NANOS); |
829 |
|
|
if (eventCount != ec || !w.isRunning() || |
830 |
|
|
runState >= TERMINATING) |
831 |
dl |
1.19 |
break; |
832 |
dl |
1.22 |
if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS) |
833 |
|
|
tryShutdownWaiter(ec); |
834 |
dl |
1.19 |
} |
835 |
dl |
1.14 |
} |
836 |
|
|
} |
837 |
dl |
1.22 |
} |
838 |
|
|
|
839 |
|
|
/** |
840 |
|
|
* Callback from the oldest waiter in awaitEvent waking up after a |
841 |
|
|
* period of non-use. Tries (once) to shutdown an event waiter (or |
842 |
|
|
* a spare, if one exists). Note that we don't need CAS or locks |
843 |
|
|
* here because the method is called only from one thread |
844 |
|
|
* occasionally waking (and even misfires are OK). Note that |
845 |
|
|
* until the shutdown worker fully terminates, workerCounts |
846 |
|
|
* will overestimate total count, which is tolerable. |
847 |
|
|
* |
848 |
|
|
* @param ec the event count waited on by caller (to abort |
849 |
|
|
* attempt if count has since changed). |
850 |
|
|
*/ |
851 |
|
|
private void tryShutdownWaiter(int ec) { |
852 |
|
|
if (spareWaiters != 0) { // prefer killing spares |
853 |
|
|
tryShutdownSpare(); |
854 |
|
|
return; |
855 |
|
|
} |
856 |
|
|
ForkJoinWorkerThread[] ws = workers; |
857 |
|
|
int n = ws.length; |
858 |
|
|
long h = eventWaiters; |
859 |
|
|
ForkJoinWorkerThread w; int id; long nh; |
860 |
|
|
if (runState == 0 && |
861 |
|
|
submissionQueue.isEmpty() && |
862 |
|
|
eventCount == ec && |
863 |
|
|
(id = ((int)(h & WAITER_ID_MASK)) - 1) >= 0 && |
864 |
|
|
id < n && (w = ws[id]) != null && |
865 |
|
|
(nh = w.nextWaiter) != 0L && // keep at least one worker |
866 |
|
|
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) { |
867 |
|
|
w.shutdown(); |
868 |
|
|
LockSupport.unpark(w); |
869 |
|
|
} |
870 |
|
|
releaseEventWaiters(); |
871 |
dl |
1.19 |
} |
872 |
|
|
|
873 |
|
|
// Maintaining spares |
874 |
|
|
|
875 |
|
|
/** |
876 |
|
|
* Pushes worker onto the spare stack |
877 |
|
|
*/ |
878 |
|
|
final void pushSpare(ForkJoinWorkerThread w) { |
879 |
dl |
1.22 |
int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1); |
880 |
dl |
1.19 |
do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
881 |
|
|
w.nextSpare = spareWaiters,ns)); |
882 |
dl |
1.14 |
} |
883 |
|
|
|
884 |
|
|
/** |
885 |
dl |
1.22 |
* Callback from oldest spare occasionally waking up. Tries |
886 |
|
|
* (once) to shutdown a spare. Same idea as tryShutdownWaiter. |
887 |
dl |
1.14 |
*/ |
888 |
dl |
1.22 |
final void tryShutdownSpare() { |
889 |
dl |
1.19 |
int sw, id; |
890 |
|
|
ForkJoinWorkerThread w; |
891 |
|
|
ForkJoinWorkerThread[] ws; |
892 |
|
|
if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 && |
893 |
|
|
id < (ws = workers).length && (w = ws[id]) != null && |
894 |
dl |
1.22 |
(workerCounts & RUNNING_COUNT_MASK) >= parallelism && |
895 |
dl |
1.19 |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
896 |
dl |
1.22 |
sw, w.nextSpare)) { |
897 |
|
|
w.shutdown(); |
898 |
dl |
1.19 |
LockSupport.unpark(w); |
899 |
dl |
1.22 |
advanceEventCount(); |
900 |
|
|
} |
901 |
|
|
} |
902 |
|
|
|
903 |
|
|
/** |
904 |
|
|
* Tries (once) to resume a spare if worker counts match |
905 |
|
|
* the given count. |
906 |
|
|
* |
907 |
|
|
* @param wc workerCounts value on invocation of this method |
908 |
|
|
*/ |
909 |
|
|
private void tryResumeSpare(int wc) { |
910 |
|
|
ForkJoinWorkerThread[] ws = workers; |
911 |
|
|
int n = ws.length; |
912 |
|
|
int sw, id, rs; ForkJoinWorkerThread w; |
913 |
|
|
if ((id = ((sw = spareWaiters) & SPARE_ID_MASK) - 1) >= 0 && |
914 |
|
|
id < n && (w = ws[id]) != null && |
915 |
|
|
(rs = runState) < TERMINATING && |
916 |
|
|
eventWaiters == 0L && workerCounts == wc) { |
917 |
|
|
// In case all workers busy, heuristically back off to let settle |
918 |
|
|
Thread.yield(); |
919 |
|
|
if (eventWaiters == 0L && runState == rs && // recheck |
920 |
|
|
workerCounts == wc && spareWaiters == sw && |
921 |
|
|
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
922 |
|
|
sw, w.nextSpare)) { |
923 |
|
|
int c; // increment running count before resume |
924 |
|
|
do {} while(!UNSAFE.compareAndSwapInt |
925 |
|
|
(this, workerCountsOffset, |
926 |
|
|
c = workerCounts, c + ONE_RUNNING)); |
927 |
|
|
if (w.tryUnsuspend()) |
928 |
|
|
LockSupport.unpark(w); |
929 |
|
|
else // back out if w was shutdown |
930 |
|
|
decrementWorkerCounts(ONE_RUNNING, 0); |
931 |
dl |
1.14 |
} |
932 |
|
|
} |
933 |
dl |
1.19 |
} |
934 |
|
|
|
935 |
dl |
1.22 |
// adding workers on demand |
936 |
|
|
|
937 |
dl |
1.19 |
/** |
938 |
dl |
1.22 |
* Adds one or more workers if needed to establish target parallelism. |
939 |
|
|
* Retries upon contention. |
940 |
|
|
*/ |
941 |
|
|
private void addWorkerIfBelowTarget() { |
942 |
|
|
int pc = parallelism; |
943 |
|
|
int wc; |
944 |
|
|
while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < pc && |
945 |
|
|
runState < TERMINATING) { |
946 |
|
|
if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
947 |
|
|
wc + (ONE_RUNNING|ONE_TOTAL))) { |
948 |
|
|
if (addWorker() == null) |
949 |
|
|
break; |
950 |
|
|
} |
951 |
|
|
} |
952 |
|
|
} |
953 |
|
|
|
954 |
|
|
/** |
955 |
|
|
* Tries (once) to add a new worker if all existing workers are |
956 |
|
|
* busy, and there are either no running workers or the deficit is |
957 |
|
|
* at least twice the surplus. |
958 |
|
|
* |
959 |
|
|
* @param wc workerCounts value on invocation of this method |
960 |
|
|
*/ |
961 |
|
|
private void tryAddWorkerIfBusy(int wc) { |
962 |
|
|
int tc, rc, rs; |
963 |
dl |
1.19 |
int pc = parallelism; |
964 |
dl |
1.22 |
if ((tc = wc >>> TOTAL_COUNT_SHIFT) < MAX_WORKERS && |
965 |
|
|
((rc = wc & RUNNING_COUNT_MASK) == 0 || |
966 |
|
|
rc < pc - ((tc - pc) << 1)) && |
967 |
|
|
(rs = runState) < TERMINATING && |
968 |
|
|
(rs & ACTIVE_COUNT_MASK) == tc) { |
969 |
|
|
// Since all workers busy, heuristically back off to let settle |
970 |
|
|
Thread.yield(); |
971 |
|
|
if (eventWaiters == 0L && spareWaiters == 0 && // recheck |
972 |
|
|
runState == rs && workerCounts == wc && |
973 |
|
|
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
974 |
|
|
wc + (ONE_RUNNING|ONE_TOTAL))) |
975 |
|
|
addWorker(); |
976 |
|
|
} |
977 |
dl |
1.19 |
} |
978 |
|
|
|
979 |
|
|
/** |
980 |
|
|
* Does at most one of: |
981 |
|
|
* |
982 |
|
|
* 1. Help wake up existing workers waiting for work via |
983 |
dl |
1.22 |
* releaseEventWaiters. (If any exist, then it doesn't |
984 |
dl |
1.19 |
* matter right now if under target parallelism level.) |
985 |
|
|
* |
986 |
dl |
1.22 |
* 2. If a spare exists, try (once) to resume it via tryResumeSpare. |
987 |
dl |
1.19 |
* |
988 |
dl |
1.22 |
* 3. If there are not enough total workers, add some |
989 |
|
|
* via addWorkerIfBelowTarget; |
990 |
|
|
* |
991 |
|
|
* 4. Try (once) to add a new worker if all existing workers |
992 |
|
|
* are busy, via tryAddWorkerIfBusy |
993 |
dl |
1.19 |
*/ |
994 |
|
|
private void helpMaintainParallelism() { |
995 |
dl |
1.22 |
long h; int pc, wc; |
996 |
|
|
if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0) { |
997 |
dl |
1.19 |
if ((int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
998 |
dl |
1.22 |
releaseEventWaiters(); // avoid useless call |
999 |
dl |
1.19 |
} |
1000 |
|
|
else if ((pc = parallelism) > |
1001 |
dl |
1.22 |
((wc = workerCounts) & RUNNING_COUNT_MASK)) { |
1002 |
dl |
1.19 |
if (spareWaiters != 0) |
1003 |
dl |
1.22 |
tryResumeSpare(wc); |
1004 |
|
|
else if ((wc >>> TOTAL_COUNT_SHIFT) < pc) |
1005 |
|
|
addWorkerIfBelowTarget(); |
1006 |
|
|
else |
1007 |
|
|
tryAddWorkerIfBusy(wc); |
1008 |
dl |
1.19 |
} |
1009 |
dl |
1.14 |
} |
1010 |
|
|
|
1011 |
|
|
/** |
1012 |
|
|
* Callback from workers invoked upon each top-level action (i.e., |
1013 |
dl |
1.22 |
* stealing a task or taking a submission and running it). |
1014 |
|
|
* Performs one or more of the following: |
1015 |
dl |
1.19 |
* |
1016 |
dl |
1.22 |
* 1. If the worker is active, try to set its active status to |
1017 |
|
|
* inactive and update activeCount. On contention, we may try |
1018 |
|
|
* again on this or subsequent call. |
1019 |
dl |
1.19 |
* |
1020 |
dl |
1.22 |
* 2. Release any existing event waiters that are now relesable |
1021 |
dl |
1.19 |
* |
1022 |
dl |
1.22 |
* 3. If there are too many running threads, suspend this worker |
1023 |
|
|
* (first forcing inactive if necessary). If it is not |
1024 |
dl |
1.19 |
* needed, it may be killed while suspended via |
1025 |
dl |
1.22 |
* tryShutdownSpare. Otherwise, upon resume it rechecks to make |
1026 |
dl |
1.19 |
* sure that it is still needed. |
1027 |
dl |
1.14 |
* |
1028 |
dl |
1.22 |
* 4. If more than 1 miss, await the next task event via |
1029 |
|
|
* eventSync (first forcing inactivation if necessary), upon |
1030 |
|
|
* which worker may also be killed, via tryShutdownWaiter. |
1031 |
|
|
* |
1032 |
|
|
* 5. Help reactivate other workers via helpMaintainParallelism |
1033 |
dl |
1.14 |
* |
1034 |
|
|
* @param w the worker |
1035 |
dl |
1.19 |
* @param misses the number of scans by caller failing to find work |
1036 |
dl |
1.22 |
* (saturating at 2 to avoid wraparound) |
1037 |
dl |
1.14 |
*/ |
1038 |
dl |
1.19 |
final void preStep(ForkJoinWorkerThread w, int misses) { |
1039 |
dl |
1.14 |
boolean active = w.active; |
1040 |
dl |
1.19 |
int pc = parallelism; |
1041 |
dl |
1.14 |
for (;;) { |
1042 |
dl |
1.22 |
int rs, wc, rc, ec; long h; |
1043 |
|
|
if (active && UNSAFE.compareAndSwapInt(this, runStateOffset, |
1044 |
|
|
rs = runState, rs - 1)) |
1045 |
|
|
active = w.active = false; |
1046 |
|
|
if (((int)((h = eventWaiters) & WAITER_ID_MASK)) != 0 && |
1047 |
|
|
(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) { |
1048 |
|
|
releaseEventWaiters(); |
1049 |
|
|
if (misses > 1) |
1050 |
|
|
continue; // clear before sync below |
1051 |
dl |
1.19 |
} |
1052 |
dl |
1.22 |
if ((rc = ((wc = workerCounts) & RUNNING_COUNT_MASK)) > pc) { |
1053 |
|
|
if (!active && // must inactivate to suspend |
1054 |
|
|
workerCounts == wc && // try to suspend as spare |
1055 |
dl |
1.19 |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1056 |
dl |
1.22 |
wc, wc - ONE_RUNNING)) { |
1057 |
|
|
w.suspendAsSpare(); |
1058 |
|
|
if (!w.isRunning()) |
1059 |
|
|
break; // was killed while spare |
1060 |
|
|
} |
1061 |
|
|
continue; |
1062 |
dl |
1.19 |
} |
1063 |
dl |
1.22 |
if (misses > 0) { |
1064 |
|
|
if ((ec = eventCount) == w.lastEventCount && misses > 1) { |
1065 |
|
|
if (!active) { // must inactivate to sync |
1066 |
|
|
eventSync(w); |
1067 |
|
|
if (w.isRunning()) |
1068 |
|
|
misses = 1; // don't re-sync |
1069 |
|
|
else |
1070 |
|
|
break; // was killed while waiting |
1071 |
|
|
} |
1072 |
|
|
continue; |
1073 |
|
|
} |
1074 |
|
|
w.lastEventCount = ec; |
1075 |
dl |
1.14 |
} |
1076 |
dl |
1.22 |
if (rc < pc) |
1077 |
|
|
helpMaintainParallelism(); |
1078 |
|
|
break; |
1079 |
dl |
1.14 |
} |
1080 |
|
|
} |
1081 |
|
|
|
1082 |
|
|
/** |
1083 |
dl |
1.19 |
* Helps and/or blocks awaiting join of the given task. |
1084 |
|
|
* Alternates between helpJoinTask() and helpMaintainParallelism() |
1085 |
|
|
* as many times as there is a deficit in running count (or longer |
1086 |
|
|
* if running count would become zero), then blocks if task still |
1087 |
|
|
* not done. |
1088 |
dl |
1.17 |
* |
1089 |
|
|
* @param joinMe the task to join |
1090 |
dl |
1.14 |
*/ |
1091 |
dl |
1.19 |
final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker) { |
1092 |
|
|
int threshold = parallelism; // descend blocking thresholds |
1093 |
|
|
while (joinMe.status >= 0) { |
1094 |
|
|
boolean block; int wc; |
1095 |
|
|
worker.helpJoinTask(joinMe); |
1096 |
|
|
if (joinMe.status < 0) |
1097 |
|
|
break; |
1098 |
|
|
if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) { |
1099 |
|
|
if (threshold > 0) |
1100 |
|
|
--threshold; |
1101 |
|
|
else |
1102 |
|
|
advanceEventCount(); // force release |
1103 |
|
|
block = false; |
1104 |
|
|
} |
1105 |
|
|
else |
1106 |
|
|
block = UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1107 |
|
|
wc, wc - ONE_RUNNING); |
1108 |
|
|
helpMaintainParallelism(); |
1109 |
|
|
if (block) { |
1110 |
|
|
int c; |
1111 |
|
|
joinMe.internalAwaitDone(); |
1112 |
|
|
do {} while (!UNSAFE.compareAndSwapInt |
1113 |
|
|
(this, workerCountsOffset, |
1114 |
|
|
c = workerCounts, c + ONE_RUNNING)); |
1115 |
|
|
break; |
1116 |
dl |
1.14 |
} |
1117 |
|
|
} |
1118 |
|
|
} |
1119 |
|
|
|
1120 |
|
|
/** |
1121 |
dl |
1.19 |
* Same idea as awaitJoin, but no helping |
1122 |
dl |
1.14 |
*/ |
1123 |
dl |
1.18 |
final void awaitBlocker(ManagedBlocker blocker) |
1124 |
dl |
1.14 |
throws InterruptedException { |
1125 |
dl |
1.19 |
int threshold = parallelism; |
1126 |
|
|
while (!blocker.isReleasable()) { |
1127 |
|
|
boolean block; int wc; |
1128 |
|
|
if (((wc = workerCounts) & RUNNING_COUNT_MASK) <= threshold) { |
1129 |
|
|
if (threshold > 0) |
1130 |
|
|
--threshold; |
1131 |
|
|
else |
1132 |
|
|
advanceEventCount(); |
1133 |
|
|
block = false; |
1134 |
dl |
1.14 |
} |
1135 |
dl |
1.19 |
else |
1136 |
|
|
block = UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1137 |
|
|
wc, wc - ONE_RUNNING); |
1138 |
|
|
helpMaintainParallelism(); |
1139 |
|
|
if (block) { |
1140 |
|
|
try { |
1141 |
|
|
do {} while (!blocker.isReleasable() && !blocker.block()); |
1142 |
|
|
} finally { |
1143 |
dl |
1.14 |
int c; |
1144 |
dl |
1.17 |
do {} while (!UNSAFE.compareAndSwapInt |
1145 |
|
|
(this, workerCountsOffset, |
1146 |
|
|
c = workerCounts, c + ONE_RUNNING)); |
1147 |
dl |
1.14 |
} |
1148 |
dl |
1.18 |
break; |
1149 |
|
|
} |
1150 |
dl |
1.14 |
} |
1151 |
dl |
1.19 |
} |
1152 |
dl |
1.15 |
|
1153 |
|
|
/** |
1154 |
dl |
1.14 |
* Possibly initiates and/or completes termination. |
1155 |
|
|
* |
1156 |
|
|
* @param now if true, unconditionally terminate, else only |
1157 |
|
|
* if shutdown and empty queue and no active workers |
1158 |
|
|
* @return true if now terminating or terminated |
1159 |
jsr166 |
1.1 |
*/ |
1160 |
dl |
1.14 |
private boolean tryTerminate(boolean now) { |
1161 |
|
|
if (now) |
1162 |
|
|
advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN |
1163 |
|
|
else if (runState < SHUTDOWN || |
1164 |
|
|
!submissionQueue.isEmpty() || |
1165 |
|
|
(runState & ACTIVE_COUNT_MASK) != 0) |
1166 |
jsr166 |
1.1 |
return false; |
1167 |
dl |
1.14 |
|
1168 |
|
|
if (advanceRunLevel(TERMINATING)) |
1169 |
|
|
startTerminating(); |
1170 |
|
|
|
1171 |
|
|
// Finish now if all threads terminated; else in some subsequent call |
1172 |
|
|
if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) { |
1173 |
|
|
advanceRunLevel(TERMINATED); |
1174 |
dl |
1.18 |
termination.arrive(); |
1175 |
dl |
1.14 |
} |
1176 |
jsr166 |
1.1 |
return true; |
1177 |
|
|
} |
1178 |
|
|
|
1179 |
|
|
/** |
1180 |
dl |
1.14 |
* Actions on transition to TERMINATING |
1181 |
dl |
1.19 |
* |
1182 |
|
|
* Runs up to four passes through workers: (0) shutting down each |
1183 |
dl |
1.22 |
* (without waking up if parked) to quickly spread notifications |
1184 |
|
|
* without unnecessary bouncing around event queues etc (1) wake |
1185 |
|
|
* up and help cancel tasks (2) interrupt (3) mop up races with |
1186 |
|
|
* interrupted workers |
1187 |
dl |
1.14 |
*/ |
1188 |
|
|
private void startTerminating() { |
1189 |
dl |
1.19 |
cancelSubmissions(); |
1190 |
|
|
for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) { |
1191 |
|
|
advanceEventCount(); |
1192 |
|
|
eventWaiters = 0L; // clobber lists |
1193 |
|
|
spareWaiters = 0; |
1194 |
|
|
ForkJoinWorkerThread[] ws = workers; |
1195 |
|
|
int n = ws.length; |
1196 |
|
|
for (int i = 0; i < n; ++i) { |
1197 |
|
|
ForkJoinWorkerThread w = ws[i]; |
1198 |
|
|
if (w != null) { |
1199 |
dl |
1.22 |
w.shutdown(); |
1200 |
dl |
1.19 |
if (passes > 0 && !w.isTerminated()) { |
1201 |
|
|
w.cancelTasks(); |
1202 |
|
|
LockSupport.unpark(w); |
1203 |
|
|
if (passes > 1) { |
1204 |
|
|
try { |
1205 |
|
|
w.interrupt(); |
1206 |
|
|
} catch (SecurityException ignore) { |
1207 |
|
|
} |
1208 |
|
|
} |
1209 |
|
|
} |
1210 |
|
|
} |
1211 |
|
|
} |
1212 |
dl |
1.17 |
} |
1213 |
|
|
} |
1214 |
|
|
|
1215 |
|
|
/** |
1216 |
|
|
* Clear out and cancel submissions, ignoring exceptions |
1217 |
|
|
*/ |
1218 |
|
|
private void cancelSubmissions() { |
1219 |
dl |
1.14 |
ForkJoinTask<?> task; |
1220 |
|
|
while ((task = submissionQueue.poll()) != null) { |
1221 |
|
|
try { |
1222 |
|
|
task.cancel(false); |
1223 |
|
|
} catch (Throwable ignore) { |
1224 |
|
|
} |
1225 |
|
|
} |
1226 |
dl |
1.17 |
} |
1227 |
|
|
|
1228 |
dl |
1.14 |
// misc support for ForkJoinWorkerThread |
1229 |
|
|
|
1230 |
|
|
/** |
1231 |
|
|
* Returns pool number |
1232 |
jsr166 |
1.1 |
*/ |
1233 |
dl |
1.14 |
final int getPoolNumber() { |
1234 |
|
|
return poolNumber; |
1235 |
jsr166 |
1.1 |
} |
1236 |
|
|
|
1237 |
|
|
/** |
1238 |
dl |
1.19 |
* Tries to accumulates steal count from a worker, clearing |
1239 |
|
|
* the worker's value. |
1240 |
|
|
* |
1241 |
|
|
* @return true if worker steal count now zero |
1242 |
jsr166 |
1.1 |
*/ |
1243 |
dl |
1.19 |
final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) { |
1244 |
dl |
1.14 |
int sc = w.stealCount; |
1245 |
dl |
1.19 |
long c = stealCount; |
1246 |
|
|
// CAS even if zero, for fence effects |
1247 |
|
|
if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) { |
1248 |
|
|
if (sc != 0) |
1249 |
|
|
w.stealCount = 0; |
1250 |
|
|
return true; |
1251 |
jsr166 |
1.1 |
} |
1252 |
dl |
1.19 |
return sc == 0; |
1253 |
jsr166 |
1.1 |
} |
1254 |
|
|
|
1255 |
|
|
/** |
1256 |
dl |
1.14 |
* Returns the approximate (non-atomic) number of idle threads per |
1257 |
|
|
* active thread. |
1258 |
|
|
*/ |
1259 |
|
|
final int idlePerActive() { |
1260 |
dl |
1.19 |
int pc = parallelism; // use parallelism, not rc |
1261 |
dl |
1.14 |
int ac = runState; // no mask -- artifically boosts during shutdown |
1262 |
|
|
// Use exact results for small values, saturate past 4 |
1263 |
|
|
return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3; |
1264 |
|
|
} |
1265 |
|
|
|
1266 |
|
|
// Public and protected methods |
1267 |
jsr166 |
1.1 |
|
1268 |
|
|
// Constructors |
1269 |
|
|
|
1270 |
|
|
/** |
1271 |
jsr166 |
1.9 |
* Creates a {@code ForkJoinPool} with parallelism equal to {@link |
1272 |
dl |
1.18 |
* java.lang.Runtime#availableProcessors}, using the {@linkplain |
1273 |
|
|
* #defaultForkJoinWorkerThreadFactory default thread factory}, |
1274 |
|
|
* no UncaughtExceptionHandler, and non-async LIFO processing mode. |
1275 |
jsr166 |
1.1 |
* |
1276 |
|
|
* @throws SecurityException if a security manager exists and |
1277 |
|
|
* the caller is not permitted to modify threads |
1278 |
|
|
* because it does not hold {@link |
1279 |
|
|
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1280 |
|
|
*/ |
1281 |
|
|
public ForkJoinPool() { |
1282 |
|
|
this(Runtime.getRuntime().availableProcessors(), |
1283 |
dl |
1.18 |
defaultForkJoinWorkerThreadFactory, null, false); |
1284 |
jsr166 |
1.1 |
} |
1285 |
|
|
|
1286 |
|
|
/** |
1287 |
jsr166 |
1.9 |
* Creates a {@code ForkJoinPool} with the indicated parallelism |
1288 |
dl |
1.18 |
* level, the {@linkplain |
1289 |
|
|
* #defaultForkJoinWorkerThreadFactory default thread factory}, |
1290 |
|
|
* no UncaughtExceptionHandler, and non-async LIFO processing mode. |
1291 |
jsr166 |
1.1 |
* |
1292 |
jsr166 |
1.9 |
* @param parallelism the parallelism level |
1293 |
jsr166 |
1.1 |
* @throws IllegalArgumentException if parallelism less than or |
1294 |
jsr166 |
1.11 |
* equal to zero, or greater than implementation limit |
1295 |
jsr166 |
1.1 |
* @throws SecurityException if a security manager exists and |
1296 |
|
|
* the caller is not permitted to modify threads |
1297 |
|
|
* because it does not hold {@link |
1298 |
|
|
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1299 |
|
|
*/ |
1300 |
|
|
public ForkJoinPool(int parallelism) { |
1301 |
dl |
1.18 |
this(parallelism, defaultForkJoinWorkerThreadFactory, null, false); |
1302 |
jsr166 |
1.1 |
} |
1303 |
|
|
|
1304 |
|
|
/** |
1305 |
dl |
1.18 |
* Creates a {@code ForkJoinPool} with the given parameters. |
1306 |
jsr166 |
1.1 |
* |
1307 |
dl |
1.18 |
* @param parallelism the parallelism level. For default value, |
1308 |
|
|
* use {@link java.lang.Runtime#availableProcessors}. |
1309 |
|
|
* @param factory the factory for creating new threads. For default value, |
1310 |
|
|
* use {@link #defaultForkJoinWorkerThreadFactory}. |
1311 |
dl |
1.19 |
* @param handler the handler for internal worker threads that |
1312 |
|
|
* terminate due to unrecoverable errors encountered while executing |
1313 |
dl |
1.18 |
* tasks. For default value, use <code>null</code>. |
1314 |
dl |
1.19 |
* @param asyncMode if true, |
1315 |
dl |
1.18 |
* establishes local first-in-first-out scheduling mode for forked |
1316 |
|
|
* tasks that are never joined. This mode may be more appropriate |
1317 |
|
|
* than default locally stack-based mode in applications in which |
1318 |
|
|
* worker threads only process event-style asynchronous tasks. |
1319 |
|
|
* For default value, use <code>false</code>. |
1320 |
jsr166 |
1.1 |
* @throws IllegalArgumentException if parallelism less than or |
1321 |
jsr166 |
1.11 |
* equal to zero, or greater than implementation limit |
1322 |
|
|
* @throws NullPointerException if the factory is null |
1323 |
jsr166 |
1.1 |
* @throws SecurityException if a security manager exists and |
1324 |
|
|
* the caller is not permitted to modify threads |
1325 |
|
|
* because it does not hold {@link |
1326 |
|
|
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1327 |
|
|
*/ |
1328 |
dl |
1.19 |
public ForkJoinPool(int parallelism, |
1329 |
dl |
1.18 |
ForkJoinWorkerThreadFactory factory, |
1330 |
|
|
Thread.UncaughtExceptionHandler handler, |
1331 |
|
|
boolean asyncMode) { |
1332 |
dl |
1.14 |
checkPermission(); |
1333 |
|
|
if (factory == null) |
1334 |
|
|
throw new NullPointerException(); |
1335 |
dl |
1.19 |
if (parallelism <= 0 || parallelism > MAX_WORKERS) |
1336 |
jsr166 |
1.1 |
throw new IllegalArgumentException(); |
1337 |
dl |
1.14 |
this.parallelism = parallelism; |
1338 |
jsr166 |
1.1 |
this.factory = factory; |
1339 |
dl |
1.18 |
this.ueh = handler; |
1340 |
|
|
this.locallyFifo = asyncMode; |
1341 |
|
|
int arraySize = initialArraySizeFor(parallelism); |
1342 |
dl |
1.14 |
this.workers = new ForkJoinWorkerThread[arraySize]; |
1343 |
|
|
this.submissionQueue = new LinkedTransferQueue<ForkJoinTask<?>>(); |
1344 |
jsr166 |
1.1 |
this.workerLock = new ReentrantLock(); |
1345 |
dl |
1.18 |
this.termination = new Phaser(1); |
1346 |
|
|
this.poolNumber = poolNumberGenerator.incrementAndGet(); |
1347 |
jsr166 |
1.1 |
} |
1348 |
|
|
|
1349 |
|
|
/** |
1350 |
dl |
1.14 |
* Returns initial power of two size for workers array. |
1351 |
|
|
* @param pc the initial parallelism level |
1352 |
|
|
*/ |
1353 |
|
|
private static int initialArraySizeFor(int pc) { |
1354 |
dl |
1.19 |
// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16) |
1355 |
|
|
int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS; |
1356 |
dl |
1.14 |
size |= size >>> 1; |
1357 |
|
|
size |= size >>> 2; |
1358 |
|
|
size |= size >>> 4; |
1359 |
|
|
size |= size >>> 8; |
1360 |
|
|
return size + 1; |
1361 |
jsr166 |
1.1 |
} |
1362 |
|
|
|
1363 |
|
|
// Execution methods |
1364 |
|
|
|
1365 |
|
|
/** |
1366 |
|
|
* Common code for execute, invoke and submit |
1367 |
|
|
*/ |
1368 |
|
|
private <T> void doSubmit(ForkJoinTask<T> task) { |
1369 |
jsr166 |
1.2 |
if (task == null) |
1370 |
|
|
throw new NullPointerException(); |
1371 |
dl |
1.14 |
if (runState >= SHUTDOWN) |
1372 |
jsr166 |
1.1 |
throw new RejectedExecutionException(); |
1373 |
dl |
1.19 |
submissionQueue.offer(task); |
1374 |
|
|
advanceEventCount(); |
1375 |
dl |
1.22 |
if (eventWaiters != 0L) |
1376 |
|
|
releaseEventWaiters(); |
1377 |
|
|
if ((workerCounts >>> TOTAL_COUNT_SHIFT) < parallelism) |
1378 |
|
|
addWorkerIfBelowTarget(); |
1379 |
jsr166 |
1.1 |
} |
1380 |
|
|
|
1381 |
|
|
/** |
1382 |
|
|
* Performs the given task, returning its result upon completion. |
1383 |
|
|
* |
1384 |
|
|
* @param task the task |
1385 |
|
|
* @return the task's result |
1386 |
jsr166 |
1.11 |
* @throws NullPointerException if the task is null |
1387 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1388 |
|
|
* scheduled for execution |
1389 |
jsr166 |
1.1 |
*/ |
1390 |
|
|
public <T> T invoke(ForkJoinTask<T> task) { |
1391 |
|
|
doSubmit(task); |
1392 |
|
|
return task.join(); |
1393 |
|
|
} |
1394 |
|
|
|
1395 |
|
|
/** |
1396 |
|
|
* Arranges for (asynchronous) execution of the given task. |
1397 |
|
|
* |
1398 |
|
|
* @param task the task |
1399 |
jsr166 |
1.11 |
* @throws NullPointerException if the task is null |
1400 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1401 |
|
|
* scheduled for execution |
1402 |
jsr166 |
1.1 |
*/ |
1403 |
jsr166 |
1.8 |
public void execute(ForkJoinTask<?> task) { |
1404 |
jsr166 |
1.1 |
doSubmit(task); |
1405 |
|
|
} |
1406 |
|
|
|
1407 |
|
|
// AbstractExecutorService methods |
1408 |
|
|
|
1409 |
jsr166 |
1.11 |
/** |
1410 |
|
|
* @throws NullPointerException if the task is null |
1411 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1412 |
|
|
* scheduled for execution |
1413 |
|
|
*/ |
1414 |
jsr166 |
1.1 |
public void execute(Runnable task) { |
1415 |
jsr166 |
1.2 |
ForkJoinTask<?> job; |
1416 |
jsr166 |
1.3 |
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1417 |
|
|
job = (ForkJoinTask<?>) task; |
1418 |
jsr166 |
1.2 |
else |
1419 |
jsr166 |
1.7 |
job = ForkJoinTask.adapt(task, null); |
1420 |
jsr166 |
1.2 |
doSubmit(job); |
1421 |
jsr166 |
1.1 |
} |
1422 |
|
|
|
1423 |
jsr166 |
1.11 |
/** |
1424 |
dl |
1.18 |
* 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 |
|
|
doSubmit(task); |
1434 |
|
|
return task; |
1435 |
|
|
} |
1436 |
|
|
|
1437 |
|
|
/** |
1438 |
jsr166 |
1.11 |
* @throws NullPointerException if the task is null |
1439 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1440 |
|
|
* scheduled for execution |
1441 |
|
|
*/ |
1442 |
jsr166 |
1.1 |
public <T> ForkJoinTask<T> submit(Callable<T> task) { |
1443 |
jsr166 |
1.7 |
ForkJoinTask<T> job = ForkJoinTask.adapt(task); |
1444 |
jsr166 |
1.1 |
doSubmit(job); |
1445 |
|
|
return job; |
1446 |
|
|
} |
1447 |
|
|
|
1448 |
jsr166 |
1.11 |
/** |
1449 |
|
|
* @throws NullPointerException if the task is null |
1450 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1451 |
|
|
* scheduled for execution |
1452 |
|
|
*/ |
1453 |
jsr166 |
1.1 |
public <T> ForkJoinTask<T> submit(Runnable task, T result) { |
1454 |
jsr166 |
1.7 |
ForkJoinTask<T> job = ForkJoinTask.adapt(task, result); |
1455 |
jsr166 |
1.1 |
doSubmit(job); |
1456 |
|
|
return job; |
1457 |
|
|
} |
1458 |
|
|
|
1459 |
jsr166 |
1.11 |
/** |
1460 |
|
|
* @throws NullPointerException if the task is null |
1461 |
|
|
* @throws RejectedExecutionException if the task cannot be |
1462 |
|
|
* scheduled for execution |
1463 |
|
|
*/ |
1464 |
jsr166 |
1.1 |
public ForkJoinTask<?> submit(Runnable task) { |
1465 |
jsr166 |
1.2 |
ForkJoinTask<?> job; |
1466 |
jsr166 |
1.3 |
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1467 |
|
|
job = (ForkJoinTask<?>) task; |
1468 |
jsr166 |
1.2 |
else |
1469 |
jsr166 |
1.7 |
job = ForkJoinTask.adapt(task, null); |
1470 |
jsr166 |
1.1 |
doSubmit(job); |
1471 |
|
|
return job; |
1472 |
|
|
} |
1473 |
|
|
|
1474 |
|
|
/** |
1475 |
jsr166 |
1.11 |
* @throws NullPointerException {@inheritDoc} |
1476 |
|
|
* @throws RejectedExecutionException {@inheritDoc} |
1477 |
|
|
*/ |
1478 |
jsr166 |
1.1 |
public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) { |
1479 |
|
|
ArrayList<ForkJoinTask<T>> forkJoinTasks = |
1480 |
|
|
new ArrayList<ForkJoinTask<T>>(tasks.size()); |
1481 |
|
|
for (Callable<T> task : tasks) |
1482 |
jsr166 |
1.7 |
forkJoinTasks.add(ForkJoinTask.adapt(task)); |
1483 |
jsr166 |
1.1 |
invoke(new InvokeAll<T>(forkJoinTasks)); |
1484 |
|
|
|
1485 |
|
|
@SuppressWarnings({"unchecked", "rawtypes"}) |
1486 |
dl |
1.15 |
List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks; |
1487 |
jsr166 |
1.1 |
return futures; |
1488 |
|
|
} |
1489 |
|
|
|
1490 |
|
|
static final class InvokeAll<T> extends RecursiveAction { |
1491 |
|
|
final ArrayList<ForkJoinTask<T>> tasks; |
1492 |
|
|
InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; } |
1493 |
|
|
public void compute() { |
1494 |
|
|
try { invokeAll(tasks); } |
1495 |
|
|
catch (Exception ignore) {} |
1496 |
|
|
} |
1497 |
|
|
private static final long serialVersionUID = -7914297376763021607L; |
1498 |
|
|
} |
1499 |
|
|
|
1500 |
|
|
/** |
1501 |
|
|
* Returns the factory used for constructing new workers. |
1502 |
|
|
* |
1503 |
|
|
* @return the factory used for constructing new workers |
1504 |
|
|
*/ |
1505 |
|
|
public ForkJoinWorkerThreadFactory getFactory() { |
1506 |
|
|
return factory; |
1507 |
|
|
} |
1508 |
|
|
|
1509 |
|
|
/** |
1510 |
|
|
* Returns the handler for internal worker threads that terminate |
1511 |
|
|
* due to unrecoverable errors encountered while executing tasks. |
1512 |
|
|
* |
1513 |
jsr166 |
1.4 |
* @return the handler, or {@code null} if none |
1514 |
jsr166 |
1.1 |
*/ |
1515 |
|
|
public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() { |
1516 |
dl |
1.14 |
return ueh; |
1517 |
jsr166 |
1.1 |
} |
1518 |
|
|
|
1519 |
|
|
/** |
1520 |
jsr166 |
1.9 |
* Returns the targeted parallelism level of this pool. |
1521 |
jsr166 |
1.1 |
* |
1522 |
jsr166 |
1.9 |
* @return the targeted parallelism level of this pool |
1523 |
jsr166 |
1.1 |
*/ |
1524 |
|
|
public int getParallelism() { |
1525 |
|
|
return parallelism; |
1526 |
|
|
} |
1527 |
|
|
|
1528 |
|
|
/** |
1529 |
|
|
* Returns the number of worker threads that have started but not |
1530 |
|
|
* yet terminated. This result returned by this method may differ |
1531 |
jsr166 |
1.4 |
* from {@link #getParallelism} when threads are created to |
1532 |
jsr166 |
1.1 |
* maintain parallelism when others are cooperatively blocked. |
1533 |
|
|
* |
1534 |
|
|
* @return the number of worker threads |
1535 |
|
|
*/ |
1536 |
|
|
public int getPoolSize() { |
1537 |
dl |
1.14 |
return workerCounts >>> TOTAL_COUNT_SHIFT; |
1538 |
jsr166 |
1.1 |
} |
1539 |
|
|
|
1540 |
|
|
/** |
1541 |
jsr166 |
1.4 |
* Returns {@code true} if this pool uses local first-in-first-out |
1542 |
jsr166 |
1.1 |
* scheduling mode for forked tasks that are never joined. |
1543 |
|
|
* |
1544 |
jsr166 |
1.4 |
* @return {@code true} if this pool uses async mode |
1545 |
jsr166 |
1.1 |
*/ |
1546 |
|
|
public boolean getAsyncMode() { |
1547 |
|
|
return locallyFifo; |
1548 |
|
|
} |
1549 |
|
|
|
1550 |
|
|
/** |
1551 |
|
|
* Returns an estimate of the number of worker threads that are |
1552 |
|
|
* not blocked waiting to join tasks or for other managed |
1553 |
dl |
1.14 |
* synchronization. This method may overestimate the |
1554 |
|
|
* number of running threads. |
1555 |
jsr166 |
1.1 |
* |
1556 |
|
|
* @return the number of worker threads |
1557 |
|
|
*/ |
1558 |
|
|
public int getRunningThreadCount() { |
1559 |
dl |
1.14 |
return workerCounts & RUNNING_COUNT_MASK; |
1560 |
jsr166 |
1.1 |
} |
1561 |
|
|
|
1562 |
|
|
/** |
1563 |
|
|
* Returns an estimate of the number of threads that are currently |
1564 |
|
|
* stealing or executing tasks. This method may overestimate the |
1565 |
|
|
* number of active threads. |
1566 |
|
|
* |
1567 |
|
|
* @return the number of active threads |
1568 |
|
|
*/ |
1569 |
|
|
public int getActiveThreadCount() { |
1570 |
dl |
1.14 |
return runState & ACTIVE_COUNT_MASK; |
1571 |
jsr166 |
1.1 |
} |
1572 |
|
|
|
1573 |
|
|
/** |
1574 |
jsr166 |
1.4 |
* Returns {@code true} if all worker threads are currently idle. |
1575 |
|
|
* An idle worker is one that cannot obtain a task to execute |
1576 |
|
|
* because none are available to steal from other threads, and |
1577 |
|
|
* there are no pending submissions to the pool. This method is |
1578 |
|
|
* conservative; it might not return {@code true} immediately upon |
1579 |
|
|
* idleness of all threads, but will eventually become true if |
1580 |
|
|
* threads remain inactive. |
1581 |
jsr166 |
1.1 |
* |
1582 |
jsr166 |
1.4 |
* @return {@code true} if all threads are currently idle |
1583 |
jsr166 |
1.1 |
*/ |
1584 |
|
|
public boolean isQuiescent() { |
1585 |
dl |
1.14 |
return (runState & ACTIVE_COUNT_MASK) == 0; |
1586 |
jsr166 |
1.1 |
} |
1587 |
|
|
|
1588 |
|
|
/** |
1589 |
|
|
* Returns an estimate of the total number of tasks stolen from |
1590 |
|
|
* one thread's work queue by another. The reported value |
1591 |
|
|
* underestimates the actual total number of steals when the pool |
1592 |
|
|
* is not quiescent. This value may be useful for monitoring and |
1593 |
|
|
* tuning fork/join programs: in general, steal counts should be |
1594 |
|
|
* high enough to keep threads busy, but low enough to avoid |
1595 |
|
|
* overhead and contention across threads. |
1596 |
|
|
* |
1597 |
|
|
* @return the number of steals |
1598 |
|
|
*/ |
1599 |
|
|
public long getStealCount() { |
1600 |
dl |
1.14 |
return stealCount; |
1601 |
jsr166 |
1.1 |
} |
1602 |
|
|
|
1603 |
|
|
/** |
1604 |
|
|
* Returns an estimate of the total number of tasks currently held |
1605 |
|
|
* in queues by worker threads (but not including tasks submitted |
1606 |
|
|
* to the pool that have not begun executing). This value is only |
1607 |
|
|
* an approximation, obtained by iterating across all threads in |
1608 |
|
|
* the pool. This method may be useful for tuning task |
1609 |
|
|
* granularities. |
1610 |
|
|
* |
1611 |
|
|
* @return the number of queued tasks |
1612 |
|
|
*/ |
1613 |
|
|
public long getQueuedTaskCount() { |
1614 |
|
|
long count = 0; |
1615 |
dl |
1.17 |
ForkJoinWorkerThread[] ws = workers; |
1616 |
dl |
1.19 |
int n = ws.length; |
1617 |
|
|
for (int i = 0; i < n; ++i) { |
1618 |
dl |
1.17 |
ForkJoinWorkerThread w = ws[i]; |
1619 |
dl |
1.14 |
if (w != null) |
1620 |
|
|
count += w.getQueueSize(); |
1621 |
jsr166 |
1.1 |
} |
1622 |
|
|
return count; |
1623 |
|
|
} |
1624 |
|
|
|
1625 |
|
|
/** |
1626 |
jsr166 |
1.8 |
* Returns an estimate of the number of tasks submitted to this |
1627 |
|
|
* pool that have not yet begun executing. This method takes time |
1628 |
jsr166 |
1.1 |
* proportional to the number of submissions. |
1629 |
|
|
* |
1630 |
|
|
* @return the number of queued submissions |
1631 |
|
|
*/ |
1632 |
|
|
public int getQueuedSubmissionCount() { |
1633 |
|
|
return submissionQueue.size(); |
1634 |
|
|
} |
1635 |
|
|
|
1636 |
|
|
/** |
1637 |
jsr166 |
1.4 |
* Returns {@code true} if there are any tasks submitted to this |
1638 |
|
|
* pool that have not yet begun executing. |
1639 |
jsr166 |
1.1 |
* |
1640 |
|
|
* @return {@code true} if there are any queued submissions |
1641 |
|
|
*/ |
1642 |
|
|
public boolean hasQueuedSubmissions() { |
1643 |
|
|
return !submissionQueue.isEmpty(); |
1644 |
|
|
} |
1645 |
|
|
|
1646 |
|
|
/** |
1647 |
|
|
* Removes and returns the next unexecuted submission if one is |
1648 |
|
|
* available. This method may be useful in extensions to this |
1649 |
|
|
* class that re-assign work in systems with multiple pools. |
1650 |
|
|
* |
1651 |
jsr166 |
1.4 |
* @return the next submission, or {@code null} if none |
1652 |
jsr166 |
1.1 |
*/ |
1653 |
|
|
protected ForkJoinTask<?> pollSubmission() { |
1654 |
|
|
return submissionQueue.poll(); |
1655 |
|
|
} |
1656 |
|
|
|
1657 |
|
|
/** |
1658 |
|
|
* Removes all available unexecuted submitted and forked tasks |
1659 |
|
|
* from scheduling queues and adds them to the given collection, |
1660 |
|
|
* without altering their execution status. These may include |
1661 |
jsr166 |
1.8 |
* artificially generated or wrapped tasks. This method is |
1662 |
|
|
* designed to be invoked only when the pool is known to be |
1663 |
jsr166 |
1.1 |
* quiescent. Invocations at other times may not remove all |
1664 |
|
|
* tasks. A failure encountered while attempting to add elements |
1665 |
|
|
* to collection {@code c} may result in elements being in |
1666 |
|
|
* neither, either or both collections when the associated |
1667 |
|
|
* exception is thrown. The behavior of this operation is |
1668 |
|
|
* undefined if the specified collection is modified while the |
1669 |
|
|
* operation is in progress. |
1670 |
|
|
* |
1671 |
|
|
* @param c the collection to transfer elements into |
1672 |
|
|
* @return the number of elements transferred |
1673 |
|
|
*/ |
1674 |
jsr166 |
1.5 |
protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) { |
1675 |
dl |
1.19 |
int count = submissionQueue.drainTo(c); |
1676 |
dl |
1.18 |
ForkJoinWorkerThread[] ws = workers; |
1677 |
dl |
1.19 |
int n = ws.length; |
1678 |
|
|
for (int i = 0; i < n; ++i) { |
1679 |
dl |
1.18 |
ForkJoinWorkerThread w = ws[i]; |
1680 |
|
|
if (w != null) |
1681 |
dl |
1.19 |
count += w.drainTasksTo(c); |
1682 |
dl |
1.18 |
} |
1683 |
|
|
return count; |
1684 |
|
|
} |
1685 |
|
|
|
1686 |
|
|
/** |
1687 |
jsr166 |
1.1 |
* 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 |
dl |
1.14 |
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 |
jsr166 |
1.1 |
return super.toString() + |
1704 |
dl |
1.14 |
"[" + runLevelToString(rs) + |
1705 |
|
|
", parallelism = " + pc + |
1706 |
|
|
", size = " + tc + |
1707 |
|
|
", active = " + ac + |
1708 |
|
|
", running = " + rc + |
1709 |
jsr166 |
1.1 |
", steals = " + st + |
1710 |
|
|
", tasks = " + qt + |
1711 |
|
|
", submissions = " + qs + |
1712 |
|
|
"]"; |
1713 |
|
|
} |
1714 |
|
|
|
1715 |
dl |
1.14 |
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 |
jsr166 |
1.1 |
} |
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 |
dl |
1.14 |
advanceRunLevel(SHUTDOWN); |
1737 |
|
|
tryTerminate(false); |
1738 |
jsr166 |
1.1 |
} |
1739 |
|
|
|
1740 |
|
|
/** |
1741 |
jsr166 |
1.9 |
* 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 |
jsr166 |
1.1 |
* |
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 |
dl |
1.14 |
tryTerminate(true); |
1759 |
jsr166 |
1.1 |
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 |
dl |
1.14 |
return runState >= TERMINATED; |
1769 |
jsr166 |
1.1 |
} |
1770 |
|
|
|
1771 |
|
|
/** |
1772 |
|
|
* Returns {@code true} if the process of termination has |
1773 |
jsr166 |
1.9 |
* 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. |
1778 |
jsr166 |
1.1 |
* |
1779 |
jsr166 |
1.9 |
* @return {@code true} if terminating but not yet terminated |
1780 |
jsr166 |
1.1 |
*/ |
1781 |
|
|
public boolean isTerminating() { |
1782 |
dl |
1.14 |
return (runState & (TERMINATING|TERMINATED)) == TERMINATING; |
1783 |
jsr166 |
1.1 |
} |
1784 |
|
|
|
1785 |
|
|
/** |
1786 |
|
|
* Returns {@code true} if this pool has been shut down. |
1787 |
|
|
* |
1788 |
|
|
* @return {@code true} if this pool has been shut down |
1789 |
|
|
*/ |
1790 |
|
|
public boolean isShutdown() { |
1791 |
dl |
1.14 |
return runState >= SHUTDOWN; |
1792 |
jsr166 |
1.9 |
} |
1793 |
|
|
|
1794 |
|
|
/** |
1795 |
jsr166 |
1.1 |
* Blocks until all tasks have completed execution after a shutdown |
1796 |
|
|
* request, or the timeout occurs, or the current thread is |
1797 |
|
|
* interrupted, whichever happens first. |
1798 |
|
|
* |
1799 |
|
|
* @param timeout the maximum time to wait |
1800 |
|
|
* @param unit the time unit of the timeout argument |
1801 |
|
|
* @return {@code true} if this executor terminated and |
1802 |
|
|
* {@code false} if the timeout elapsed before termination |
1803 |
|
|
* @throws InterruptedException if interrupted while waiting |
1804 |
|
|
*/ |
1805 |
|
|
public boolean awaitTermination(long timeout, TimeUnit unit) |
1806 |
|
|
throws InterruptedException { |
1807 |
dl |
1.18 |
try { |
1808 |
|
|
return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0; |
1809 |
|
|
} catch(TimeoutException ex) { |
1810 |
|
|
return false; |
1811 |
|
|
} |
1812 |
jsr166 |
1.1 |
} |
1813 |
|
|
|
1814 |
|
|
/** |
1815 |
|
|
* Interface for extending managed parallelism for tasks running |
1816 |
jsr166 |
1.8 |
* in {@link ForkJoinPool}s. |
1817 |
|
|
* |
1818 |
dl |
1.19 |
* <p>A {@code ManagedBlocker} provides two methods. Method |
1819 |
|
|
* {@code isReleasable} must return {@code true} if blocking is |
1820 |
|
|
* not necessary. Method {@code block} blocks the current thread |
1821 |
|
|
* if necessary (perhaps internally invoking {@code isReleasable} |
1822 |
|
|
* before actually blocking). The unusual methods in this API |
1823 |
|
|
* accommodate synchronizers that may, but don't usually, block |
1824 |
|
|
* for long periods. Similarly, they allow more efficient internal |
1825 |
|
|
* handling of cases in which additional workers may be, but |
1826 |
|
|
* usually are not, needed to ensure sufficient parallelism. |
1827 |
|
|
* Toward this end, implementations of method {@code isReleasable} |
1828 |
|
|
* must be amenable to repeated invocation. |
1829 |
jsr166 |
1.1 |
* |
1830 |
|
|
* <p>For example, here is a ManagedBlocker based on a |
1831 |
|
|
* ReentrantLock: |
1832 |
|
|
* <pre> {@code |
1833 |
|
|
* class ManagedLocker implements ManagedBlocker { |
1834 |
|
|
* final ReentrantLock lock; |
1835 |
|
|
* boolean hasLock = false; |
1836 |
|
|
* ManagedLocker(ReentrantLock lock) { this.lock = lock; } |
1837 |
|
|
* public boolean block() { |
1838 |
|
|
* if (!hasLock) |
1839 |
|
|
* lock.lock(); |
1840 |
|
|
* return true; |
1841 |
|
|
* } |
1842 |
|
|
* public boolean isReleasable() { |
1843 |
|
|
* return hasLock || (hasLock = lock.tryLock()); |
1844 |
|
|
* } |
1845 |
|
|
* }}</pre> |
1846 |
dl |
1.19 |
* |
1847 |
|
|
* <p>Here is a class that possibly blocks waiting for an |
1848 |
|
|
* item on a given queue: |
1849 |
|
|
* <pre> {@code |
1850 |
|
|
* class QueueTaker<E> implements ManagedBlocker { |
1851 |
|
|
* final BlockingQueue<E> queue; |
1852 |
|
|
* volatile E item = null; |
1853 |
|
|
* QueueTaker(BlockingQueue<E> q) { this.queue = q; } |
1854 |
|
|
* public boolean block() throws InterruptedException { |
1855 |
|
|
* if (item == null) |
1856 |
|
|
* item = queue.take |
1857 |
|
|
* return true; |
1858 |
|
|
* } |
1859 |
|
|
* public boolean isReleasable() { |
1860 |
|
|
* return item != null || (item = queue.poll) != null; |
1861 |
|
|
* } |
1862 |
|
|
* public E getItem() { // call after pool.managedBlock completes |
1863 |
|
|
* return item; |
1864 |
|
|
* } |
1865 |
|
|
* }}</pre> |
1866 |
jsr166 |
1.1 |
*/ |
1867 |
|
|
public static interface ManagedBlocker { |
1868 |
|
|
/** |
1869 |
|
|
* Possibly blocks the current thread, for example waiting for |
1870 |
|
|
* a lock or condition. |
1871 |
|
|
* |
1872 |
jsr166 |
1.4 |
* @return {@code true} if no additional blocking is necessary |
1873 |
|
|
* (i.e., if isReleasable would return true) |
1874 |
jsr166 |
1.1 |
* @throws InterruptedException if interrupted while waiting |
1875 |
|
|
* (the method is not required to do so, but is allowed to) |
1876 |
|
|
*/ |
1877 |
|
|
boolean block() throws InterruptedException; |
1878 |
|
|
|
1879 |
|
|
/** |
1880 |
jsr166 |
1.4 |
* Returns {@code true} if blocking is unnecessary. |
1881 |
jsr166 |
1.1 |
*/ |
1882 |
|
|
boolean isReleasable(); |
1883 |
|
|
} |
1884 |
|
|
|
1885 |
|
|
/** |
1886 |
|
|
* Blocks in accord with the given blocker. If the current thread |
1887 |
jsr166 |
1.8 |
* is a {@link ForkJoinWorkerThread}, this method possibly |
1888 |
|
|
* arranges for a spare thread to be activated if necessary to |
1889 |
dl |
1.18 |
* ensure sufficient parallelism while the current thread is blocked. |
1890 |
jsr166 |
1.1 |
* |
1891 |
jsr166 |
1.8 |
* <p>If the caller is not a {@link ForkJoinTask}, this method is |
1892 |
|
|
* behaviorally equivalent to |
1893 |
jsr166 |
1.1 |
* <pre> {@code |
1894 |
|
|
* while (!blocker.isReleasable()) |
1895 |
|
|
* if (blocker.block()) |
1896 |
|
|
* return; |
1897 |
|
|
* }</pre> |
1898 |
jsr166 |
1.8 |
* |
1899 |
|
|
* If the caller is a {@code ForkJoinTask}, then the pool may |
1900 |
|
|
* first be expanded to ensure parallelism, and later adjusted. |
1901 |
jsr166 |
1.1 |
* |
1902 |
|
|
* @param blocker the blocker |
1903 |
|
|
* @throws InterruptedException if blocker.block did so |
1904 |
|
|
*/ |
1905 |
dl |
1.18 |
public static void managedBlock(ManagedBlocker blocker) |
1906 |
jsr166 |
1.1 |
throws InterruptedException { |
1907 |
|
|
Thread t = Thread.currentThread(); |
1908 |
dl |
1.19 |
if (t instanceof ForkJoinWorkerThread) { |
1909 |
|
|
ForkJoinWorkerThread w = (ForkJoinWorkerThread) t; |
1910 |
|
|
w.pool.awaitBlocker(blocker); |
1911 |
|
|
} |
1912 |
dl |
1.18 |
else { |
1913 |
|
|
do {} while (!blocker.isReleasable() && !blocker.block()); |
1914 |
|
|
} |
1915 |
jsr166 |
1.1 |
} |
1916 |
|
|
|
1917 |
jsr166 |
1.7 |
// AbstractExecutorService overrides. These rely on undocumented |
1918 |
|
|
// fact that ForkJoinTask.adapt returns ForkJoinTasks that also |
1919 |
|
|
// implement RunnableFuture. |
1920 |
jsr166 |
1.1 |
|
1921 |
|
|
protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) { |
1922 |
jsr166 |
1.7 |
return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value); |
1923 |
jsr166 |
1.1 |
} |
1924 |
|
|
|
1925 |
|
|
protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) { |
1926 |
jsr166 |
1.7 |
return (RunnableFuture<T>) ForkJoinTask.adapt(callable); |
1927 |
jsr166 |
1.1 |
} |
1928 |
|
|
|
1929 |
|
|
// Unsafe mechanics |
1930 |
|
|
|
1931 |
|
|
private static final sun.misc.Unsafe UNSAFE = sun.misc.Unsafe.getUnsafe(); |
1932 |
dl |
1.14 |
private static final long workerCountsOffset = |
1933 |
|
|
objectFieldOffset("workerCounts", ForkJoinPool.class); |
1934 |
|
|
private static final long runStateOffset = |
1935 |
|
|
objectFieldOffset("runState", ForkJoinPool.class); |
1936 |
jsr166 |
1.2 |
private static final long eventCountOffset = |
1937 |
jsr166 |
1.3 |
objectFieldOffset("eventCount", ForkJoinPool.class); |
1938 |
dl |
1.14 |
private static final long eventWaitersOffset = |
1939 |
|
|
objectFieldOffset("eventWaiters",ForkJoinPool.class); |
1940 |
|
|
private static final long stealCountOffset = |
1941 |
|
|
objectFieldOffset("stealCount",ForkJoinPool.class); |
1942 |
dl |
1.19 |
private static final long spareWaitersOffset = |
1943 |
|
|
objectFieldOffset("spareWaiters",ForkJoinPool.class); |
1944 |
jsr166 |
1.3 |
|
1945 |
|
|
private static long objectFieldOffset(String field, Class<?> klazz) { |
1946 |
|
|
try { |
1947 |
|
|
return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field)); |
1948 |
|
|
} catch (NoSuchFieldException e) { |
1949 |
|
|
// Convert Exception to corresponding Error |
1950 |
|
|
NoSuchFieldError error = new NoSuchFieldError(field); |
1951 |
|
|
error.initCause(e); |
1952 |
|
|
throw error; |
1953 |
|
|
} |
1954 |
|
|
} |
1955 |
jsr166 |
1.1 |
} |