5亿整数的大文件,怎么排?
本人免费整理了Java高级资料,涵盖了Java、Redis、MongoDB、MySQL、Zookeeper、Spring Cloud、Dubbo高并发分布式等教程,一共30G,需要自己领取。
传送门:https://mp.weixin.qq.com/s/osB-BOl6W-ZLTSttTkqMPQ
问题
给你1个文件bigdata,大小4663M,5亿个数,文件中的数据随机,如下一行一个整数:
<pre class="" style="margin: 0px; padding: 8px 0px 6px; max-width: 100%; box-sizing: border-box !important; word-wrap: break-word !important; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: 0.544px; orphans: 2; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background: rgb(241, 239, 238); border-radius: 0px; overflow-y: auto; color: rgb(80, 97, 109); text-align: start; font-size: 10px; line-height: 12px; font-family: consolas, menlo, courier, monospace, "Microsoft Yahei" !important; border-width: 1px !important; border-style: solid !important; border-color: rgb(226, 226, 226) !important;">
-
6196302
-
3557681
-
6121580
-
2039345
-
2095006
-
1746773
-
7934312
-
2016371
-
7123302
-
8790171
-
2966901
-
...
-
7005375
</pre>
现在要对这个文件进行排序,怎么搞?
内部排序
先尝试内排,选2种排序方式:
<pre class="" style="margin: 0px; padding: 8px 0px 6px; max-width: 100%; box-sizing: border-box !important; word-wrap: break-word !important; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: 0.544px; orphans: 2; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background: rgb(241, 239, 238); border-radius: 0px; overflow-y: auto; color: rgb(80, 97, 109); text-align: start; font-size: 10px; line-height: 12px; font-family: consolas, menlo, courier, monospace, "Microsoft Yahei" !important; border-width: 1px !important; border-style: solid !important; border-color: rgb(226, 226, 226) !important;">
1. `private final int cutoff = 8;`
3. `public <T> void perform(Comparable<T>[] a) {`
4. `perform(a,0,a.length - 1);`
5. `}`
7. `private <T> int median3(Comparable<T>[] a,int x,int y,int z) {`
8. `if(lessThan(a[x],a[y])) {`
9. `if(lessThan(a[y],a[z])) {`
10. `return y;`
11. `}`
12. `else if(lessThan(a[x],a[z])) {`
13. `return z;`
14. `}else {`
15. `return x;`
16. `}`
17. `}else {`
18. `if(lessThan(a[z],a[y])){`
19. `return y;`
20. `}else if(lessThan(a[z],a[x])) {`
21. `return z;`
22. `}else {`
23. `return x;`
24. `}`
25. `}`
26. `}`
28. `private <T> void perform(Comparable<T>[] a,int low,int high) {`
29. `int n = high - low + 1;`
30. `//当序列非常小,用插入排序`
31. `if(n <= cutoff) {`
32. `InsertionSort insertionSort = SortFactory.createInsertionSort();`
33. `insertionSort.perform(a,low,high);`
34. `//当序列中小时,使用median3`
35. `}else if(n <= 100) {`
36. `int m = median3(a,low,low + (n >>> 1),high);`
37. `exchange(a,m,low);`
38. `//当序列比较大时,使用ninther`
39. `}else {`
40. `int gap = n >>> 3;`
41. `int m = low + (n >>> 1);`
42. `int m1 = median3(a,low,low + gap,low + (gap << 1));`
43. `int m2 = median3(a,m - gap,m,m + gap);`
44. `int m3 = median3(a,high - (gap << 1),high - gap,high);`
45. `int ninther = median3(a,m1,m2,m3);`
46. `exchange(a,ninther,low);`
47. `}`
49. `if(high <= low)`
50. `return;`
51. `//lessThan`
52. `int lt = low;`
53. `//greaterThan`
54. `int gt = high;`
55. `//中心点`
56. `Comparable<T> pivot = a[low];`
57. `int i = low + 1;`
59. `/*`
60. `* 不变式:`
61. `* a[low..lt-1] 小于pivot -> 前部(first)`
62. `* a[lt..i-1] 等于 pivot -> 中部(middle)`
63. `* a[gt+1..n-1] 大于 pivot -> 后部(final)`
64. `*`
65. `* a[i..gt] 待考察区域`
66. `*/`
68. `while (i <= gt) {`
69. `if(lessThan(a[i],pivot)) {`
70. `//i-> ,lt ->`
71. `exchange(a,lt++,i++);`
72. `}else if(lessThan(pivot,a[i])) {`
73. `exchange(a,i,gt--);`
74. `}else{`
75. `i++;`
76. `}`
77. `}`
79. `// a[low..lt-1] < v = a[lt..gt] < a[gt+1..high].`
80. `perform(a,low,lt - 1);`
81. `perform(a,gt + 1,high);`
82. `}`
</pre>
归并排序:
<pre class="" style="margin: 0px; padding: 8px 0px 6px; max-width: 100%; box-sizing: border-box !important; word-wrap: break-word !important; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: 0.544px; orphans: 2; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background: rgb(241, 239, 238); border-radius: 0px; overflow-y: auto; color: rgb(80, 97, 109); text-align: start; font-size: 10px; line-height: 12px; font-family: consolas, menlo, courier, monospace, "Microsoft Yahei" !important; border-width: 1px !important; border-style: solid !important; border-color: rgb(226, 226, 226) !important;">
1. `/**`
2. `* 小于等于这个值的时候,交给插入排序`
3. `*/`
4. `private final int cutoff = 8;`
6. `/**`
7. `* 对给定的元素序列进行排序`
8. `*`
9. `* @param a 给定元素序列`
10. `*/`
11. `@Override`
12. `public <T> void perform(Comparable<T>[] a) {`
13. `Comparable<T>[] b = a.clone();`
14. `perform(b, a, 0, a.length - 1);`
15. `}`
17. `private <T> void perform(Comparable<T>[] src,Comparable<T>[] dest,int low,int high) {`
18. `if(low >= high)`
19. `return;`
21. `//小于等于cutoff的时候,交给插入排序`
22. `if(high - low <= cutoff) {`
23. `SortFactory.createInsertionSort().perform(dest,low,high);`
24. `return;`
25. `}`
27. `int mid = low + ((high - low) >>> 1);`
28. `perform(dest,src,low,mid);`
29. `perform(dest,src,mid + 1,high);`
31. `//考虑局部有序 src[mid] <= src[mid+1]`
32. `if(lessThanOrEqual(src[mid],src[mid+1])) {`
33. `System.arraycopy(src,low,dest,low,high - low + 1);`
34. `}`
36. `//src[low .. mid] + src[mid+1 .. high] -> dest[low .. high]`
37. `merge(src,dest,low,mid,high);`
38. `}`
40. `private <T> void merge(Comparable<T>[] src,Comparable<T>[] dest,int low,int mid,int high) {`
42. `for(int i = low,v = low,w = mid + 1; i <= high; i++) {`
43. `if(w > high || v <= mid && lessThanOrEqual(src[v],src[w])) {`
44. `dest[i] = src[v++];`
45. `}else {`
46. `dest[i] = src[w++];`
47. `}`
48. `}`
49. `}`
</pre>
数据太多,递归太深 ->栈溢出?加大Xss?数据太多,数组太长 -> OOM?加大Xmx?
耐心不足,没跑出来.而且要将这么大的文件读入内存,在堆中维护这么大个数据量,还有内排中不断的拷贝,对栈和堆都是很大的压力,不具备通用性。
sort命令来跑
跑了多久呢?24分钟.
为什么这么慢?
粗略的看下我们的资源:
内存 jvm-heap/stack,native-heap/stack,page-cache,block-buffer 外存 swap + 磁盘 数据量很大,函数调用很多,系统调用很多,内核/用户缓冲区拷贝很多,脏页回写很多,io-wait很高,io很繁忙,堆栈数据不断交换至swap,线程切换很多,每个环节的锁也很多. 总之,内存吃紧,问磁盘要空间,脏数据持久化过多导致cache频繁失效,引发大量回写,回写线程高,导致cpu大量时间用于上下文切换,一切,都很糟糕,所以24分钟不细看了,无法忍受.
位图法
<pre class="" style="margin: 0px; padding: 8px 0px 6px; max-width: 100%; box-sizing: border-box !important; word-wrap: break-word !important; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: 0.544px; orphans: 2; text-indent: 0px; text-transform: none; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background: rgb(241, 239, 238); border-radius: 0px; overflow-y: auto; color: rgb(80, 97, 109); text-align: start; font-size: 10px; line-height: 12px; font-family: consolas, menlo, courier, monospace, "Microsoft Yahei" !important; border-width: 1px !important; border-style: solid !important; border-color: rgb(226, 226, 226) !important;">
1. `private BitSet bits;`
3. `public void perform(`
4. `String largeFileName,`
5. `int total,`
6. `String destLargeFileName,`
7. `Castor<Integer> castor,`
8. `int readerBufferSize,`
9. `int writerBufferSize,`
10. `boolean asc) throws IOException {`
12. `System.out.println("BitmapSort Started.");`
13. `long start = System.currentTimeMillis();`
14. `bits = new BitSet(total);`
15. `InputPart<Integer> largeIn = PartFactory.createCharBufferedInputPart(largeFileName, readerBufferSize);`
16. `OutputPart<Integer> largeOut = PartFactory.createCharBufferedOutputPart(destLargeFileName, writerBufferSize);`
17. `largeOut.delete();`
19. `Integer data;`
20. `int off = 0;`
21. `try {`
22. `while (true) {`
23. `data = largeIn.read();`
24. `if (data == null)`
25. `break;`
26. `int v = data;`
27. `set(v);`
28. `off++;`
29. `}`
30. `largeIn.close();`
31. `int size = bits.size();`
32. `System.out.println(String.format("lines : %d ,bits : %d", off, size));`
34. `if(asc) {`
35. `for (int i = 0; i < size; i++) {`
36. `if (get(i)) {`
37. `largeOut.write(i);`
38. `}`
39. `}`
40. `}else {`
41. `for (int i = size - 1; i >= 0; i--) {`
42. `if (get(i)) {`
43. `largeOut.write(i);`
44. `}`
45. `}`
46. `}`
48. `largeOut.close();`
49. `long stop = System.currentTimeMillis();`
50. `long elapsed = stop - start;`
51. `System.out.println(String.format("BitmapSort Completed.elapsed : %dms",elapsed));`
52. `}finally {`
53. `largeIn.close();`
54. `largeOut.close();`
55. `}`
56. `}`
58. `private void set(int i) {`
59. `bits.set(i);`
60. `}`
62. `private boolean get(int v) {`
63. `return bits.get(v);`
64. `}`
</pre>
nice!跑了190秒,3分来钟. 以核心内存4663M/32大小的空间跑出这么个结果,而且大量时间在用于I/O,不错.
问题是,如果这个时候突然内存条坏了1、2根,或者只有极少的内存空间怎么搞?
外部排序
该外部排序上场了. 外部排序干嘛的?
内存极少的情况下,利用分治策略,利用外存保存中间结果,再用多路归并来排序;
map-reduce的嫡系.
![image](https://img.haomeiwen.com/i20407517/26f3397ebc9dac7a?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
![image](https://img.haomeiwen.com/i20407517/85e4552f639e93e5?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
1.分
内存中维护一个极小的核心缓冲区memBuffer,将大文件bigdata按行读入,搜集到memBuffer满或者大文件读完时,对memBuffer中的数据调用内排进行排序,排序后将有序结果写入磁盘文件bigdata.xxx.part.sorted. 循环利用memBuffer直到大文件处理完毕,得到n个有序的磁盘文件:
![image](https://img.haomeiwen.com/i20407517/60aad7a5ddbebd7f?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
2.合
现在有了n个有序的小文件,怎么合并成1个有序的大文件?把所有小文件读入内存,然后内排?(⊙o⊙)… no!
利用如下原理进行归并排序:
![image](https://img.haomeiwen.com/i20407517/e284034e0a324086?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
我们举个简单的例子:
文件1:3,6,9 文件2:2,4,8 文件3:1,5,7
第一回合:文件1的最小值:3 , 排在文件1的第1行 文件2的最小值:2,排在文件2的第1行 文件3的最小值:1,排在文件3的第1行 那么,这3个文件中的最小值是:min(1,2,3) = 1 也就是说,最终大文件的当前最小值,是文件1、2、3的当前最小值的最小值,绕么?上面拿出了最小值1,写入大文件.
第二回合:文件1的最小值:3 , 排在文件1的第1行 文件2的最小值:2,排在文件2的第1行 文件3的最小值:5,排在文件3的第2行 那么,这3个文件中的最小值是:min(5,2,3) = 2 将2写入大文件.
也就是说,最小值属于哪个文件,那么就从哪个文件当中取下一行数据.(因为小文件内部有序,下一行数据代表了它当前的最小值)
最终的时间,跑了771秒,13分钟左右.