1 |
dl |
1.1 |
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN"> |
2 |
|
|
<html> <head> |
3 |
|
|
<title>Concurrency Utilities</title> |
4 |
|
|
</head> |
5 |
|
|
|
6 |
|
|
<body> |
7 |
|
|
|
8 |
dl |
1.3 |
<p> Utility classes commonly useful in concurrent programming. This |
9 |
|
|
package includes a few small standardized extensible frameworks, as |
10 |
|
|
well as some classes that provide useful functionality and are |
11 |
|
|
otherwise tedious or difficult to implement. Here are brief |
12 |
|
|
descriptions of the main components. See also the <tt>locks</tt> and |
13 |
|
|
<tt>atomic</tt> packages. |
14 |
dl |
1.1 |
|
15 |
|
|
<h2>Executors</h2> |
16 |
|
|
|
17 |
dl |
1.12 |
<b>Interfaces.</b> {@link java.util.concurrent.Executor} is a simple |
18 |
|
|
standardized interface for defining custom thread-like subsystems, |
19 |
|
|
including thread pools, asynchronous IO, and lightweight task |
20 |
|
|
frameworks. Depending on which concrete Executor class is being used, |
21 |
|
|
tasks may execute in a newly created thread, an existing |
22 |
|
|
task-execution thread, or the thread calling <tt>execute()</tt>, and |
23 |
|
|
may execute sequentially or concurrently. {@link |
24 |
|
|
java.util.concurrent.ExecutorService} provides a more complete |
25 |
|
|
asynchronous task execution framework. An ExecutorService manages |
26 |
dl |
1.14 |
queuing and scheduling of tasks, and allows controlled shutdown. The |
27 |
dl |
1.12 |
{@link java.util.concurrent.ScheduledExecutorService} subinterface |
28 |
dl |
1.19 |
and associated interfaces add support for delayed and periodic task execution. |
29 |
dl |
1.12 |
ExecutorServices provide methods arranging asynchronous execution of |
30 |
|
|
any function expressed as {@link java.util.concurrent.Callable}, the |
31 |
|
|
result-bearing analog of {@link java.lang.Runnable}. A {@link |
32 |
|
|
java.util.concurrent.Future} returns the results of a function, allows |
33 |
dl |
1.16 |
determination of whether execution has completed, and provides a means to |
34 |
jsr166 |
1.20 |
cancel execution. A {@link java.util.concurrent.RunnableFuture} is |
35 |
dl |
1.19 |
a Future that possesses a <tt>run</tt> method that upon execution, |
36 |
|
|
sets its results. |
37 |
dl |
1.12 |
|
38 |
|
|
<p> |
39 |
|
|
|
40 |
dl |
1.16 |
<b>Implementations.</b> Classes {@link |
41 |
dl |
1.12 |
java.util.concurrent.ThreadPoolExecutor} and {@link |
42 |
dl |
1.16 |
java.util.concurrent.ScheduledThreadPoolExecutor} provide tunable, |
43 |
|
|
flexible thread pools. The {@link java.util.concurrent.Executors} |
44 |
|
|
class provides factory methods for the most common kinds and |
45 |
|
|
configurations of Executors, as well as a few utility methods for |
46 |
|
|
using them. Other utilities based on Executors include the concrete |
47 |
|
|
class {@link java.util.concurrent.FutureTask} providing a common |
48 |
|
|
extensible implementation of Futures, and {@link |
49 |
dl |
1.12 |
java.util.concurrent.ExecutorCompletionService}, that assists in |
50 |
|
|
coordinating the processing of groups of asynchronous tasks. |
51 |
dl |
1.1 |
|
52 |
|
|
<h2>Queues</h2> |
53 |
|
|
|
54 |
dl |
1.6 |
The java.util.concurrent {@link |
55 |
dl |
1.1 |
java.util.concurrent.ConcurrentLinkedQueue} class supplies an |
56 |
dl |
1.6 |
efficient scalable thread-safe non-blocking FIFO queue. Five |
57 |
|
|
implementations in java.util.concurrent support the extended {@link |
58 |
|
|
java.util.concurrent.BlockingQueue} interface, that defines blocking |
59 |
|
|
versions of put and take: {@link |
60 |
dl |
1.1 |
java.util.concurrent.LinkedBlockingQueue}, {@link |
61 |
|
|
java.util.concurrent.ArrayBlockingQueue}, {@link |
62 |
|
|
java.util.concurrent.SynchronousQueue}, {@link |
63 |
|
|
java.util.concurrent.PriorityBlockingQueue}, and {@link |
64 |
dl |
1.2 |
java.util.concurrent.DelayQueue}. The different classes cover the most |
65 |
|
|
common usage contexts for producer-consumer, messaging, parallel |
66 |
dl |
1.17 |
tasking, and related concurrent designs. The {@link |
67 |
|
|
java.util.concurrent.BlockingDeque} interface extends |
68 |
|
|
<tt>BlockingQueue</tt> to support both FIFO and LIFO (stack-based) |
69 |
|
|
operations. Class {@link java.util.concurrent.LinkedBlockingDeque} |
70 |
|
|
provides an implementation. |
71 |
dl |
1.1 |
|
72 |
|
|
|
73 |
|
|
<h2>Timing</h2> |
74 |
|
|
|
75 |
|
|
The {@link java.util.concurrent.TimeUnit} class provides multiple |
76 |
dl |
1.2 |
granularities (including nanoseconds) for specifying and controlling |
77 |
jsr166 |
1.24 |
time-out based operations. Most classes in the package contain |
78 |
dl |
1.9 |
operations based on time-outs in addition to indefinite waits. In all |
79 |
|
|
cases that time-outs are used, the time-out specifies the minimum time |
80 |
|
|
that the method should wait before indicating that it |
81 |
jsr166 |
1.24 |
timed-out. Implementations make a "best effort" to detect |
82 |
|
|
time-outs as soon as possible after they occur. However, an indefinite |
83 |
dl |
1.9 |
amount of time may elapse between a time-out being detected and a |
84 |
jsr166 |
1.24 |
thread actually executing again after that time-out. All methods |
85 |
dl |
1.18 |
that accept timeout parameters treat values less than or equal to |
86 |
jsr166 |
1.24 |
zero to mean not to wait at all. To wait "forever", you can use |
87 |
dl |
1.18 |
a value of <tt>Long.MAX_VALUE</tt>. |
88 |
dholmes |
1.4 |
|
89 |
dl |
1.1 |
<h2>Synchronizers</h2> |
90 |
|
|
|
91 |
dl |
1.13 |
Four classes aid common special-purpose synchronization idioms. |
92 |
dl |
1.10 |
{@link java.util.concurrent.Semaphore} is a classic concurrency tool. |
93 |
dl |
1.16 |
{@link java.util.concurrent.CountDownLatch} is a very simple yet very |
94 |
dl |
1.11 |
common utility for blocking until a given number of signals, events, |
95 |
|
|
or conditions hold. A {@link java.util.concurrent.CyclicBarrier} is a |
96 |
dl |
1.13 |
resettable multiway synchronization point useful in some styles of |
97 |
dl |
1.11 |
parallel programming. An {@link java.util.concurrent.Exchanger} allows |
98 |
dl |
1.13 |
two threads to exchange objects at a rendezvous point, and is useful |
99 |
|
|
in several pipeline designs. |
100 |
dl |
1.1 |
|
101 |
|
|
<h2>Concurrent Collections</h2> |
102 |
|
|
|
103 |
dl |
1.17 |
Besides Queues, this package supplies Collection implementations |
104 |
|
|
designed for use in multithreaded contexts: |
105 |
|
|
{@link java.util.concurrent.ConcurrentHashMap}, |
106 |
|
|
{@link java.util.concurrent.ConcurrentSkipListMap}, |
107 |
|
|
{@link java.util.concurrent.ConcurrentSkipListSet}, |
108 |
|
|
{@link java.util.concurrent.CopyOnWriteArrayList}, and |
109 |
|
|
{@link java.util.concurrent.CopyOnWriteArraySet}. |
110 |
|
|
When many threads are expected to access a given collection, |
111 |
|
|
a <tt>ConcurrentHashMap</tt> is normally preferable to |
112 |
|
|
a synchronized <tt>HashMap</tt>, and a |
113 |
|
|
<tt>ConcurrentSkipListMap</tt> is normally preferable |
114 |
|
|
to a synchronized <tt>TreeMap</tt>. A |
115 |
|
|
<tt>CopyOnWriteArrayList</tt> is preferable to |
116 |
|
|
a synchronized <tt>ArrayList</tt> when the expected number of reads |
117 |
|
|
and traversals greatly outnumber the number of updates to a list. |
118 |
dl |
1.1 |
|
119 |
dl |
1.6 |
<p>The "Concurrent" prefix used with some classes in this package is a |
120 |
|
|
shorthand indicating several differences from similar "synchronized" |
121 |
dl |
1.1 |
classes. For example <tt>java.util.Hashtable</tt> and |
122 |
|
|
<tt>Collections.synchronizedMap(new HashMap())</tt> are |
123 |
dl |
1.6 |
synchronized. But {@link java.util.concurrent.ConcurrentHashMap} is |
124 |
|
|
"concurrent". A concurrent collection is thread-safe, but not |
125 |
|
|
governed by a single exclusion lock. In the particular case of |
126 |
|
|
ConcurrentHashMap, it safely permits any number of concurrent reads as |
127 |
dl |
1.9 |
well as a tunable number of concurrent writes. "Synchronized" classes |
128 |
|
|
can be useful when you need to prevent all access to a collection via |
129 |
|
|
a single lock, at the expense of poorer scalability. In other cases in |
130 |
|
|
which multiple threads are expected to access a common collection, |
131 |
|
|
"concurrent" versions are normally preferable. And unsynchronized |
132 |
|
|
collections are preferable when either collections are unshared, or |
133 |
|
|
are accessible only when holding other locks. |
134 |
dl |
1.1 |
|
135 |
|
|
<p> Most concurrent Collection implementations (including most Queues) |
136 |
|
|
also differ from the usual java.util conventions in that their Iterators |
137 |
|
|
provide <em>weakly consistent</em> rather than fast-fail traversal. A |
138 |
|
|
weakly consistent iterator is thread-safe, but does not necessarily |
139 |
|
|
freeze the collection while iterating, so it may (or may not) reflect |
140 |
|
|
any updates since the iterator was created. |
141 |
|
|
|
142 |
brian |
1.26 |
<a name="MemoryVisibility"> |
143 |
dl |
1.21 |
<h2> Memory Consistency Properties </h2> |
144 |
|
|
|
145 |
|
|
<a href="http://java.sun.com/docs/books/jls/third_edition/html/memory.html"> |
146 |
|
|
Chapter 17 of the Java Language Specification</a> defines the |
147 |
|
|
<i>happens-before</i> relation on memory operations such as reads and |
148 |
|
|
writes of shared variables. The results of a write by one thread are |
149 |
|
|
guaranteed to be visible to a read by another thread only if the write |
150 |
|
|
operation <i>happens-before</i> the read operation. The |
151 |
|
|
<tt>synchronized</tt> and <tt>volatile</tt> constructs, as well as the |
152 |
|
|
<tt>Thread.start()</tt> and <tt>Thread.join()</tt> methods, can form |
153 |
|
|
<i>happens-before</i> relationships. In particular: |
154 |
|
|
|
155 |
|
|
<ul> |
156 |
brian |
1.26 |
<li>Each action in a thread <i>happens-before</i> every action in that |
157 |
dl |
1.21 |
thread that comes later in the program's order. |
158 |
|
|
|
159 |
jsr166 |
1.24 |
<li>An unlock (<tt>synchronized</tt> block or method exit) of a |
160 |
brian |
1.26 |
monitor <i>happens-before</i> every subsequent lock (<tt>synchronized</tt> |
161 |
dl |
1.23 |
block or method entry) of that same monitor. And because |
162 |
|
|
the <i>happens-before</i> relation is transitive, all actions |
163 |
brian |
1.27 |
of a thread prior to unlocking <i>happen-before</i> all actions |
164 |
dl |
1.23 |
subsequent to any thread locking that monitor. |
165 |
dl |
1.21 |
|
166 |
brian |
1.26 |
<li>A write to a <tt>volatile</tt> field <i>happens-before</i> every |
167 |
dl |
1.23 |
subsequent read of that same field. Writes and reads of |
168 |
|
|
<tt>volatile</tt> fields have similar memory consistency effects |
169 |
|
|
as entering and exiting monitors, but do <em>not</em> entail |
170 |
|
|
mutual exclusion locking. |
171 |
dl |
1.21 |
|
172 |
brian |
1.26 |
<li>A call to <tt>start</tt> on a thread <i>happens-before</i> any action in the |
173 |
dl |
1.21 |
started thread. |
174 |
|
|
|
175 |
brian |
1.26 |
<li>All actions in a thread <i>happen-before</i> any other thread |
176 |
dl |
1.21 |
successfully returns from a <tt>join</tt> on that thread. |
177 |
|
|
|
178 |
|
|
</ul> |
179 |
|
|
|
180 |
dl |
1.23 |
|
181 |
dl |
1.21 |
The methods of all classes in <tt>java.util.concurrent</tt> and its |
182 |
|
|
subpackages extend these guarantees to higher-level |
183 |
|
|
synchronization. In particular: |
184 |
|
|
|
185 |
|
|
<ul> |
186 |
|
|
|
187 |
|
|
<li>State changes to any object made prior to placement into any |
188 |
brian |
1.27 |
concurrent collection <i>happen-before</i> this element is accessed via or |
189 |
dl |
1.21 |
removed from that collection. |
190 |
|
|
|
191 |
brian |
1.27 |
<li>State changes to a <tt>Runnable</tt> object made by the |
192 |
|
|
submitting thread prior to submission to an <tt>Executor</tt> |
193 |
|
|
<i>happen-before</i> its execution. Similarly, state changes to a |
194 |
|
|
<tt>Callable</tt> object made by the submitting thread before being |
195 |
|
|
submitted to an <tt>ExecutorService</tt> <i>happen-before</i> its |
196 |
|
|
execution. |
197 |
dl |
1.21 |
|
198 |
dl |
1.28 |
<li>State changes to a <tt>Future</tt> that |
199 |
|
|
<i>happen-before</i> it is made available |
200 |
|
|
<i>happen-before</i> access via <tt>Future.get()</tt>. |
201 |
dl |
1.21 |
|
202 |
|
|
<li>Actions prior to "releasing" synchronizer methods such as |
203 |
dl |
1.28 |
<tt>Lock.unlock</tt>, <tt>Semaphore.release</tt>, and |
204 |
|
|
<tt>CountDownLatch.countDown</tt> |
205 |
brian |
1.26 |
<i>happen-before</i> actions subsequent to a successful "acquiring" method such as |
206 |
brian |
1.27 |
<tt>Lock.lock</tt>, <tt>Semaphore.acquire</tt>, <tt>Condition.await</tt>, and |
207 |
jsr166 |
1.29 |
<tt>CountDownLatch.await</tt> on the same synchronizer object. |
208 |
dl |
1.21 |
|
209 |
brian |
1.27 |
<li>For each pair of threads that successfully exchange objects via |
210 |
jsr166 |
1.30 |
an <tt>Exchanger</tt>, actions prior to the <tt>exchange()</tt> |
211 |
brian |
1.27 |
in each thread <i>happen-before</i> those subsequent to the |
212 |
|
|
corresponding <tt>exchange()</tt> in the other thread. |
213 |
brian |
1.26 |
|
214 |
|
|
<li>Actions prior to calling <tt>CyclicBarrier.await</tt> |
215 |
|
|
<i>happen-before</i> actions performed by the barrier action, and |
216 |
|
|
actions performed by the barrier action <i>happen-before</i> actions |
217 |
jsr166 |
1.29 |
subsequent to a successful return from the corresponding <tt>await</tt> |
218 |
brian |
1.27 |
in other threads. |
219 |
dl |
1.21 |
|
220 |
|
|
</ul> |
221 |
|
|
|
222 |
dl |
1.15 |
@since 1.5 |
223 |
|
|
|
224 |
dl |
1.1 |
</body> </html> |