[20231204]快速删除大量文件测试(rsync版本)4.txt

  • [20231204]快速删除大量文件测试(rsync版本)4.txt已关闭评论
  • 53 次浏览
  • A+
所属分类:linux技术

[20231204]快速删除大量文件测试(rsync版本)4.txt

--//前个星期做了大量文件测试,发现只要顺着inode节点顺序删除文件效率最高.
--//还有一种方式使用rsync,利用建立空目录使用同步功能,删除对应目录里面的全部文件.
--//我记忆里以前测试效率不错,今天做具体探究.

1.建立测试环境:
$ mount | column | grep u01
/dev/cciss/c0d0p6 on /u01 type ext3 (rw)
--// /u01的文件系统是ext3.

--//按照顺序建立文件。
$ cd /u01/testrm
$ time for i in $(seq -f "%06g" 300000); do >| $i.aud; done

$ time perl -e 'unlink for ( <*.aud> )'
real    0m4.077s
user    0m0.445s
sys     0m2.961s

2.rsync测试:
$ mkdir /u01/blanktest
--//建立一个空目录.
$ cd /u01/testrm
$ time for i in $(seq -f "%06g" 300000); do >| $i.aud; done
real    0m7.174s
user    0m2.179s
sys     0m4.407s

$ time rsync -a --delete /u01/blanktest /u01/testrm
real    0m0.048s
user    0m0.003s
sys     0m0.003s
--//奇怪不可能这样快,发现/u01/blanktest漏写一个/.

$ time rsync -a --delete /u01/blanktest/ /u01/testrm
real    0m4.889s
user    0m0.360s
sys     0m3.831s
--//确实非常快,与perl的测试基本接近。

3.使用strace跟踪看看.

$ time for i in $(seq -f "%06g" 300000); do >| $i.aud; done
real    0m6.508s
user    0m2.272s
sys     0m3.614s

$ ls -f | head
.
..
026879.aud
078531.aud
084876.aud
034744.aud
280025.aud
063529.aud
137869.aud
163787.aud
--//显示顺序与前面的测试一样!!

$ strace -T -tt -o /tmp/yy4.txt rsync -a --delete /u01/blanktest/ /u01/testrm
--//你根本看不到unlink操作。

--//跟踪内容如下:
....
16:01:20.455672 fcntl(3, F_SETFD, FD_CLOEXEC) = 0 <0.000015>
16:01:20.455739 getdents(3, /* 3 entries */, 32768) = 80 <0.000026>
16:01:20.455824 lstat("000001.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000018>
16:01:20.455924 mmap(NULL, 266240, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6ebf790000 <0.000017>
16:01:20.456008 getdents(3, /* 0 entries */, 32768) = 0 <0.000016>
16:01:20.456070 close(3)                = 0 <0.000018>
16:01:20.456144 mmap(NULL, 266240, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6ebf74f000 <0.000016>
16:01:20.456216 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999997}) <0.000018>
16:01:20.456306 write(4, "=75f1.20e&*p355A2013666oracle20136610o"..., 65) = 65 <0.000617>
16:01:20.457033 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {54, 864503}) <5.135527>
16:01:25.592682 read(5, "37", 4)  = 4 <0.000020>
16:01:25.622656 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999997}) <0.000015>
16:01:25.622717 read(5, "110", 3)   = 3 <0.000016>
16:01:25.622776 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999997}) <0.000013>
16:01:25.622824 write(4, "37110", 7) = 7 <0.000016>
16:01:25.622870 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999996}) <0.000026>
16:01:25.622978 read(5, "237", 4) = 4 <0.000024>
16:01:25.623065 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999997}) <0.000025>
16:01:25.623159 read(5, "210200", 19) = 19 <0.000025>
16:01:25.623260 open("000001.aud", O_RDONLY) = 3 <0.000049>
16:01:25.623354 fstat(3, {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000011>
16:01:25.623470 close(3)                = 0 <0.000013>
16:01:25.623512 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999995}) <0.000037>
16:01:25.623604 write(4, "'721020032435214331217"..., 43) = 43 <0.000027>
16:01:25.623704 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999753}) <0.000262>
16:01:25.624019 read(5, "17", 4)  = 4 <0.000035>
16:01:25.624108 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999997}) <0.000012>
16:01:25.624167 read(5, "", 1)        = 1 <0.000023>
16:01:25.624257 munmap(0x7f6ebf790000, 266240) = 0 <0.000023>
16:01:25.624323 munmap(0x7f6ebf7d1000, 266240) = 0 <0.000026>
16:01:25.624415 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999996}) <0.000024>
16:01:25.624509 write(4, "17", 5) = 5 <0.000029>
16:01:25.624597 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999931}) <0.000083>
16:01:25.624731 read(5, "27", 4)  = 4 <0.000011>
16:01:25.624772 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999998}) <0.000025>
16:01:25.624850 read(5, "", 2)      = 2 <0.000011>
16:01:25.624909 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999999}) <0.000010>
16:01:25.624970 write(4, "17", 5) = 5 <0.000012>
16:01:25.625019 select(5, NULL, [4], [4], {60, 0}) = 1 (out [4], left {59, 999998}) <0.000010>
16:01:25.625060 write(4, "17", 5) = 5 <0.000013>
16:01:25.625106 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999999}) <0.000013>
16:01:25.625194 read(5, "17", 4)  = 4 <0.000023>
16:01:25.625279 select(6, [5], [], NULL, {60, 0}) = 1 (in [5], left {59, 999998}) <0.000012>
16:01:25.625378 read(5, "", 1)        = 1 <0.000032>
16:01:25.625462 wait4(4794, 0x7fff4321847c, WNOHANG, NULL) = 0 <0.000022>
16:01:25.625543 select(0, NULL, NULL, NULL, {0, 20000}) = 0 (Timeout) <0.020087>
16:01:25.645697 wait4(4794, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], WNOHANG, NULL) = 4794 <0.000027>
16:01:25.645787 --- SIGCHLD (Child exited) @ 0 (0) ---
16:01:25.645833 wait4(-1, 0x7fff43217e6c, WNOHANG, NULL) = -1 ECHILD (No child processes) <0.000018>
16:01:25.645900 rt_sigreturn(0xffffffffffffffff) = 4794 <0.000017>
16:01:25.645958 rt_sigaction(SIGUSR1, {0x1, [], SA_RESTORER, 0x3799030330}, NULL, 8) = 0 <0.000022>
16:01:25.646064 rt_sigaction(SIGUSR2, {0x1, [], SA_RESTORER, 0x3799030330}, NULL, 8) = 0 <0.000021>
16:01:25.646149 wait4(4794, 0x7fff43218474, WNOHANG, NULL) = -1 ECHILD (No child processes) <0.000020>
16:01:25.646267 exit_group(0)           = ?
--//我不知道句柄5,6表示怎么。发现rync实际上建立子进程来执行删除操作。重新测试,strace加入-f参数。

--//补充文件大小不为0看看。
$ time for i in $(seq -f "%06g" 300000); do echo test $i >| $i.aud; done
real    0m44.288s
user    0m3.335s
sys     0m7.946s

$ time  rsync -a --delete /u01/blanktest/ /u01/testrm
real    0m31.823s
user    0m0.351s
sys     0m4.612s

--//继续strace的测试:
$ time for i in $(seq -f "%06g" 300000); do >| $i.aud; done
real    0m6.989s
user    0m2.238s
sys     0m4.112s

$ strace -f -T -tt -o /tmp/yy5.txt rsync -a --delete /u01/blanktest/ /u01/testrm

$ awk '{print $1}' /tmp/yy5.txt | sort | uniq -c
    325 5158
 600728 5159
     72 5160

--//分成3个进程,主要有pid=5159完成。
--//查看跟踪文件:
5159  16:15:08.681724 getdents(0, /* 1024 entries */, 32768) = 32752 <0.001046>
--//开始读取目录
5159  16:15:08.686451 lstat("026879.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000024>
5159  16:15:08.686571 mmap(NULL, 135168, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f413480e000 <0.000021>
5159  16:15:08.686664 mmap(NULL, 266240, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f41347cd000 <0.000017>
5159  16:15:08.686738 lstat("078531.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000020>
5159  16:15:08.686830 lstat("084876.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000020>
5159  16:15:08.686921 lstat("034744.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.687009 lstat("280025.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000020>
5159  16:15:08.687099 lstat("063529.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.687188 lstat("137869.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.687277 lstat("163787.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.687378 lstat("111415.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000021>
5159  16:15:08.687468 lstat("177021.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000020>
--//与前面perl执行看到非常相似。
--//你可以发现rsync的特点是getdents之后跟着lstat,注意看lstat里面的文件与前面ls -f顺序一致.
...
5159  16:15:08.778996 getdents(0, /* 1024 entries */, 32768) = 32768 <0.000950>
5159  16:15:08.783589 lstat("121204.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000020>
5159  16:15:08.783689 lstat("006462.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000018>
5159  16:15:08.783782 lstat("235760.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.783871 lstat("187943.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.783959 lstat("235116.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000018>
5159  16:15:08.784047 lstat("077735.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000018>
5159  16:15:08.784134 lstat("126397.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.784223 lstat("098220.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.784319 lstat("209702.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
5159  16:15:08.784410 lstat("277317.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000018>
5159  16:15:08.784497 lstat("269846.aud", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0 <0.000019>
...
--//接着getdents之后跟着lstat,如此反复.
5159  16:15:37.270786 getdents(0, /* 0 entries */, 32768) = 0 <0.000017>
5159  16:15:37.270854 close(0)          = 0 <0.000030>
5159  16:15:37.270938 mmap(NULL, 1200128, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f41331fe000 <0.000019>
5159  16:15:37.537610 munmap(0x7f41331fe000, 1200128) = 0 <0.000140>
5159  16:15:37.552144 unlink("300000.aud") = 0 <0.000098>
5159  16:15:37.552364 unlink("299999.aud") = 0 <0.000060>
5159  16:15:37.552466 unlink("299998.aud") = 0 <0.000023>
5159  16:15:37.552512 unlink("299997.aud") = 0 <0.000051>
5159  16:15:37.552608 unlink("299996.aud") = 0 <0.000038>
5159  16:15:37.552685 unlink("299995.aud") = 0 <0.000051>
5159  16:15:37.552769 unlink("299994.aud") = 0 <0.000035>
5159  16:15:37.552843 unlink("299993.aud") = 0 <0.000054>
5159  16:15:37.552930 unlink("299992.aud") = 0 <0.000020>
--//最后集中做unlink操作,注意看unlink的文件按照文件名反序排序删除的,这样删除文件最快,这也是prsync与perl一样,测试删除
--//文件快的原因。
--//只要按照文件名顺序建立或者反序建立做删除操作,perl或者rsync删除文件都是最快的。
--//补充实际上按照inode的顺序删除最快(反向类似)。

4.乱序建立文件呢?

--//按照乱序建立文件,我不知道bash shell如何实现,我先通过oracle的sql语句建立乱序数字输出。
set pagesize 0
spool aa.txt
select lpad(level,6,'0') from dual connect by level <=3e5 order by dbms_random.value;
spool off
--//注意设置set pagesize 0,不然中间出现多次页头。注意保持aa.txt的备份,测试要重复多次。

$ time for i in $(cat /tmp/aa.txt ); do >| $i.aud; done
real    0m6.880s
user    0m1.986s
sys     0m4.282s

$ ls -f|head
.
..
026879.aud
078531.aud
084876.aud
034744.aud
280025.aud
063529.aud
137869.aud
163787.aud
--//显示顺序与前面的测试一样!!

$ time  rsync -a --delete /u01/blanktest/ /u01/testrm
real    1m53.208s
user    0m0.459s
sys     0m7.391s

--//差不多2分钟,使用strace跟踪可以看到前面unlink的顺序,可以推断,unlink按照文件名从大到小的顺序删除,这样inode的顺序是打乱的,rsync也不会很快
--//删除全部文件,再次验证我前面的判断:按照inode的顺序(不管正序还是反序),做删除操作效率最高。

$ strace -f -T -tt -o /tmp/yy6.txt rsync -a --delete /u01/blanktest/ /u01/testrm

--//跟踪文件内容如下:
9769  08:41:59.387415 getdents(0, /* 0 entries */, 32768) = 0 <0.000025>
9769  08:41:59.387503 close(0)          = 0 <0.000036>
9769  08:41:59.387608 mmap(NULL, 1200128, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fa39724d000 <0.000024>
9769  08:41:59.628093 munmap(0x7fa39724d000, 1200128) = 0 <0.000174>
9769  08:41:59.643044 unlink("300000.aud") = 0 <0.000086>
9769  08:41:59.643229 unlink("299999.aud") = 0 <0.000084>
9769  08:41:59.643361 unlink("299998.aud") = 0 <0.000085>
9769  08:41:59.643476 unlink("299997.aud") = 0 <0.000066>
9769  08:41:59.643588 unlink("299996.aud") = 0 <0.000029>
...

$ time for i in $(cat /tmp/aa.txt ); do >| $i.aud; done
real    0m6.881s
user    0m1.965s
sys     0m4.312s

$ time sed "s/$/.aud/" /tmp/aa.txt |  xargs rm -f
real    0m6.191s
user    0m0.552s
sys     0m5.403s

$ time for i in $(cat /tmp/aa.txt ); do >| $i.aud; done
real    0m6.889s
user    0m1.990s
sys     0m4.280s

$  time sed "s/$/.aud/" /tmp/aa.txt|tac | xargs rm -f
real    0m7.161s
user    0m0.459s
sys     0m5.969s
--//反序始终慢一点,也许执行tac的原因。
--//按照inode的顺序(不管正序还是反序),做删除操作效率最高。

5.总结:
--//可以得出结论,要快速删除某个目录的大量文件,最快的方式是按照inode的顺序执行删除操作,这样最快.