Slow rename()

gistfile1.txt

SUMMARY

If you generate a lot of data and then immediately rename a file on top of an
existing file (even if the destination file is empty, you'll see several
seconds delay).

$ uname -a
Linux ip-10-0-0-127 4.15.0-1021-aws #21-Ubuntu SMP Tue Aug 28 10:23:07 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
$ grep ext4 < /proc/mounts
/dev/nvme0n1p1 / ext4 rw,relatime,discard,data=ordered 0 0

$ mkdir ~/slow-mv
$ cd ~/slow-mv
$ dd if=/dev/urandom of=5GB bs=1MB count=5120
5120+0 records in
5120+0 records out
5120000000 bytes (5.1 GB, 4.8 GiB) copied, 31.4953 s, 163 MB/s


NORMAL CASE



$ time mv 5GB 5GB.rename

real	0m0.001s
user	0m0.000s
sys	0m0.001s

$ time mv 5GB.rename 5GB

real	0m0.001s
user	0m0.001s
sys	0m0.000s



SLOW DOWN CASE


However, if we copy the file to a new file, and then rename it over the
original file, we see a slow down:

$ time cp 5GB 5GB.copy

real	0m27.597s
user	0m0.000s
sys	0m3.364s

This is the part I don't get:

$ time mv 5GB.copy 5GB

real	0m4.540s
user	0m0.000s
sys	0m0.321s

Even if the destination file is empty it still is slow:

$ cp 5GB 5GB.copy
$ rm 5GB
$ touch 5GB
$ time mv 5GB.copy 5GB

real	0m2.283s
user	0m0.000s
sys	0m0.041s

Just for completeness, the strace shows the slowdown
is in the rename syscall.


$ sudo strace -t mv 5GB.copy 5GB
   ....
    05:05:10 stat("5GB", {st_mode=S_IFREG|0664, st_size=5120000000, ...}) = 0
    05:05:10 lstat("5GB.copy", {st_mode=S_IFREG|0664, st_size=5120000000, ...}) = 0
    05:05:10 lstat("5GB", {st_mode=S_IFREG|0664, st_size=5120000000, ...}) = 0
    05:05:10 geteuid()                      = 0
*** 05:05:10 rename("5GB.copy", "5GB")      = 0
*** 05:05:18 lseek(0, 0, SEEK_CUR)          = -1 ESPIPE (Illegal seek)
    05:05:18 close(0)                       = 0
    05:05:18 close(1)                       = 0
    05:05:18 close(2)                       = 0
    05:05:18 exit_group(0)                  = ?
    05:05:18 +++ exited with 0 +++