--- jsr166/src/jsr166y/ForkJoinPool.java 2009/07/20 21:45:06 1.7 +++ jsr166/src/jsr166y/ForkJoinPool.java 2012/01/31 01:32:25 1.120 @@ -1,546 +1,2007 @@ /* * Written by Doug Lea with assistance from members of JCP JSR-166 * Expert Group and released to the public domain, as explained at - * http://creativecommons.org/licenses/publicdomain + * http://creativecommons.org/publicdomain/zero/1.0/ */ package jsr166y; -import java.util.*; -import java.util.concurrent.*; -import java.util.concurrent.locks.*; -import java.util.concurrent.atomic.*; -import sun.misc.Unsafe; -import java.lang.reflect.*; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collection; +import java.util.Collections; +import java.util.List; +import java.util.Random; +import java.util.concurrent.AbstractExecutorService; +import java.util.concurrent.Callable; +import java.util.concurrent.ExecutorService; +import java.util.concurrent.Future; +import java.util.concurrent.RejectedExecutionException; +import java.util.concurrent.RunnableFuture; +import java.util.concurrent.TimeUnit; +import java.util.concurrent.atomic.AtomicInteger; +import java.util.concurrent.atomic.AtomicLong; +import java.util.concurrent.locks.AbstractQueuedSynchronizer; +import java.util.concurrent.locks.Condition; /** - * An {@link ExecutorService} for running {@link ForkJoinTask}s. A - * ForkJoinPool provides the entry point for submissions from - * non-ForkJoinTasks, as well as management and monitoring operations. - * Normally a single ForkJoinPool is used for a large number of - * submitted tasks. Otherwise, use would not usually outweigh the - * construction and bookkeeping overhead of creating a large set of - * threads. + * An {@link ExecutorService} for running {@link ForkJoinTask}s. + * A {@code ForkJoinPool} provides the entry point for submissions + * from non-{@code ForkJoinTask} clients, as well as management and + * monitoring operations. * - *
ForkJoinPools differ from other kinds of Executors mainly in - * that they provide work-stealing: all threads in the pool - * attempt to find and execute subtasks created by other active tasks - * (eventually blocking if none exist). This makes them efficient when - * most tasks spawn other subtasks (as do most ForkJoinTasks), as well - * as the mixed execution of some plain Runnable- or Callable- based - * activities along with ForkJoinTasks. When setting - * setAsyncMode, a ForkJoinPools may also be appropriate for - * use with fine-grained tasks that are never joined. Otherwise, other - * ExecutorService implementations are typically more appropriate - * choices. + *
A {@code ForkJoinPool} differs from other kinds of {@link + * ExecutorService} mainly by virtue of employing + * work-stealing: all threads in the pool attempt to find and + * execute tasks submitted to the pool and/or created by other active + * tasks (eventually blocking waiting for work if none exist). This + * enables efficient processing when most tasks spawn other subtasks + * (as do most {@code ForkJoinTask}s), as well as when many small + * tasks are submitted to the pool from external clients. Especially + * when setting asyncMode to true in constructors, {@code + * ForkJoinPool}s may also be appropriate for use with event-style + * tasks that are never joined. * - *
A ForkJoinPool may be constructed with a given parallelism level
- * (target pool size), which it attempts to maintain by dynamically
- * adding, suspending, or resuming threads, even if some tasks are
- * waiting to join others. However, no such adjustments are performed
- * in the face of blocked IO or other unmanaged synchronization. The
- * nested ManagedBlocker
interface enables extension of
- * the kinds of synchronization accommodated. The target parallelism
- * level may also be changed dynamically (setParallelism
)
- * and thread construction can be limited using methods
- * setMaximumPoolSize
and/or
- * setMaintainsParallelism
.
+ *
A {@code ForkJoinPool} is constructed with a given target + * parallelism level; by default, equal to the number of available + * processors. The pool attempts to maintain enough active (or + * available) threads by dynamically adding, suspending, or resuming + * internal worker threads, even if some tasks are stalled waiting to + * join others. However, no such adjustments are guaranteed in the + * face of blocked IO or other unmanaged synchronization. The nested + * {@link ManagedBlocker} interface enables extension of the kinds of + * synchronization accommodated. * *
In addition to execution and lifecycle control methods, this
* class provides status check methods (for example
- * getStealCount
) that are intended to aid in developing,
+ * {@link #getStealCount}) that are intended to aid in developing,
* tuning, and monitoring fork/join applications. Also, method
- * toString
returns indications of pool state in a
+ * {@link #toString} returns indications of pool state in a
* convenient form for informal monitoring.
*
+ *
As is the case with other ExecutorServices, there are three + * main task execution methods summarized in the following table. + * These are designed to be used primarily by clients not already + * engaged in fork/join computations in the current pool. The main + * forms of these methods accept instances of {@code ForkJoinTask}, + * but overloaded forms also allow mixed execution of plain {@code + * Runnable}- or {@code Callable}- based activities as well. However, + * tasks that are already executing in a pool should normally instead + * use the within-computation forms listed in the table unless using + * async event-style tasks that are not usually joined, in which case + * there is little difference among choice of methods. + * + *
+ * | Call from non-fork/join clients | + *Call from within fork/join computations | + *
Arrange async execution | + *{@link #execute(ForkJoinTask)} | + *{@link ForkJoinTask#fork} | + *
Await and obtain result | + *{@link #invoke(ForkJoinTask)} | + *{@link ForkJoinTask#invoke} | + *
Arrange exec and obtain Future | + *{@link #submit(ForkJoinTask)} | + *{@link ForkJoinTask#fork} (ForkJoinTasks are Futures) | + *
Sample Usage. Normally a single {@code ForkJoinPool} is + * used for all parallel task execution in a program or subsystem. + * Otherwise, use would not usually outweigh the construction and + * bookkeeping overhead of creating a large set of threads. For + * example, a common pool could be used for the {@code SortTasks} + * illustrated in {@link RecursiveAction}. Because {@code + * ForkJoinPool} uses threads in {@linkplain java.lang.Thread#isDaemon + * daemon} mode, there is typically no need to explicitly {@link + * #shutdown} such a pool upon program exit. + * + *
{@code + * static final ForkJoinPool mainPool = new ForkJoinPool(); + * ... + * public void sort(long[] array) { + * mainPool.invoke(new SortTask(array, 0, array.length)); + * }}+ * *
Implementation notes: This implementation restricts the * maximum number of running threads to 32767. Attempts to create - * pools with greater than the maximum result in - * IllegalArgumentExceptions. + * pools with greater than the maximum number result in + * {@code IllegalArgumentException}. + * + *
This implementation rejects submitted tasks (that is, by throwing
+ * {@link RejectedExecutionException}) only when the pool is shut down
+ * or internal resources have been exhausted.
+ *
+ * @since 1.7
+ * @author Doug Lea
*/
public class ForkJoinPool extends AbstractExecutorService {
/*
- * See the extended comments interspersed below for design,
- * rationale, and walkthroughs.
- */
+ * Implementation Overview
+ *
+ * This class and its nested classes provide the main
+ * functionality and control for a set of worker threads:
+ * Submissions from non-FJ threads enter into submission queues.
+ * Workers take these tasks and typically split them into subtasks
+ * that may be stolen by other workers. Preference rules give
+ * first priority to processing tasks from their own queues (LIFO
+ * or FIFO, depending on mode), then to randomized FIFO steals of
+ * tasks in other queues.
+ *
+ * WorkQueues
+ * ==========
+ *
+ * Most operations occur within work-stealing queues (in nested
+ * class WorkQueue). These are special forms of Deques that
+ * support only three of the four possible end-operations -- push,
+ * pop, and poll (aka steal), under the further constraints that
+ * push and pop are called only from the owning thread (or, as
+ * extended here, under a lock), while poll may be called from
+ * other threads. (If you are unfamiliar with them, you probably
+ * want to read Herlihy and Shavit's book "The Art of
+ * Multiprocessor programming", chapter 16 describing these in
+ * more detail before proceeding.) The main work-stealing queue
+ * design is roughly similar to those in the papers "Dynamic
+ * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
+ * (http://research.sun.com/scalable/pubs/index.html) and
+ * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
+ * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
+ * The main differences ultimately stem from GC requirements that
+ * we null out taken slots as soon as we can, to maintain as small
+ * a footprint as possible even in programs generating huge
+ * numbers of tasks. To accomplish this, we shift the CAS
+ * arbitrating pop vs poll (steal) from being on the indices
+ * ("base" and "top") to the slots themselves. So, both a
+ * successful pop and poll mainly entail a CAS of a slot from
+ * non-null to null. Because we rely on CASes of references, we
+ * do not need tag bits on base or top. They are simple ints as
+ * used in any circular array-based queue (see for example
+ * ArrayDeque). Updates to the indices must still be ordered in a
+ * way that guarantees that top == base means the queue is empty,
+ * but otherwise may err on the side of possibly making the queue
+ * appear nonempty when a push, pop, or poll have not fully
+ * committed. Note that this means that the poll operation,
+ * considered individually, is not wait-free. One thief cannot
+ * successfully continue until another in-progress one (or, if
+ * previously empty, a push) completes. However, in the
+ * aggregate, we ensure at least probabilistic non-blockingness.
+ * If an attempted steal fails, a thief always chooses a different
+ * random victim target to try next. So, in order for one thief to
+ * progress, it suffices for any in-progress poll or new push on
+ * any empty queue to complete.
+ *
+ * This approach also enables support of a user mode in which local
+ * task processing is in FIFO, not LIFO order, simply by using
+ * poll rather than pop. This can be useful in message-passing
+ * frameworks in which tasks are never joined. However neither
+ * mode considers affinities, loads, cache localities, etc, so
+ * rarely provide the best possible performance on a given
+ * machine, but portably provide good throughput by averaging over
+ * these factors. (Further, even if we did try to use such
+ * information, we do not usually have a basis for exploiting it.
+ * For example, some sets of tasks profit from cache affinities,
+ * but others are harmed by cache pollution effects.)
+ *
+ * WorkQueues are also used in a similar way for tasks submitted
+ * to the pool. We cannot mix these tasks in the same queues used
+ * for work-stealing (this would contaminate lifo/fifo
+ * processing). Instead, we loosely associate submission queues
+ * with submitting threads, using a form of hashing. The
+ * ThreadLocal Submitter class contains a value initially used as
+ * a hash code for choosing existing queues, but may be randomly
+ * repositioned upon contention with other submitters. In
+ * essence, submitters act like workers except that they never
+ * take tasks, and they are multiplexed on to a finite number of
+ * shared work queues. However, classes are set up so that future
+ * extensions could allow submitters to optionally help perform
+ * tasks as well. Insertion of tasks in shared mode requires a
+ * lock (mainly to protect in the case of resizing) but we use
+ * only a simple spinlock (using bits in field runState), because
+ * submitters encountering a busy queue move on to try or create
+ * other queues, so never block.
+ *
+ * Management
+ * ==========
+ *
+ * The main throughput advantages of work-stealing stem from
+ * decentralized control -- workers mostly take tasks from
+ * themselves or each other. We cannot negate this in the
+ * implementation of other management responsibilities. The main
+ * tactic for avoiding bottlenecks is packing nearly all
+ * essentially atomic control state into two volatile variables
+ * that are by far most often read (not written) as status and
+ * consistency checks.
+ *
+ * Field "ctl" contains 64 bits holding all the information needed
+ * to atomically decide to add, inactivate, enqueue (on an event
+ * queue), dequeue, and/or re-activate workers. To enable this
+ * packing, we restrict maximum parallelism to (1<<15)-1 (which is
+ * far in excess of normal operating range) to allow ids, counts,
+ * and their negations (used for thresholding) to fit into 16bit
+ * fields.
+ *
+ * Field "runState" contains 32 bits needed to register and
+ * deregister WorkQueues, as well as to enable shutdown. It is
+ * only modified under a lock (normally briefly held, but
+ * occasionally protecting allocations and resizings) but even
+ * when locked remains available to check consistency. An
+ * auxiliary field "growHints", also only modified under lock,
+ * contains a candidate index for the next WorkQueue and
+ * a mask for submission queue indices.
+ *
+ * Recording WorkQueues. WorkQueues are recorded in the
+ * "workQueues" array that is created upon pool construction and
+ * expanded if necessary. Updates to the array while recording
+ * new workers and unrecording terminated ones are protected from
+ * each other by a lock but the array is otherwise concurrently
+ * readable, and accessed directly. To simplify index-based
+ * operations, the array size is always a power of two, and all
+ * readers must tolerate null slots. Shared (submission) queues
+ * are at even indices, worker queues at odd indices. Grouping
+ * them together in this way simplifies and speeds up task
+ * scanning. To avoid flailing during start-up, the array is
+ * presized to hold twice #parallelism workers (which is unlikely
+ * to need further resizing during execution). But to avoid
+ * dealing with so many null slots, variable runState includes a
+ * mask for the nearest power of two that contains all currently
+ * used indices.
+ *
+ * All worker thread creation is on-demand, triggered by task
+ * submissions, replacement of terminated workers, and/or
+ * compensation for blocked workers. However, all other support
+ * code is set up to work with other policies. To ensure that we
+ * do not hold on to worker references that would prevent GC, ALL
+ * accesses to workQueues are via indices into the workQueues
+ * array (which is one source of some of the messy code
+ * constructions here). In essence, the workQueues array serves as
+ * a weak reference mechanism. Thus for example the wait queue
+ * field of ctl stores indices, not references. Access to the
+ * workQueues in associated methods (for example signalWork) must
+ * both index-check and null-check the IDs. All such accesses
+ * ignore bad IDs by returning out early from what they are doing,
+ * since this can only be associated with termination, in which
+ * case it is OK to give up. All uses of the workQueues array
+ * also check that it is non-null (even if previously
+ * non-null). This allows nulling during termination, which is
+ * currently not necessary, but remains an option for
+ * resource-revocation-based shutdown schemes. It also helps
+ * reduce JIT issuance of uncommon-trap code, which tends to
+ * unnecessarily complicate control flow in some methods.
+ *
+ * Event Queuing. Unlike HPC work-stealing frameworks, we cannot
+ * let workers spin indefinitely scanning for tasks when none can
+ * be found immediately, and we cannot start/resume workers unless
+ * there appear to be tasks available. On the other hand, we must
+ * quickly prod them into action when new tasks are submitted or
+ * generated. In many usages, ramp-up time to activate workers is
+ * the main limiting factor in overall performance (this is
+ * compounded at program start-up by JIT compilation and
+ * allocation). So we try to streamline this as much as possible.
+ * We park/unpark workers after placing in an event wait queue
+ * when they cannot find work. This "queue" is actually a simple
+ * Treiber stack, headed by the "id" field of ctl, plus a 15bit
+ * counter value (that reflects the number of times a worker has
+ * been inactivated) to avoid ABA effects (we need only as many
+ * version numbers as worker threads). Successors are held in
+ * field WorkQueue.nextWait. Queuing deals with several intrinsic
+ * races, mainly that a task-producing thread can miss seeing (and
+ * signalling) another thread that gave up looking for work but
+ * has not yet entered the wait queue. We solve this by requiring
+ * a full sweep of all workers (via repeated calls to method
+ * scan()) both before and after a newly waiting worker is added
+ * to the wait queue. During a rescan, the worker might release
+ * some other queued worker rather than itself, which has the same
+ * net effect. Because enqueued workers may actually be rescanning
+ * rather than waiting, we set and clear the "parker" field of
+ * WorkQueues to reduce unnecessary calls to unpark. (This
+ * requires a secondary recheck to avoid missed signals.) Note
+ * the unusual conventions about Thread.interrupts surrounding
+ * parking and other blocking: Because interrupts are used solely
+ * to alert threads to check termination, which is checked anyway
+ * upon blocking, we clear status (using Thread.interrupted)
+ * before any call to park, so that park does not immediately
+ * return due to status being set via some other unrelated call to
+ * interrupt in user code.
+ *
+ * Signalling. We create or wake up workers only when there
+ * appears to be at least one task they might be able to find and
+ * execute. When a submission is added or another worker adds a
+ * task to a queue that previously had fewer than two tasks, they
+ * signal waiting workers (or trigger creation of new ones if
+ * fewer than the given parallelism level -- see signalWork).
+ * These primary signals are buttressed by signals during rescans;
+ * together these cover the signals needed in cases when more
+ * tasks are pushed but untaken, and improve performance compared
+ * to having one thread wake up all workers.
+ *
+ * Trimming workers. To release resources after periods of lack of
+ * use, a worker starting to wait when the pool is quiescent will
+ * time out and terminate if the pool has remained quiescent for
+ * SHRINK_RATE nanosecs. This will slowly propagate, eventually
+ * terminating all workers after long periods of non-use.
+ *
+ * Shutdown and Termination. A call to shutdownNow atomically sets
+ * a runState bit and then (non-atomically) sets each worker's
+ * runState status, cancels all unprocessed tasks, and wakes up
+ * all waiting workers. Detecting whether termination should
+ * commence after a non-abrupt shutdown() call requires more work
+ * and bookkeeping. We need consensus about quiescence (i.e., that
+ * there is no more work). The active count provides a primary
+ * indication but non-abrupt shutdown still requires a rechecking
+ * scan for any workers that are inactive but not queued.
+ *
+ * Joining Tasks
+ * =============
+ *
+ * Any of several actions may be taken when one worker is waiting
+ * to join a task stolen (or always held) by another. Because we
+ * are multiplexing many tasks on to a pool of workers, we can't
+ * just let them block (as in Thread.join). We also cannot just
+ * reassign the joiner's run-time stack with another and replace
+ * it later, which would be a form of "continuation", that even if
+ * possible is not necessarily a good idea since we sometimes need
+ * both an unblocked task and its continuation to progress.
+ * Instead we combine two tactics:
+ *
+ * Helping: Arranging for the joiner to execute some task that it
+ * would be running if the steal had not occurred.
+ *
+ * Compensating: Unless there are already enough live threads,
+ * method tryCompensate() may create or re-activate a spare
+ * thread to compensate for blocked joiners until they unblock.
+ *
+ * A third form (implemented in tryRemoveAndExec and
+ * tryPollForAndExec) amounts to helping a hypothetical
+ * compensator: If we can readily tell that a possible action of a
+ * compensator is to steal and execute the task being joined, the
+ * joining thread can do so directly, without the need for a
+ * compensation thread (although at the expense of larger run-time
+ * stacks, but the tradeoff is typically worthwhile).
+ *
+ * The ManagedBlocker extension API can't use helping so relies
+ * only on compensation in method awaitBlocker.
+ *
+ * The algorithm in tryHelpStealer entails a form of "linear"
+ * helping: Each worker records (in field currentSteal) the most
+ * recent task it stole from some other worker. Plus, it records
+ * (in field currentJoin) the task it is currently actively
+ * joining. Method tryHelpStealer uses these markers to try to
+ * find a worker to help (i.e., steal back a task from and execute
+ * it) that could hasten completion of the actively joined task.
+ * In essence, the joiner executes a task that would be on its own
+ * local deque had the to-be-joined task not been stolen. This may
+ * be seen as a conservative variant of the approach in Wagner &
+ * Calder "Leapfrogging: a portable technique for implementing
+ * efficient futures" SIGPLAN Notices, 1993
+ * (http://portal.acm.org/citation.cfm?id=155354). It differs in
+ * that: (1) We only maintain dependency links across workers upon
+ * steals, rather than use per-task bookkeeping. This sometimes
+ * requires a linear scan of workQueues array to locate stealers, but
+ * often doesn't because stealers leave hints (that may become
+ * stale/wrong) of where to locate them. A stealHint is only a
+ * hint because a worker might have had multiple steals and the
+ * hint records only one of them (usually the most current).
+ * Hinting isolates cost to when it is needed, rather than adding
+ * to per-task overhead. (2) It is "shallow", ignoring nesting
+ * and potentially cyclic mutual steals. (3) It is intentionally
+ * racy: field currentJoin is updated only while actively joining,
+ * which means that we miss links in the chain during long-lived
+ * tasks, GC stalls etc (which is OK since blocking in such cases
+ * is usually a good idea). (4) We bound the number of attempts
+ * to find work (see MAX_HELP_DEPTH) and fall back to suspending
+ * the worker and if necessary replacing it with another.
+ *
+ * It is impossible to keep exactly the target parallelism number
+ * of threads running at any given time. Determining the
+ * existence of conservatively safe helping targets, the
+ * availability of already-created spares, and the apparent need
+ * to create new spares are all racy, so we rely on multiple
+ * retries of each. Currently, in keeping with on-demand
+ * signalling policy, we compensate only if blocking would leave
+ * less than one active (non-waiting, non-blocked) worker.
+ * Additionally, to avoid some false alarms due to GC, lagging
+ * counters, system activity, etc, compensated blocking for joins
+ * is only attempted after rechecks stabilize in
+ * ForkJoinTask.awaitJoin. (Retries are interspersed with
+ * Thread.yield, for good citizenship.)
+ *
+ * Style notes: There is a lot of representation-level coupling
+ * among classes ForkJoinPool, ForkJoinWorkerThread, and
+ * ForkJoinTask. The fields of WorkQueue maintain data structures
+ * managed by ForkJoinPool, so are directly accessed. There is
+ * little point trying to reduce this, since any associated future
+ * changes in representations will need to be accompanied by
+ * algorithmic changes anyway. Several methods intrinsically
+ * sprawl because they must accumulate sets of consistent reads of
+ * volatiles held in local variables. Methods signalWork() and
+ * scan() are the main bottlenecks, so are especially heavily
+ * micro-optimized/mangled. There are lots of inline assignments
+ * (of form "while ((local = field) != 0)") which are usually the
+ * simplest way to ensure the required read orderings (which are
+ * sometimes critical). This leads to a "C"-like style of listing
+ * declarations of these locals at the heads of methods or blocks.
+ * There are several occurrences of the unusual "do {} while
+ * (!cas...)" which is the simplest way to force an update of a
+ * CAS'ed variable. There are also other coding oddities that help
+ * some methods perform reasonably even when interpreted (not
+ * compiled).
+ *
+ * The order of declarations in this file is:
+ * (1) Static utility functions
+ * (2) Nested (static) classes
+ * (3) Static fields
+ * (4) Fields, along with constants used when unpacking some of them
+ * (5) Internal control methods
+ * (6) Callbacks and other support for ForkJoinTask methods
+ * (7) Exported methods
+ * (8) Static block initializing statics in minimally dependent order
+ */
+
+ // Static utilities
+
+ /**
+ * Computes an initial hash code (also serving as a non-zero
+ * random seed) for a thread id. This method is expected to
+ * provide higher-quality hash codes than using method hashCode().
+ */
+ static final int hashId(long id) {
+ int h = (int)id ^ (int)(id >>> 32); // Use MurmurHash of thread id
+ h ^= h >>> 16; h *= 0x85ebca6b;
+ h ^= h >>> 13; h *= 0xc2b2ae35;
+ h ^= h >>> 16;
+ return (h == 0)? 1 : h; // ensure nonzero
+ }
- /** Mask for packing and unpacking shorts */
- private static final int shortMask = 0xffff;
+ /**
+ * If there is a security manager, makes sure caller has
+ * permission to modify threads.
+ */
+ private static void checkPermission() {
+ SecurityManager security = System.getSecurityManager();
+ if (security != null)
+ security.checkPermission(modifyThreadPermission);
+ }
- /** Max pool size -- must be a power of two minus 1 */
- private static final int MAX_THREADS = 0x7FFF;
+ // Nested classes
/**
- * Factory for creating new ForkJoinWorkerThreads. A
- * ForkJoinWorkerThreadFactory must be defined and used for
- * ForkJoinWorkerThread subclasses that extend base functionality
- * or initialize threads with different contexts.
+ * Factory for creating new {@link ForkJoinWorkerThread}s.
+ * A {@code ForkJoinWorkerThreadFactory} must be defined and used
+ * for {@code ForkJoinWorkerThread} subclasses that extend base
+ * functionality or initialize threads with different contexts.
*/
public static interface ForkJoinWorkerThreadFactory {
/**
* Returns a new worker thread operating in the given pool.
*
* @param pool the pool this thread works in
- * @throws NullPointerException if pool is null;
+ * @throws NullPointerException if the pool is null
*/
public ForkJoinWorkerThread newThread(ForkJoinPool pool);
}
/**
- * Default ForkJoinWorkerThreadFactory implementation, creates a
+ * Default ForkJoinWorkerThreadFactory implementation; creates a
* new ForkJoinWorkerThread.
*/
- static class DefaultForkJoinWorkerThreadFactory
+ static class DefaultForkJoinWorkerThreadFactory
implements ForkJoinWorkerThreadFactory {
public ForkJoinWorkerThread newThread(ForkJoinPool pool) {
- try {
- return new ForkJoinWorkerThread(pool);
- } catch (OutOfMemoryError oom) {
+ return new ForkJoinWorkerThread(pool);
+ }
+ }
+
+ /**
+ * A simple non-reentrant lock used for exclusion when managing
+ * queues and workers. We use a custom lock so that we can readily
+ * probe lock state in constructions that check among alternative
+ * actions. The lock is normally only very briefly held, and
+ * sometimes treated as a spinlock, but other usages block to
+ * reduce overall contention in those cases where locked code
+ * bodies perform allocation/resizing.
+ */
+ static final class Mutex extends AbstractQueuedSynchronizer {
+ public final boolean tryAcquire(int ignore) {
+ return compareAndSetState(0, 1);
+ }
+ public final boolean tryRelease(int ignore) {
+ setState(0);
+ return true;
+ }
+ public final void lock() { acquire(0); }
+ public final void unlock() { release(0); }
+ public final boolean isHeldExclusively() { return getState() == 1; }
+ public final Condition newCondition() { return new ConditionObject(); }
+ }
+
+ /**
+ * Class for artificial tasks that are used to replace the target
+ * of local joins if they are removed from an interior queue slot
+ * in WorkQueue.tryRemoveAndExec. We don't need the proxy to
+ * actually do anything beyond having a unique identity.
+ */
+ static final class EmptyTask extends ForkJoinTask A {@code ManagedBlocker} provides two methods. Method
+ * {@code isReleasable} must return {@code true} if blocking is
+ * not necessary. Method {@code block} blocks the current thread
+ * if necessary (perhaps internally invoking {@code isReleasable}
+ * before actually blocking). These actions are performed by any
+ * thread invoking {@link ForkJoinPool#managedBlock}. The
+ * unusual methods in this API accommodate synchronizers that may,
+ * but don't usually, block for long periods. Similarly, they
+ * allow more efficient internal handling of cases in which
+ * additional workers may be, but usually are not, needed to
+ * ensure sufficient parallelism. Toward this end,
+ * implementations of method {@code isReleasable} must be amenable
+ * to repeated invocation.
*
- * Conceptually, all we need to do here is add or resume a
- * spare thread when one is about to block (and remove or
- * suspend it later when unblocked -- see suspendIfSpare).
- * However, implementing this idea requires coping with
- * several problems: We have imperfect information about the
- * states of threads. Some count updates can and usually do
- * lag run state changes, despite arrangements to keep them
- * accurate (for example, when possible, updating counts
- * before signalling or resuming), especially when running on
- * dynamic JVMs that don't optimize the infrequent paths that
- * update counts. Generating too many threads can make these
- * problems become worse, because excess threads are more
- * likely to be context-switched with others, slowing them all
- * down, especially if there is no work available, so all are
- * busy scanning or idling. Also, excess spare threads can
- * only be suspended or removed when they are idle, not
- * immediately when they aren't needed. So adding threads will
- * raise parallelism level for longer than necessary. Also,
- * FJ applications often enounter highly transient peaks when
- * many threads are blocked joining, but for less time than it
- * takes to create or resume spares.
- *
- * @param joinMe if non-null, return early if done
- * @param maintainParallelism if true, try to stay within
- * target counts, else create only to avoid starvation
- * @return true if joinMe known to be done
- */
- final boolean preJoin(ForkJoinTask> joinMe, boolean maintainParallelism) {
- maintainParallelism &= maintainsParallelism; // overrride
- boolean dec = false; // true when running count decremented
- while (spareStack == null || !tryResumeSpare(dec)) {
- int counts = workerCounts;
- if (dec || (dec = casWorkerCounts(counts, --counts))) { // CAS cheat
- if (!needSpare(counts, maintainParallelism))
- break;
- if (joinMe.status < 0)
- return true;
- if (tryAddSpare(counts))
- break;
- }
- }
- return false;
- }
-
- /**
- * Same idea as preJoin
- */
- final boolean preBlock(ManagedBlocker blocker, boolean maintainParallelism){
- maintainParallelism &= maintainsParallelism;
- boolean dec = false;
- while (spareStack == null || !tryResumeSpare(dec)) {
- int counts = workerCounts;
- if (dec || (dec = casWorkerCounts(counts, --counts))) {
- if (!needSpare(counts, maintainParallelism))
- break;
- if (blocker.isReleasable())
- return true;
- if (tryAddSpare(counts))
- break;
- }
- }
- return false;
- }
-
- /**
- * Returns true if a spare thread appears to be needed. If
- * maintaining parallelism, returns true when the deficit in
- * running threads is more than the surplus of total threads, and
- * there is apparently some work to do. This self-limiting rule
- * means that the more threads that have already been added, the
- * less parallelism we will tolerate before adding another.
- * @param counts current worker counts
- * @param maintainParallelism try to maintain parallelism
- */
- private boolean needSpare(int counts, boolean maintainParallelism) {
- int ps = parallelism;
- int rc = runningCountOf(counts);
- int tc = totalCountOf(counts);
- int runningDeficit = ps - rc;
- int totalSurplus = tc - ps;
- return (tc < maxPoolSize &&
- (rc == 0 || totalSurplus < 0 ||
- (maintainParallelism &&
- runningDeficit > totalSurplus &&
- ForkJoinWorkerThread.hasQueuedTasks(workers))));
- }
-
- /**
- * Add a spare worker if lock available and no more than the
- * expected numbers of threads exist
- * @return true if successful
- */
- private boolean tryAddSpare(int expectedCounts) {
- final ReentrantLock lock = this.workerLock;
- int expectedRunning = runningCountOf(expectedCounts);
- int expectedTotal = totalCountOf(expectedCounts);
- boolean success = false;
- boolean locked = false;
- // confirm counts while locking; CAS after obtaining lock
- try {
- for (;;) {
- int s = workerCounts;
- int tc = totalCountOf(s);
- int rc = runningCountOf(s);
- if (rc > expectedRunning || tc > expectedTotal)
- break;
- if (!locked && !(locked = lock.tryLock()))
- break;
- if (casWorkerCounts(s, workerCountsFor(tc+1, rc+1))) {
- createAndStartSpare(tc);
- success = true;
- break;
- }
- }
- } finally {
- if (locked)
- lock.unlock();
- }
- return success;
- }
-
- /**
- * Add the kth spare worker. On entry, pool coounts are already
- * adjusted to reflect addition.
- */
- private void createAndStartSpare(int k) {
- ForkJoinWorkerThread w = null;
- ForkJoinWorkerThread[] ws = ensureWorkerArrayCapacity(k + 1);
- int len = ws.length;
- // Probably, we can place at slot k. If not, find empty slot
- if (k < len && ws[k] != null) {
- for (k = 0; k < len && ws[k] != null; ++k)
- ;
- }
- if (k < len && !isTerminating() && (w = createWorker(k)) != null) {
- ws[k] = w;
- w.start();
- }
- else
- updateWorkerCount(-1); // adjust on failure
- signalIdleWorkers();
- }
-
- /**
- * Suspend calling thread w if there are excess threads. Called
- * only from sync. Spares are enqueued in a Treiber stack
- * using the same WaitQueueNodes as barriers. They are resumed
- * mainly in preJoin, but are also woken on pool events that
- * require all threads to check run state.
- * @param w the caller
- */
- private boolean suspendIfSpare(ForkJoinWorkerThread w) {
- WaitQueueNode node = null;
- int s;
- while (parallelism < runningCountOf(s = workerCounts)) {
- if (node == null)
- node = new WaitQueueNode(0, w);
- if (casWorkerCounts(s, s-1)) { // representation-dependent
- // push onto stack
- do;while (!casSpareStack(node.next = spareStack, node));
- // block until released by resumeSpare
- node.awaitSpareRelease();
- return true;
- }
- }
- return false;
- }
-
- /**
- * Try to pop and resume a spare thread.
- * @param updateCount if true, increment running count on success
- * @return true if successful
- */
- private boolean tryResumeSpare(boolean updateCount) {
- WaitQueueNode q;
- while ((q = spareStack) != null) {
- if (casSpareStack(q, q.next)) {
- if (updateCount)
- updateRunningCount(1);
- q.signal();
- return true;
- }
- }
- return false;
- }
-
- /**
- * Pop and resume all spare threads. Same idea as ensureSync.
- * @return true if any spares released
- */
- private boolean resumeAllSpares() {
- WaitQueueNode q;
- while ( (q = spareStack) != null) {
- if (casSpareStack(q, null)) {
- do {
- updateRunningCount(1);
- q.signal();
- } while ((q = q.next) != null);
- return true;
- }
- }
- return false;
- }
-
- /**
- * Pop and shutdown excessive spare threads. Call only while
- * holding lock. This is not guaranteed to eliminate all excess
- * threads, only those suspended as spares, which are the ones
- * unlikely to be needed in the future.
- */
- private void trimSpares() {
- int surplus = totalCountOf(workerCounts) - parallelism;
- WaitQueueNode q;
- while (surplus > 0 && (q = spareStack) != null) {
- if (casSpareStack(q, null)) {
- do {
- updateRunningCount(1);
- ForkJoinWorkerThread w = q.thread;
- if (w != null && surplus > 0 &&
- runningCountOf(workerCounts) > 0 && w.shutdown())
- --surplus;
- q.signal();
- } while ((q = q.next) != null);
- }
- }
- }
-
- /**
- * Interface for extending managed parallelism for tasks running
- * in ForkJoinPools. A ManagedBlocker provides two methods.
- * Method For example, here is a ManagedBlocker based on a
* ReentrantLock:
- * Here is a class that possibly blocks waiting for an
+ * item on a given queue:
+ * If the caller is not a ForkJoinTask, this method is behaviorally
- * equivalent to
- * If the caller is not a {@link ForkJoinTask}, this method is
+ * behaviorally equivalent to
+ * ("modifyThread")
,
+ * java.lang.RuntimePermission}{@code ("modifyThread")}
*/
public ForkJoinPool() {
this(Runtime.getRuntime().availableProcessors(),
- defaultForkJoinWorkerThreadFactory);
+ defaultForkJoinWorkerThreadFactory, null, false);
}
/**
- * Creates a ForkJoinPool with the indicated parellelism level
- * threads, and using the default ForkJoinWorkerThreadFactory,
- * @param parallelism the number of worker threads
+ * Creates a {@code ForkJoinPool} with the indicated parallelism
+ * level, the {@linkplain
+ * #defaultForkJoinWorkerThreadFactory default thread factory},
+ * no UncaughtExceptionHandler, and non-async LIFO processing mode.
+ *
+ * @param parallelism the parallelism level
* @throws IllegalArgumentException if parallelism less than or
- * equal to zero
+ * equal to zero, or greater than implementation limit
* @throws SecurityException if a security manager exists and
* the caller is not permitted to modify threads
* because it does not hold {@link
- * java.lang.RuntimePermission}("modifyThread")
,
+ * java.lang.RuntimePermission}{@code ("modifyThread")}
*/
public ForkJoinPool(int parallelism) {
- this(parallelism, defaultForkJoinWorkerThreadFactory);
- }
-
- /**
- * Creates a ForkJoinPool with parallelism equal to the number of
- * processors available on the system and using the given
- * ForkJoinWorkerThreadFactory,
- * @param factory the factory for creating new threads
- * @throws NullPointerException if factory is null
- * @throws SecurityException if a security manager exists and
- * the caller is not permitted to modify threads
- * because it does not hold {@link
- * java.lang.RuntimePermission}("modifyThread")
,
- */
- public ForkJoinPool(ForkJoinWorkerThreadFactory factory) {
- this(Runtime.getRuntime().availableProcessors(), factory);
+ this(parallelism, defaultForkJoinWorkerThreadFactory, null, false);
}
/**
- * Creates a ForkJoinPool with the given parallelism and factory.
+ * Creates a {@code ForkJoinPool} with the given parameters.
*
- * @param parallelism the targeted number of worker threads
- * @param factory the factory for creating new threads
+ * @param parallelism the parallelism level. For default value,
+ * use {@link java.lang.Runtime#availableProcessors}.
+ * @param factory the factory for creating new threads. For default value,
+ * use {@link #defaultForkJoinWorkerThreadFactory}.
+ * @param handler the handler for internal worker threads that
+ * terminate due to unrecoverable errors encountered while executing
+ * tasks. For default value, use {@code null}.
+ * @param asyncMode if true,
+ * establishes local first-in-first-out scheduling mode for forked
+ * tasks that are never joined. This mode may be more appropriate
+ * than default locally stack-based mode in applications in which
+ * worker threads only process event-style asynchronous tasks.
+ * For default value, use {@code false}.
* @throws IllegalArgumentException if parallelism less than or
- * equal to zero, or greater than implementation limit.
- * @throws NullPointerException if factory is null
+ * equal to zero, or greater than implementation limit
+ * @throws NullPointerException if the factory is null
* @throws SecurityException if a security manager exists and
* the caller is not permitted to modify threads
* because it does not hold {@link
- * java.lang.RuntimePermission}("modifyThread")
,
+ * java.lang.RuntimePermission}{@code ("modifyThread")}
*/
- public ForkJoinPool(int parallelism, ForkJoinWorkerThreadFactory factory) {
- if (parallelism <= 0 || parallelism > MAX_THREADS)
- throw new IllegalArgumentException();
+ public ForkJoinPool(int parallelism,
+ ForkJoinWorkerThreadFactory factory,
+ Thread.UncaughtExceptionHandler handler,
+ boolean asyncMode) {
+ checkPermission();
if (factory == null)
throw new NullPointerException();
- checkPermission();
- this.factory = factory;
+ if (parallelism <= 0 || parallelism > POOL_MAX)
+ throw new IllegalArgumentException();
this.parallelism = parallelism;
- this.maxPoolSize = MAX_THREADS;
- this.maintainsParallelism = true;
- this.poolNumber = poolNumberGenerator.incrementAndGet();
- this.workerLock = new ReentrantLock();
- this.termination = workerLock.newCondition();
- this.stealCount = new AtomicLong();
- this.submissionQueue = new LinkedTransferQueue("modifyThread")
,
- */
- public Thread.UncaughtExceptionHandler
- setUncaughtExceptionHandler(Thread.UncaughtExceptionHandler h) {
- checkPermission();
- Thread.UncaughtExceptionHandler old = null;
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- old = ueh;
- ueh = h;
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread w = ws[i];
- if (w != null)
- w.setUncaughtExceptionHandler(h);
- }
- }
- } finally {
- lock.unlock();
- }
- return old;
- }
-
-
- /**
- * Sets the target paralleism level of this pool.
- * @param parallelism the target parallelism
- * @throws IllegalArgumentException if parallelism less than or
- * equal to zero or greater than maximum size bounds.
- * @throws SecurityException if a security manager exists and
- * the caller is not permitted to modify threads
- * because it does not hold {@link
- * java.lang.RuntimePermission}("modifyThread")
,
+ * @return the handler, or {@code null} if none
*/
- public void setParallelism(int parallelism) {
- checkPermission();
- if (parallelism <= 0 || parallelism > maxPoolSize)
- throw new IllegalArgumentException();
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- if (!isTerminating()) {
- int p = this.parallelism;
- this.parallelism = parallelism;
- if (parallelism > p)
- createAndStartAddedWorkers();
- else
- trimSpares();
- }
- } finally {
- lock.unlock();
- }
- signalIdleWorkers();
+ public Thread.UncaughtExceptionHandler getUncaughtExceptionHandler() {
+ return ueh;
}
/**
- * Returns the targeted number of worker threads in this pool.
+ * Returns the targeted parallelism level of this pool.
*
- * @return the targeted number of worker threads in this pool
+ * @return the targeted parallelism level of this pool
*/
public int getParallelism() {
return parallelism;
@@ -757,141 +2157,71 @@ public class ForkJoinPool extends Abstra
/**
* Returns the number of worker threads that have started but not
- * yet terminated. This result returned by this method may differ
- * from getParallelism
when threads are created to
+ * yet terminated. The result returned by this method may differ
+ * from {@link #getParallelism} when threads are created to
* maintain parallelism when others are cooperatively blocked.
*
* @return the number of worker threads
*/
public int getPoolSize() {
- return totalCountOf(workerCounts);
+ return parallelism + (short)(ctl >>> TC_SHIFT);
}
/**
- * Returns the maximum number of threads allowed to exist in the
- * pool, even if there are insufficient unblocked running threads.
- * @return the maximum
- */
- public int getMaximumPoolSize() {
- return maxPoolSize;
- }
-
- /**
- * Sets the maximum number of threads allowed to exist in the
- * pool, even if there are insufficient unblocked running threads.
- * Setting this value has no effect on current pool size. It
- * controls construction of new threads.
- * @throws IllegalArgumentException if negative or greater then
- * internal implementation limit.
- */
- public void setMaximumPoolSize(int newMax) {
- if (newMax < 0 || newMax > MAX_THREADS)
- throw new IllegalArgumentException();
- maxPoolSize = newMax;
- }
-
-
- /**
- * Returns true if this pool dynamically maintains its target
- * parallelism level. If false, new threads are added only to
- * avoid possible starvation.
- * This setting is by default true;
- * @return true if maintains parallelism
- */
- public boolean getMaintainsParallelism() {
- return maintainsParallelism;
- }
-
- /**
- * Sets whether this pool dynamically maintains its target
- * parallelism level. If false, new threads are added only to
- * avoid possible starvation.
- * @param enable true to maintains parallelism
- */
- public void setMaintainsParallelism(boolean enable) {
- maintainsParallelism = enable;
- }
-
- /**
- * Establishes local first-in-first-out scheduling mode for forked
- * tasks that are never joined. This mode may be more appropriate
- * than default locally stack-based mode in applications in which
- * worker threads only process asynchronous tasks. This method is
- * designed to be invoked only when pool is quiescent, and
- * typically only before any tasks are submitted. The effects of
- * invocations at ather times may be unpredictable.
- *
- * @param async if true, use locally FIFO scheduling
- * @return the previous mode.
- */
- public boolean setAsyncMode(boolean async) {
- boolean oldMode = locallyFifo;
- locallyFifo = async;
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.setAsyncMode(async);
- }
- }
- return oldMode;
- }
-
- /**
- * Returns true if this pool uses local first-in-first-out
+ * Returns {@code true} if this pool uses local first-in-first-out
* scheduling mode for forked tasks that are never joined.
*
- * @return true if this pool uses async mode.
+ * @return {@code true} if this pool uses async mode
*/
public boolean getAsyncMode() {
- return locallyFifo;
+ return localMode != 0;
}
/**
* Returns an estimate of the number of worker threads that are
* not blocked waiting to join tasks or for other managed
- * synchronization.
+ * synchronization. This method may overestimate the
+ * number of running threads.
*
* @return the number of worker threads
*/
public int getRunningThreadCount() {
- return runningCountOf(workerCounts);
+ int rc = 0;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && w.isApparentlyUnblocked())
+ ++rc;
+ }
+ }
+ return rc;
}
/**
* Returns an estimate of the number of threads that are currently
* stealing or executing tasks. This method may overestimate the
* number of active threads.
- * @return the number of active threads.
+ *
+ * @return the number of active threads
*/
public int getActiveThreadCount() {
- return activeCountOf(runControl);
- }
-
- /**
- * Returns an estimate of the number of threads that are currently
- * idle waiting for tasks. This method may underestimate the
- * number of idle threads.
- * @return the number of idle threads.
- */
- final int getIdleThreadCount() {
- int c = runningCountOf(workerCounts) - activeCountOf(runControl);
- return (c <= 0)? 0 : c;
+ int r = parallelism + (int)(ctl >> AC_SHIFT);
+ return (r <= 0) ? 0 : r; // suppress momentarily negative values
}
/**
- * Returns true if all worker threads are currently idle. An idle
- * worker is one that cannot obtain a task to execute because none
- * are available to steal from other threads, and there are no
- * pending submissions to the pool. This method is conservative:
- * It might not return true immediately upon idleness of all
- * threads, but will eventually become true if threads remain
- * inactive.
- * @return true if all threads are currently idle
+ * Returns {@code true} if all worker threads are currently idle.
+ * An idle worker is one that cannot obtain a task to execute
+ * because none are available to steal from other threads, and
+ * there are no pending submissions to the pool. This method is
+ * conservative; it might not return {@code true} immediately upon
+ * idleness of all threads, but will eventually become true if
+ * threads remain inactive.
+ *
+ * @return {@code true} if all threads are currently idle
*/
public boolean isQuiescent() {
- return activeCountOf(runControl) == 0;
+ return (int)(ctl >> AC_SHIFT) + parallelism == 0;
}
/**
@@ -899,23 +2229,22 @@ public class ForkJoinPool extends Abstra
* one thread's work queue by another. The reported value
* underestimates the actual total number of steals when the pool
* is not quiescent. This value may be useful for monitoring and
- * tuning fork/join programs: In general, steal counts should be
+ * tuning fork/join programs: in general, steal counts should be
* high enough to keep threads busy, but low enough to avoid
* overhead and contention across threads.
- * @return the number of steals.
+ *
+ * @return the number of steals
*/
public long getStealCount() {
- return stealCount.get();
- }
-
- /**
- * Accumulate steal count from a worker. Call only
- * when worker known to be idle.
- */
- private void updateStealCount(ForkJoinWorkerThread w) {
- int sc = w.getAndClearStealCount();
- if (sc != 0)
- stealCount.addAndGet(sc);
+ long count = stealCount.get();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.totalSteals;
+ }
+ }
+ return count;
}
/**
@@ -925,77 +2254,106 @@ public class ForkJoinPool extends Abstra
* an approximation, obtained by iterating across all threads in
* the pool. This method may be useful for tuning task
* granularities.
- * @return the number of queued tasks.
+ *
+ * @return the number of queued tasks
*/
public long getQueuedTaskCount() {
long count = 0;
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- count += t.getQueueSize();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 1; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
}
}
return count;
}
/**
- * Returns an estimate of the number tasks submitted to this pool
- * that have not yet begun executing. This method takes time
- * proportional to the number of submissions.
- * @return the number of queued submissions.
+ * Returns an estimate of the number of tasks submitted to this
+ * pool that have not yet begun executing. This method may take
+ * time proportional to the number of submissions.
+ *
+ * @return the number of queued submissions
*/
public int getQueuedSubmissionCount() {
- return submissionQueue.size();
+ int count = 0;
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null)
+ count += w.queueSize();
+ }
+ }
+ return count;
}
/**
- * Returns true if there are any tasks submitted to this pool
- * that have not yet begun executing.
- * @return true
if there are any queued submissions.
+ * Returns {@code true} if there are any tasks submitted to this
+ * pool that have not yet begun executing.
+ *
+ * @return {@code true} if there are any queued submissions
*/
public boolean hasQueuedSubmissions() {
- return !submissionQueue.isEmpty();
+ WorkQueue[] ws; WorkQueue w;
+ if ((ws = workQueues) != null) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && w.queueSize() != 0)
+ return true;
+ }
+ }
+ return false;
}
/**
* Removes and returns the next unexecuted submission if one is
* available. This method may be useful in extensions to this
* class that re-assign work in systems with multiple pools.
- * @return the next submission, or null if none
+ *
+ * @return the next submission, or {@code null} if none
*/
protected ForkJoinTask> pollSubmission() {
- return submissionQueue.poll();
+ WorkQueue[] ws; WorkQueue w; ForkJoinTask> t;
+ if ((ws = workQueues) != null) {
+ for (int i = 0; i < ws.length; i += 2) {
+ if ((w = ws[i]) != null && (t = w.poll()) != null)
+ return t;
+ }
+ }
+ return null;
}
/**
* Removes all available unexecuted submitted and forked tasks
* from scheduling queues and adds them to the given collection,
* without altering their execution status. These may include
- * artifically generated or wrapped tasks. This method id designed
- * to be invoked only when the pool is known to be
+ * artificially generated or wrapped tasks. This method is
+ * designed to be invoked only when the pool is known to be
* quiescent. Invocations at other times may not remove all
* tasks. A failure encountered while attempting to add elements
- * to collection c may result in elements being in
+ * to collection {@code c} may result in elements being in
* neither, either or both collections when the associated
* exception is thrown. The behavior of this operation is
* undefined if the specified collection is modified while the
* operation is in progress.
+ *
* @param c the collection to transfer elements into
* @return the number of elements transferred
*/
- protected int drainTasksTo(Collection("modifyThread")
,
+ * java.lang.RuntimePermission}{@code ("modifyThread")}
*/
public void shutdown() {
checkPermission();
- transitionRunStateTo(SHUTDOWN);
- if (canTerminateOnShutdown(runControl))
- terminateOnShutdown();
+ tryTerminate(false, true);
}
/**
- * Attempts to stop all actively executing tasks, and cancels all
- * waiting tasks. Tasks that are in the process of being
- * submitted or executed concurrently during the course of this
- * method may or may not be rejected. Unlike some other executors,
- * this method cancels rather than collects non-executed tasks
- * upon termination, so always returns an empty list. However, you
- * can use method drainTasksTo
before invoking this
- * method to transfer unexecuted tasks to another collection.
+ * Attempts to cancel and/or stop all tasks, and reject all
+ * subsequently submitted tasks. Tasks that are in the process of
+ * being submitted or executed concurrently during the course of
+ * this method may or may not be rejected. This method cancels
+ * both existing and unexecuted tasks, in order to permit
+ * termination in the presence of task dependencies. So the method
+ * always returns an empty list (unlike the case for some other
+ * Executors).
+ *
* @return an empty list
* @throws SecurityException if a security manager exists and
* the caller is not permitted to modify threads
* because it does not hold {@link
- * java.lang.RuntimePermission}("modifyThread")
,
+ * java.lang.RuntimePermission}{@code ("modifyThread")}
*/
public Listtrue
if all tasks have completed following shut down.
+ * Returns {@code true} if all tasks have completed following shut down.
*
- * @return true
if all tasks have completed following shut down
+ * @return {@code true} if all tasks have completed following shut down
*/
public boolean isTerminated() {
- return runStateOf(runControl) == TERMINATED;
+ long c = ctl;
+ return ((c & STOP_BIT) != 0L &&
+ (short)(c >>> TC_SHIFT) == -parallelism);
}
/**
- * Returns true
if the process of termination has
- * commenced but possibly not yet completed.
+ * Returns {@code true} if the process of termination has
+ * commenced but not yet completed. This method may be useful for
+ * debugging. A return of {@code true} reported a sufficient
+ * period after shutdown may indicate that submitted tasks have
+ * ignored or suppressed interruption, or are waiting for IO,
+ * causing this executor not to properly terminate. (See the
+ * advisory notes for class {@link ForkJoinTask} stating that
+ * tasks should not normally entail blocking operations. But if
+ * they do, they must abort them on interrupt.)
*
- * @return true
if terminating
+ * @return {@code true} if terminating but not yet terminated
*/
public boolean isTerminating() {
- return runStateOf(runControl) >= TERMINATING;
+ long c = ctl;
+ return ((c & STOP_BIT) != 0L &&
+ (short)(c >>> TC_SHIFT) != -parallelism);
}
/**
- * Returns true
if this pool has been shut down.
+ * Returns {@code true} if this pool has been shut down.
*
- * @return true
if this pool has been shut down
+ * @return {@code true} if this pool has been shut down
*/
public boolean isShutdown() {
- return runStateOf(runControl) >= SHUTDOWN;
+ return runState < 0;
}
/**
@@ -1110,14 +2491,14 @@ public class ForkJoinPool extends Abstra
*
* @param timeout the maximum time to wait
* @param unit the time unit of the timeout argument
- * @return true
if this executor terminated and
- * false
if the timeout elapsed before termination
+ * @return {@code true} if this executor terminated and
+ * {@code false} if the timeout elapsed before termination
* @throws InterruptedException if interrupted while waiting
*/
public boolean awaitTermination(long timeout, TimeUnit unit)
throws InterruptedException {
long nanos = unit.toNanos(timeout);
- final ReentrantLock lock = this.workerLock;
+ final Mutex lock = this.lock;
lock.lock();
try {
for (;;) {
@@ -1132,729 +2513,179 @@ public class ForkJoinPool extends Abstra
}
}
- // Shutdown and termination support
-
- /**
- * Callback from terminating worker. Null out the corresponding
- * workers slot, and if terminating, try to terminate, else try to
- * shrink workers array.
- * @param w the worker
- */
- final void workerTerminated(ForkJoinWorkerThread w) {
- updateStealCount(w);
- updateWorkerCount(-1);
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- int idx = w.poolIndex;
- if (idx >= 0 && idx < ws.length && ws[idx] == w)
- ws[idx] = null;
- if (totalCountOf(workerCounts) == 0) {
- terminate(); // no-op if already terminating
- transitionRunStateTo(TERMINATED);
- termination.signalAll();
- }
- else if (!isTerminating()) {
- tryShrinkWorkerArray();
- tryResumeSpare(true); // allow replacement
- }
- }
- } finally {
- lock.unlock();
- }
- signalIdleWorkers();
- }
-
/**
- * Initiate termination.
- */
- private void terminate() {
- if (transitionRunStateTo(TERMINATING)) {
- stopAllWorkers();
- resumeAllSpares();
- signalIdleWorkers();
- cancelQueuedSubmissions();
- cancelQueuedWorkerTasks();
- interruptUnterminatedWorkers();
- signalIdleWorkers(); // resignal after interrupt
- }
- }
-
- /**
- * Possibly terminate when on shutdown state
- */
- private void terminateOnShutdown() {
- if (!hasQueuedSubmissions() && canTerminateOnShutdown(runControl))
- terminate();
- }
-
- /**
- * Clear out and cancel submissions
- */
- private void cancelQueuedSubmissions() {
- ForkJoinTask> task;
- while ((task = pollSubmission()) != null)
- task.cancel(false);
- }
-
- /**
- * Clean out worker queues.
- */
- private void cancelQueuedWorkerTasks() {
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.cancelTasks();
- }
- }
- } finally {
- lock.unlock();
- }
- }
-
- /**
- * Set each worker's status to terminating. Requires lock to avoid
- * conflicts with add/remove
- */
- private void stopAllWorkers() {
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null)
- t.shutdownNow();
- }
- }
- } finally {
- lock.unlock();
- }
- }
-
- /**
- * Interrupt all unterminated workers. This is not required for
- * sake of internal control, but may help unstick user code during
- * shutdown.
- */
- private void interruptUnterminatedWorkers() {
- final ReentrantLock lock = this.workerLock;
- lock.lock();
- try {
- ForkJoinWorkerThread[] ws = workers;
- if (ws != null) {
- for (int i = 0; i < ws.length; ++i) {
- ForkJoinWorkerThread t = ws[i];
- if (t != null && !t.isTerminated()) {
- try {
- t.interrupt();
- } catch (SecurityException ignore) {
- }
- }
- }
- }
- } finally {
- lock.unlock();
- }
- }
-
-
- /*
- * Nodes for event barrier to manage idle threads. Queue nodes
- * are basic Treiber stack nodes, also used for spare stack.
+ * Interface for extending managed parallelism for tasks running
+ * in {@link ForkJoinPool}s.
*
- * The event barrier has an event count and a wait queue (actually
- * a Treiber stack). Workers are enabled to look for work when
- * the eventCount is incremented. If they fail to find work, they
- * may wait for next count. Upon release, threads help others wake
- * up.
- *
- * Synchronization events occur only in enough contexts to
- * maintain overall liveness:
- *
- * - Submission of a new task to the pool
- * - Resizes or other changes to the workers array
- * - pool termination
- * - A worker pushing a task on an empty queue
- *
- * The case of pushing a task occurs often enough, and is heavy
- * enough compared to simple stack pushes, to require special
- * handling: Method signalWork returns without advancing count if
- * the queue appears to be empty. This would ordinarily result in
- * races causing some queued waiters not to be woken up. To avoid
- * this, the first worker enqueued in method sync (see
- * syncIsReleasable) rescans for tasks after being enqueued, and
- * helps signal if any are found. This works well because the
- * worker has nothing better to do, and so might as well help
- * alleviate the overhead and contention on the threads actually
- * doing work. Also, since event counts increments on task
- * availability exist to maintain liveness (rather than to force
- * refreshes etc), it is OK for callers to exit early if
- * contending with another signaller.
- */
- static final class WaitQueueNode {
- WaitQueueNode next; // only written before enqueued
- volatile ForkJoinWorkerThread thread; // nulled to cancel wait
- final long count; // unused for spare stack
-
- WaitQueueNode(long c, ForkJoinWorkerThread w) {
- count = c;
- thread = w;
- }
-
- /**
- * Wake up waiter, returning false if known to already
- */
- boolean signal() {
- ForkJoinWorkerThread t = thread;
- if (t == null)
- return false;
- thread = null;
- LockSupport.unpark(t);
- return true;
- }
-
- /**
- * Await release on sync
- */
- void awaitSyncRelease(ForkJoinPool p) {
- while (thread != null && !p.syncIsReleasable(this))
- LockSupport.park(this);
- }
-
- /**
- * Await resumption as spare
- */
- void awaitSpareRelease() {
- while (thread != null) {
- if (!Thread.interrupted())
- LockSupport.park(this);
- }
- }
- }
-
- /**
- * Ensures that no thread is waiting for count to advance from the
- * current value of eventCount read on entry to this method, by
- * releasing waiting threads if necessary.
- * @return the count
- */
- final long ensureSync() {
- long c = eventCount;
- WaitQueueNode q;
- while ((q = syncStack) != null && q.count < c) {
- if (casBarrierStack(q, null)) {
- do {
- q.signal();
- } while ((q = q.next) != null);
- break;
- }
- }
- return c;
- }
-
- /**
- * Increments event count and releases waiting threads.
- */
- private void signalIdleWorkers() {
- long c;
- do;while (!casEventCount(c = eventCount, c+1));
- ensureSync();
- }
-
- /**
- * Signal threads waiting to poll a task. Because method sync
- * rechecks availability, it is OK to only proceed if queue
- * appears to be non-empty, and OK to skip under contention to
- * increment count (since some other thread succeeded).
- */
- final void signalWork() {
- long c;
- WaitQueueNode q;
- if (syncStack != null &&
- casEventCount(c = eventCount, c+1) &&
- (((q = syncStack) != null && q.count <= c) &&
- (!casBarrierStack(q, q.next) || !q.signal())))
- ensureSync();
- }
-
- /**
- * Waits until event count advances from last value held by
- * caller, or if excess threads, caller is resumed as spare, or
- * caller or pool is terminating. Updates caller's event on exit.
- * @param w the calling worker thread
- */
- final void sync(ForkJoinWorkerThread w) {
- updateStealCount(w); // Transfer w's count while it is idle
-
- while (!w.isShutdown() && !isTerminating() && !suspendIfSpare(w)) {
- long prev = w.lastEventCount;
- WaitQueueNode node = null;
- WaitQueueNode h;
- while (eventCount == prev &&
- ((h = syncStack) == null || h.count == prev)) {
- if (node == null)
- node = new WaitQueueNode(prev, w);
- if (casBarrierStack(node.next = h, node)) {
- node.awaitSyncRelease(this);
- break;
- }
- }
- long ec = ensureSync();
- if (ec != prev) {
- w.lastEventCount = ec;
- break;
- }
- }
- }
-
- /**
- * Returns true if worker waiting on sync can proceed:
- * - on signal (thread == null)
- * - on event count advance (winning race to notify vs signaller)
- * - on Interrupt
- * - if the first queued node, we find work available
- * If node was not signalled and event count not advanced on exit,
- * then we also help advance event count.
- * @return true if node can be released
- */
- final boolean syncIsReleasable(WaitQueueNode node) {
- long prev = node.count;
- if (!Thread.interrupted() && node.thread != null &&
- (node.next != null ||
- !ForkJoinWorkerThread.hasQueuedTasks(workers)) &&
- eventCount == prev)
- return false;
- if (node.thread != null) {
- node.thread = null;
- long ec = eventCount;
- if (prev <= ec) // help signal
- casEventCount(ec, ec+1);
- }
- return true;
- }
-
- /**
- * Returns true if a new sync event occurred since last call to
- * sync or this method, if so, updating caller's count.
- */
- final boolean hasNewSyncEvent(ForkJoinWorkerThread w) {
- long lc = w.lastEventCount;
- long ec = ensureSync();
- if (ec == lc)
- return false;
- w.lastEventCount = ec;
- return true;
- }
-
- // Parallelism maintenance
-
- /**
- * Decrement running count; if too low, add spare.
+ * isReleasable
must return true if blocking is not
- * necessary. Method block
blocks the current thread
- * if necessary (perhaps internally invoking isReleasable before
- * actually blocking.).
*
- * class ManagedLocker implements ManagedBlocker {
- * final ReentrantLock lock;
- * boolean hasLock = false;
- * ManagedLocker(ReentrantLock lock) { this.lock = lock; }
- * public boolean block() {
- * if (!hasLock)
- * lock.lock();
- * return true;
- * }
- * public boolean isReleasable() {
- * return hasLock || (hasLock = lock.tryLock());
- * }
+ *
*/
public static interface ManagedBlocker {
/**
* Possibly blocks the current thread, for example waiting for
* a lock or condition.
- * @return true if no additional blocking is necessary (i.e.,
- * if isReleasable would return true).
+ *
+ * @return {@code true} if no additional blocking is necessary
+ * (i.e., if isReleasable would return true)
* @throws InterruptedException if interrupted while waiting
- * (the method is not required to do so, but is allowe to).
+ * (the method is not required to do so, but is allowed to)
*/
boolean block() throws InterruptedException;
/**
- * Returns true if blocking is unnecessary.
+ * Returns {@code true} if blocking is unnecessary.
*/
boolean isReleasable();
}
/**
* Blocks in accord with the given blocker. If the current thread
- * is a ForkJoinWorkerThread, this method possibly arranges for a
- * spare thread to be activated if necessary to ensure parallelism
- * while the current thread is blocked. If
- * {@code
+ * class ManagedLocker implements ManagedBlocker {
+ * final ReentrantLock lock;
+ * boolean hasLock = false;
+ * ManagedLocker(ReentrantLock lock) { this.lock = lock; }
+ * public boolean block() {
+ * if (!hasLock)
+ * lock.lock();
+ * return true;
+ * }
+ * public boolean isReleasable() {
+ * return hasLock || (hasLock = lock.tryLock());
+ * }
+ * }}
+ *
+ * {@code
+ * class QueueTaker
+ * public boolean isReleasable() {
+ * return item != null || (item = queue.poll()) != null;
+ * }
+ * public E getItem() { // call after pool.managedBlock completes
+ * return item;
+ * }
+ * }}maintainParallelism
is true and the pool supports
- * it ({@link #getMaintainsParallelism}), this method attempts to
- * maintain the pool's nominal parallelism. Otherwise if activates
- * a thread only if necessary to avoid complete starvation. This
- * option may be preferable when blockages use timeouts, or are
- * almost always brief.
- *
- *
- * while (!blocker.isReleasable())
- * if (blocker.block())
- * return;
- *
- * If the caller is a ForkJoinTask, then the pool may first
- * be expanded to ensure parallelism, and later adjusted.
+ * is a {@link ForkJoinWorkerThread}, this method possibly
+ * arranges for a spare thread to be activated if necessary to
+ * ensure sufficient parallelism while the current thread is blocked.
+ *
+ * {@code
+ * while (!blocker.isReleasable())
+ * if (blocker.block())
+ * return;
+ * }
+ *
+ * If the caller is a {@code ForkJoinTask}, then the pool may
+ * first be expanded to ensure parallelism, and later adjusted.
*
* @param blocker the blocker
- * @param maintainParallelism if true and supported by this pool,
- * attempt to maintain the pool's nominal parallelism; otherwise
- * activate a thread only if necessary to avoid complete
- * starvation.
- * @throws InterruptedException if blocker.block did so.
+ * @throws InterruptedException if blocker.block did so
*/
- public static void managedBlock(ManagedBlocker blocker,
- boolean maintainParallelism)
+ public static void managedBlock(ManagedBlocker blocker)
throws InterruptedException {
Thread t = Thread.currentThread();
- ForkJoinPool pool = (t instanceof ForkJoinWorkerThread?
- ((ForkJoinWorkerThread)t).pool : null);
- if (!blocker.isReleasable()) {
- try {
- if (pool == null ||
- !pool.preBlock(blocker, maintainParallelism))
- awaitBlocker(blocker);
- } finally {
- if (pool != null)
- pool.updateRunningCount(1);
+ ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
+ ((ForkJoinWorkerThread)t).pool : null);
+ while (!blocker.isReleasable()) {
+ if (p == null || p.tryCompensate()) {
+ try {
+ do {} while (!blocker.isReleasable() && !blocker.block());
+ } finally {
+ if (p != null)
+ p.incrementActiveCount();
+ }
+ break;
}
}
}
- private static void awaitBlocker(ManagedBlocker blocker)
- throws InterruptedException {
- do;while (!blocker.isReleasable() && !blocker.block());
- }
-
- // AbstractExecutorService overrides
+ // AbstractExecutorService overrides. These rely on undocumented
+ // fact that ForkJoinTask.adapt returns ForkJoinTasks that also
+ // implement RunnableFuture.
protected