I usually use find to search for files and analyze the output.
Reading the manual page i learned about nouser & nogroup test expressions.
So i’ve tried some test to find the quicker (or a better way) to remove files with find.
First, lets create a demo dir and a lot of files
# cp -ra /usr /usr.test
# chown -R 10101.10101 /usr.test
How many files do we have ?
# time find /usr.test/ -xdev | wc -l
124298
real 0m0.575s
user 0m0.243s
sys 0m0.363s
Ok, 124.298 files are a lot!
If i want to delete the entire directory via rm, the running time will be:
# time rm -rf usr.test/
real 0m5.883s
user 0m0.287s
sys 0m5.063s
5.88 seconds !
A walk through entire tree path:
# time find /usr.test/ -xdev -nouser > /dev/null
real 0m6.480s
user 0m2.763s
sys 0m3.660s
6.48 secs. It’s faster to remove them!
We now have a base to compare our results.
We will try 3 methods:
a. -delete find option
b. -exec find option
c. xargs via pipe
First Method
# time find /usr.test/ -xdev -nouser -delete
real 0m12.739s
user 0m2.826s
sys 0m9.513s
12.74 secs. Thats the twice amount of time
Second Method
# time find /usr.test -xdev -nouser -exec rm -rf {} ;
real 0m6.307s
user 0m0.253s
sys 0m5.516s
6.3 secs. Same as rm (that was expected by the way).
Third Method
# time find /usr.test/ -xdev -nouser | xargs rm -rf
real 0m4.666s
user 0m1.117s
sys 0m3.426s
4.66 secs!
So xargs is the faster way for the above methods
Friday, September 30, 2011 - 12:11:20
Although my initial test showed otherwise I redid it with more files and it proved your point above.
I also did a bit of googling and I found the following page which explains why xargs is faster. It is the fact that -exec rm -rf forks for every file whereas with xargs its like doing rm -rf file1 file2 …
http://www.gnu.org/software/findutils/manual/html_node/find_html/Deleting-Files.html
Friday, September 30, 2011 - 12:17:50
that’s quite right my friend, its all about threads ;)
Friday, September 30, 2011 - 14:09:31
Its funny that if find was actually using threads instead of waiting for each command to finish it whould have been much faster.
Anoter reference on the issue: http://www.sunmanagers.org/pipermail/summaries/2005-March/006255.html