iOS 多线程系列 -- 基础概述
iOS 多线程系列 -- pthread
iOS 多线程系列 -- NSThread
iOS 多线程系列 -- GCD全解一(基础)
iOS 多线程系列 -- GCD全解二(常用方法)
iOS 多线程系列 -- GCD全解三(进阶)
iOS 多线程系列 -- NSOperation
测试Demo的GitHub地址
3. GCD常用方法
3.1 Dispatch Group
Dispatch Group 可以让我们很方便的控制多线程中任务执行顺序。假设这样一种需求,有三个任务OA/OB/OC,我们想让OC在OAOB执行完毕在执行,有几种实现方式?方式有很多,详细看下面总结,其中一种方式我们就可以用Dispatch Group实现。
- 数据类型:
- dispatch_group_t : 组对象,调用dispatch_group_create()创建
常用方法:
- dispatch_group_async/dispatch_group_async_f ,异步方式将代码块block放入队列queue中执行,并将队列关联到调度组group。dispatch_group会等和它关联的所有的dispatch_queue_t上的任务都执行完毕才会发出同步信号,执行dispathc_group_notify的代码块block,同时dispatch_group_wati会结束等待。一个group可以关联多个任务队列,下面示例代码中可以查看。
dispatch_group_async(dispatch_group_t group,dispatch_queue_t queue,dispatch_block_t block);
dispatch_group_async(group, dispatch_get_global_queue(0, 0), ^{
for (int i = 0; i<10; i++) {
NSLog(@"---group1--%d---%@",i,[NSThread currentThread]);
}
});
- dispatch_group_notify 调度组Dispatch Group中任务执行完毕后,会发出同步信号,执行dispatch_group_notify中的block,以此来实现线程同步,实现多线程任务顺序控制。所以上面的OA/OB/OC问题可以这样:
dispatch_group_async(group, dispatch_get_global_queue(0, 0), ^{
//任务OA
});
dispatch_group_async(group, dispatch_get_global_queue(0, 0), ^{
//任务OB
});
dispatch_group_notify(group, dispatch_get_main_queue(), ^{
//任务OC
});
- dispatch_group_wait 同步函数,会阻塞当前线程直到超时或者group任务完成。返回0表示任务完全执行完毕,非0表示超时。如果想要一直等待直到任务完成,可以把第二个时间参数设置为DISPATCH_TIME_FOREVER.因为同步,所以小心死锁,尽量不要放在主线程调用此方法
dispatch_group_wait(group, dispatch_time(DISPATCH_TIME_NOW, 5 * NSEC_PER_SEC));
- dispatch_group_enter 手动指示有一个block进入group.调用此函数表示另一个块通过加入组除dispatch_group_async()之外的一种手段。调用此函数必须是与dispatch_group_leave()平衡,否则调度组会显示一个有任务,不会调用dispatch_group_notify.
- dispatch_group_leave 手动指示group中一个block完成.调用此函数表示块已完成,并通过除dispatch_group_async()之外的方法离开调度组.
- dispatch_group_enter和dispatch_group_leave的简单理解
- 一种不同于dispatch_group_async的操作调度组的方法
- 当我们调用n次dispatch_group_enter后再调用n次dispatch_group_level时,dispatch_group_notify和dispatch_group_wait会收到同步信号
- 可以简单的看做group中有一个表示任务数的属性operationCount, dispatch_group_enter会让operationCount加1, dispatch_group_level会让operationCount减1.当operationCount为0的时候,表示group总任务执行完毕,会同步通知dispatch_group_notify和dispatch_group_wait
- 测试代码
- (void)group
{
NSLog(@"---group-begin---%@",[NSThread currentThread]);
dispatch_group_t group = dispatch_group_create();
dispatch_group_async(group, dispatch_get_global_queue(0, 0), ^{// 异步添加任务到全局并发队列,并关联到调度组
for (int i = 0; i<2; i++) {
NSLog(@"---group1--%d---%@",i,[NSThread currentThread]);
}
});
dispatch_group_async(group, dispatch_get_main_queue(), ^{// 异步添加任务到主队列,并关联到调度组
for (int i = 0; i<2; i++) {
NSLog(@"---group2--%d---%@",i,[NSThread currentThread]);
}
});
NSLog(@"---group middle %@",[NSThread currentThread]);
dispatch_group_notify(group, dispatch_get_main_queue(), ^{ // group中的任务都执行完毕后,才回执行这个block
NSLog(@"---dispatch_group_notify---%@",[NSThread currentThread]);
dispatch_group_async(group, dispatch_get_global_queue(0, 0), ^{
NSLog(@"---group3-----%@",[NSThread currentThread]);
});
NSLog(@"---dispatch_group_notify middle %@",[NSThread currentThread]);
dispatch_group_async(group, dispatch_get_main_queue(), ^{
NSLog(@"---group4----%@",[NSThread currentThread]);
});
dispatch_group_notify(group, dispatch_get_main_queue(), ^{
NSLog(@"---second dispatch_group_notify---%@",[NSThread currentThread]);
});
});
NSLog(@"---before dispatch_group_wait %@",[NSThread currentThread]);
dispatch_group_wait(group, dispatch_time(DISPATCH_TIME_NOW, 5 * NSEC_PER_SEC)); // 同步函数,会阻塞当前线程,只要超时或者group任务完成,返回0表示任务完全执行完毕,非0表示超时
NSLog(@"---group-end---%@",[NSThread currentThread]);
}
- 打印结果:
2017-06-29 10:38:15.572 Test - 多线程[21587:754475] ---group-begin---<NSThread: 0x60000007d200>{number = 1, name = main}
2017-06-29 10:38:15.573 Test - 多线程[21587:754568] ---group1--0---<NSThread: 0x6000002742c0>{number = 3, name = (null)}
2017-06-29 10:38:15.573 Test - 多线程[21587:754475] ---group middle <NSThread: 0x60000007d200>{number = 1, name = main} //①
2017-06-29 10:38:15.573 Test - 多线程[21587:754568] ---group1--1---<NSThread: 0x6000002742c0>{number = 3, name = (null)} //②
2017-06-29 10:38:15.573 Test - 多线程[21587:754475] ---before dispatch_group_wait <NSThread: 0x60000007d200>{number = 1, name = main} //③
2017-06-29 10:38:20.575 Test - 多线程[21587:754475] ---group-end---<NSThread: 0x60000007d200>{number = 1, name = main} //④
2017-06-29 10:38:20.576 Test - 多线程[21587:754475] ---group2--0---<NSThread: 0x60000007d200>{number = 1, name = main}
2017-06-29 10:38:20.576 Test - 多线程[21587:754475] ---group2--1---<NSThread: 0x60000007d200>{number = 1, name = main} //⑤
2017-06-29 10:38:20.577 Test - 多线程[21587:754475] ---dispatch_group_notify---<NSThread: 0x60000007d200>{number = 1, name = main} //⑥
2017-06-29 10:38:20.577 Test - 多线程[21587:754475] ---dispatch_group_notify middle <NSThread: 0x60000007d200>{number = 1, name = main}
2017-06-29 10:38:20.577 Test - 多线程[21587:754568] ---group3-----<NSThread: 0x6000002742c0>{number = 3, name = (null)}
2017-06-29 10:38:20.578 Test - 多线程[21587:754475] ---group4----<NSThread: 0x60000007d200>{number = 1, name = main}
2017-06-29 10:38:20.578 Test - 多线程[21587:754475] ---second dispatch_group_notify---<NSThread: 0x60000007d200>{number = 1, name = main} //⑦
- 打印结果分析:
- 从①和②打印结果看, dispatch_group_async添加任务是异步的,不会阻塞当前线程
- 从③和④的打印时间差,可以看出当前线程被阻塞了5s,说明dispatch_group_wait是同步函数,会阻塞线程. 为什么阻塞当前线程的时间内没有group2输出,查看添加group2输出的代码发现,这个打印任务是异步添加到主线程的,主线程阻塞完毕以后才会继续打印group2
- ③的打印顺序还可以说明dispatch_group_notify也是异步添加任务,不会阻塞当前线程
- ⑥的打印是在group1和group2之后得出,notify中的block任务确实是在group关联的任务执行完毕后才执行
- group3/group4的打印,以及④的输出可以看出nofity可以多层嵌套
- 所有调用dispatch_group_async的地方查看一下代码会发现,调度组group可以关联到不同的队列
3.2 栅栏
dispatch_barrier_sync
- 在一个dispatch queue分派队列上提交用于同步执行的block代码块,类似dispatch_sync()但是会标记为障碍,可以控制多线程中任务执行顺序.执行顺序为:在其之前提交的任务先执行 -> dispatch_barrier_sync提交的任务 -> dispatch_barrier_sync之后提交的任务
- dispatch_barrier_sync是和并发队列相关联的,因为串行队列中的任务本身就是顺序执行的,不需要barrier技术
- 同步提交block到指定queue,
会阻塞当前线程
直到在他前面提交到此queue的任务执行完毕,然后执行barrier block,然后当前线程才能继续往下执行任务 , 从下面测试打印结果的⑤⑥⑦,打印barrier之后才会打印after dispatch_barrier_sync可以验证这一点 - 函数声明如下:queue是指定的并发队列,block是提交的代码块
void dispatch_barrier_sync(dispatch_queue_t queue,
DISPATCH_NOESCAPE dispatch_block_t block);
- 测试代码如下:
NSLog(@"barrierSync begin");
dispatch_queue_t queue = dispatch_queue_create("12312312", DISPATCH_QUEUE_CONCURRENT);
dispatch_async(queue, ^{
NSLog(@"----1-----%@", [NSThread currentThread]);
});
dispatch_async(queue, ^{
NSLog(@"----2-----%@", [NSThread currentThread]);
});
NSLog(@"before dispatch_barrier_sync");
dispatch_barrier_sync(queue, ^{
NSLog(@"----barrier-----%@", [NSThread currentThread]);
});
NSLog(@"after dispatch_barrier_sync");
dispatch_async(queue, ^{
NSLog(@"----3-----%@", [NSThread currentThread]);
});
dispatch_async(queue, ^{
NSLog(@"----4-----%@", [NSThread currentThread]);
});
NSLog(@"barrierSync end");
- 测试代码打印如下:
2017-06-29 12:28:15.355 Test - 多线程[22813:835755] barrierSync begin //①
2017-06-29 12:28:15.355 Test - 多线程[22813:835755] before dispatch_barrier_sync //②
2017-06-29 12:28:15.355 Test - 多线程[22813:835803] ----2-----<NSThread: 0x6000002692c0>{number = 4, name = (null)} //③
2017-06-29 12:28:15.355 Test - 多线程[22813:835805] ----1-----<NSThread: 0x608000260bc0>{number = 3, name = (null)} //④
2017-06-29 12:28:15.356 Test - 多线程[22813:835755] ----barrier-----<NSThread: 0x600000071880>{number = 1, name = main} //⑤
2017-06-29 12:28:15.356 Test - 多线程[22813:835755] after dispatch_barrier_sync //⑥
2017-06-29 12:28:15.356 Test - 多线程[22813:835755] barrierSync end //⑦
2017-06-29 12:28:15.356 Test - 多线程[22813:835805] ----3-----<NSThread: 0x608000260bc0>{number = 3, name = (null)} //⑧
2017-06-29 12:28:15.356 Test - 多线程[22813:835803] ----4-----<NSThread: 0x6000002692c0>{number = 4, name = (null)} //⑨
dispatch_barrier_async,基本和dispatch_barrier_sync类似的功能,区别在于:
- dispatch_barrier_async是提交异步执行的block代码块,不会阻塞当前线程,从下面测试打印结果的⑤⑥⑦after dispatch_barrier_sync在barrier之前打印可以验证这一点
- 测试代码:
NSLog(@"barrierAsync begin");
dispatch_queue_t queue = dispatch_queue_create("12312312", DISPATCH_QUEUE_CONCURRENT);
dispatch_async(queue, ^{
NSLog(@"----1-----%@", [NSThread currentThread]);
});
dispatch_async(queue, ^{
NSLog(@"----2-----%@", [NSThread currentThread]);
});
NSLog(@"before dispatch_barrier_async");
dispatch_barrier_async(queue, ^{
NSLog(@"----barrier-----%@", [NSThread currentThread]);
});
NSLog(@"after dispatch_barrier_async");
dispatch_async(queue, ^{
NSLog(@"----3-----%@", [NSThread currentThread]);
});
dispatch_async(queue, ^{
NSLog(@"----4-----%@", [NSThread currentThread]);
});
NSLog(@"barrierAsync end");
- 测试结果打印:
2017-06-29 12:29:32.048 Test - 多线程[22813:835755] barrierAsync begin //①
2017-06-29 12:29:32.049 Test - 多线程[22813:835755] before dispatch_barrier_async //②
2017-06-29 12:29:32.049 Test - 多线程[22813:835815] ----1-----<NSThread: 0x600000268e00>{number = 5, name = (null)} //③
2017-06-29 12:29:32.049 Test - 多线程[22813:837436] ----2-----<NSThread: 0x600000267540>{number = 6, name = (null)} //④
2017-06-29 12:29:32.049 Test - 多线程[22813:835755] after dispatch_barrier_async //⑤
2017-06-29 12:29:32.049 Test - 多线程[22813:837436] ----barrier-----<NSThread: 0x600000267540>{number = 6, name = (null)} //⑥
2017-06-29 12:29:32.049 Test - 多线程[22813:835755] barrierAsync end //⑦
2017-06-29 12:29:32.050 Test - 多线程[22813:837436] ----3-----<NSThread: 0x600000267540>{number = 6, name = (null)} //⑧
2017-06-29 12:29:32.050 Test - 多线程[22813:835815] ----4-----<NSThread: 0x600000268e00>{number = 5, name = (null)} //⑨
总结:
- 上面已经得出dispatch_barrier_async和dispatch_barrier_sync的重要区别是会不会阻塞当前线程,所以按需选择使用哪种barrier技术
- 如果在主线程需要用到,推荐dispatch_barrier_async因为不会阻塞线程,如果在barrier之前提交的任务中有耗时任务,也不会带来主线程UI的卡顿
3.3 迭代dispatch_apply
dispatch_apply : 提交block任务块到调度队列进行多次调度.
三个参数解析:
- iterations表示调度总次数;
- queue任务提交的队列,如果是并发队列,那么调度可以并发执行,提高调用次数
- (^block)(size_t),带有一个size_t参数的block块, size_t类型的参数是当前迭代索引 ; block是任务块
void dispatch_apply(size_t iterations, dispatch_queue_t queue, DISPATCH_NOESCAPE void (^block)(size_t));
NOTE:
- dispatch_apply是同步函数,会阻塞当前线程,直到迭代完毕,由③可以验证.
- dispatch_apply可能会提高迭代速度.你可以指定串行或并发 queue,并发queue允许同时执行多个循环迭代,而串行queue就没太大必要使用了
- 如果当前队列是串行队列,而且在当前串行队列中使用dispatch_apply指定当前队列为迭代队列,会死锁
- 并发队列中,迭代次序是不定的,有测试打印结果①②可以验证
测试代码:
NSLog(@"apply begin");
dispatch_apply(10, dispatch_get_global_queue(0, 0), ^(size_t index) {
NSLog(@"index = %zd , thread = %@",index,[NSThread currentThread]);
});
NSLog(@"apply end");
测试打印结果:
2017-06-29 14:46:08.973 Test - 多线程[24285:925046] apply begin
2017-06-29 14:46:08.973 Test - 多线程[24285:925046] index = 0 , thread = <NSThread: 0x608000079c00>{number = 1, name = main}
2017-06-29 14:46:08.973 Test - 多线程[24285:925297] index = 1 , thread = <NSThread: 0x608000268600>{number = 7, name = (null)} //①
2017-06-29 14:46:08.973 Test - 多线程[24285:925094] index = 3 , thread = <NSThread: 0x60000026cb40>{number = 5, name = (null)} //②
2017-06-29 14:46:08.973 Test - 多线程[24285:925296] index = 2 , thread = <NSThread: 0x60800026ac80>{number = 6, name = (null)}
2017-06-29 14:46:08.974 Test - 多线程[24285:925046] apply end //③
3.4 一次执行dispatch_once
dispatch_once的作用:保证其block代码块中的代码只执行一次,是线程安全的,常用于实现单例等
数据类型
- dispatch_once_t , 其实就是long类型
typedef long dispatch_once_t;
方法,我们在使用的时候,是一个dispatch_once的宏,对应的是一个_dispatch_once的函数._dispatch_once函数两个参数,第一个predicate是标志位,第二个是代码块。
线程安全简单理解:
- 第一次执行,block需要被调用,调用结束后需要置标记变量
- 非第一次执行,而此时第一次执行尚未完成,线程需要等待第一次执行完成后才能继续往下执行
- 非第一次执行,而此时#1已经完成,线程直接跳过block而进行后续任务
- 更详细的深入探究可以看这里一步步分析dispatch_once的低负载特性
void
_dispatch_once(dispatch_once_t *predicate,
DISPATCH_NOESCAPE dispatch_block_t block)
{
if (DISPATCH_EXPECT(*predicate, ~0l) != ~0l) {
dispatch_once(predicate, block);
} else {
dispatch_compiler_barrier();
}
DISPATCH_COMPILER_CAN_ASSUME(*predicate == ~0l);
}
测试代码
dispatch_apply(4, dispatch_get_global_queue(0, 0), ^(size_t index) {
NSLog(@"------index = %zd",index);
static dispatch_once_t onceToken; //①
dispatch_once(&onceToken, ^{//②
NSLog(@"------run");
});
});
测试打印结果
2017-06-29 15:11:57.542 Test - 多线程[24619:946139] ------index = 1
2017-06-29 15:11:57.542 Test - 多线程[24619:946049] ------index = 0
2017-06-29 15:11:57.542 Test - 多线程[24619:946136] ------index = 2
2017-06-29 15:11:57.542 Test - 多线程[24619:946139] ------run
3.5 延迟 dispatch_after
dispatch_after(dispatch_time_t when,
dispatch_queue_t queue,
dispatch_block_t block);
- dispatch_after,延迟指定时间when后,提交block任务到队列queue。
注意:并不是延时执行任务
- 测试代码:
dispatch_after(dispatch_time(DISPATCH_TIME_NOW, (int64_t)(2.0 * NSEC_PER_SEC)), dispatch_get_main_queue(), ^{
NSLog(@"run-----"); // 延迟2秒后提交任务
});
3.6 线程状态操作: 挂起和恢复队列
- 当需要挂起队列时,使用dispatch_suspend方法;
- 恢复队列时,使用dispatch_resume方法
注意点:
- dispatch_suspend 并不会立即挂起队列,根据API文档可以查看到。调用dispatch_suspend函数后,当前正在执行的block任务会继续执行下面的测试打印结果①②可以验证;只是其后添加的任务会被挂起,直到调用dispatch_resume恢复队列,从下面的测试打印结果③④⑤顺序可以验证
- dispatch_suspend 可以挂起串行、并发队列,但是不能挂起全局并发队列,下面测试代码中打开test1可以自行验证
- dispatch_suspend 和dispatch_resume必须成对调用。可以简单理解为:dispatch_suspend会让队列的某个计数器加1,dispatch_resume会让这个计数器减1 , 只有当这个计数器为0的时候才回恢复队列
- 以上注意点,可以从下面测试打印结果中验证。
测试代码:
- (void)testsuplend
{
//test1
// dispatch_queue_t queue = dispatch_get_global_queue(0, 0); // dispatch_suspend对全局并发队列无效
//test2
// dispatch_queue_t queue = dispatch_queue_create(NULL, 0); // dispatch_suspend 对串行队列有效
//test3
dispatch_queue_t queue = dispatch_queue_create(NULL, DISPATCH_QUEUE_CONCURRENT); // dispatch_suspend 对自己创建的并发队列有效
dispatch_async(queue, ^{
for (int i = 0 ; i < 10; i ++) {
NSLog(@"---suplend1 -- %zd thread = %@",i,[NSThread currentThread]);
sleep(1);
}
});
dispatch_after(dispatch_time(DISPATCH_TIME_NOW, (int64_t)(2.1 * NSEC_PER_SEC)), dispatch_get_main_queue(), ^{
NSLog(@"dispatch_suspend queue one , suplend1打印任务执行完毕后会挂起队列,直到dispatch_resume调用后恢复队列");
dispatch_suspend(queue);
});
// 4s后继续添加新的任务,如果在添加之前点击挂起,此任务不会执行
dispatch_after(dispatch_time(DISPATCH_TIME_NOW, (int64_t)(4 * NSEC_PER_SEC)), dispatch_get_main_queue(), ^{
dispatch_async(queue, ^{
for (int i = 0 ; i < 10; i ++) {
sleep(2);
NSLog(@"---suplend2 -- %zd thread = %@",i,[NSThread currentThread]);
}
});
});
dispatch_after(dispatch_time(DISPATCH_TIME_NOW, (int64_t)(6 * NSEC_PER_SEC)), dispatch_get_main_queue(), ^{
NSLog(@"dispatch_resume queue one ,---suplend2 --打印任务会恢复执行");
dispatch_resume(queue);
});
}
测试打印结果:
2017-06-29 23:31:56.678 Test - 多线程[57328:8371175] ---suplend1 -- 0 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:31:57.680 Test - 多线程[57328:8371175] ---suplend1 -- 1 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:31:58.682 Test - 多线程[57328:8371175] ---suplend1 -- 2 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:31:58.777 Test - 多线程[57328:8371126] dispatch_suspend queue one , suplend1打印任务执行完毕后会挂起队列,直到dispatch_resume调用后恢复队列 //①
2017-06-29 23:31:59.687 Test - 多线程[57328:8371175] ---suplend1 -- 3 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}//②
2017-06-29 23:32:00.689 Test - 多线程[57328:8371175] ---suplend1 -- 4 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:32:01.690 Test - 多线程[57328:8371175] ---suplend1 -- 5 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:32:02.677 Test - 多线程[57328:8371126] dispatch_resume queue one ,---suplend2 --打印任务会恢复执行//③
2017-06-29 23:32:02.692 Test - 多线程[57328:8371175] ---suplend1 -- 6 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:32:03.695 Test - 多线程[57328:8371175] ---suplend1 -- 7 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}//④
2017-06-29 23:32:04.682 Test - 多线程[57328:8371177] ---suplend2 -- 0 thread = <NSThread: 0x60000027b7c0>{number = 4, name = (null)}//⑤
2017-06-29 23:32:04.699 Test - 多线程[57328:8371175] ---suplend1 -- 8 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}
2017-06-29 23:32:05.702 Test - 多线程[57328:8371175] ---suplend1 -- 9 thread = <NSThread: 0x60000027b740>{number = 3, name = (null)}