file_exists() is too slow in PHP. Can anyone suggest a faster alternative?

Question

When displaying images on our website, we check if the file exists with a call to file_exists(). We fall back to a dummy image if the file was missing.

Howe

Answer 1

Old question, I'm going to add an answer here. For php 5.3.8, is_file() (for an existing file) is an order of magnitude faster. For a non-existing file, the times are nearly identical. For PHP 5.1 with eaccelerator, they are a little closer.

PHP 5.3.8 w & w/o APC

time ratio (1000 iterations)
Array
(
    [3."is_file('exists')"] => 1.00x    (0.002305269241333)
    [5."is_link('exists')"] => 1.21x    (0.0027914047241211)
    [7."stream_resolve_inclu"(exists)] => 2.79x (0.0064241886138916)
    [1."file_exists('exists')"] => 13.35x   (0.030781030654907)
    [8."stream_resolve_inclu"(nonexists)] => 14.19x (0.032708406448364)
    [4."is_file('nonexists)"] => 14.23x (0.032796382904053)
    [6."is_link('nonexists)"] => 14.33x (0.033039808273315)
    [2."file_exists('nonexists)"] => 14.77x (0.034039735794067)
)

PHP 5.1 w/ eaccelerator

time ratio (1000x)
Array
(
    [3."is_file('exists')"] => 1.00x    (0.000458002090454)
    [5."is_link('exists')"] => 1.22x    (0.000559568405151)
    [6."is_link('nonexists')"] => 3.27x (0.00149989128113)
    [4."is_file('nonexists')"] => 3.36x (0.00153875350952)
    [2."file_exists('nonexists')"] => 3.92x (0.00179600715637)
    [1."file_exists('exists"] => 4.22x  (0.00193166732788)
)

There are a couple of caveats.
1) Not all "files" are files, is_file() tests for regular files, not symlinks. So on a *nix system, you can't get away with just is_file() unless you are sure that you are only dealing with regular files. For uploads, etc, this may be a fair assumption, or if the server is Windows based, which does not actually have symlinks. Otherwise, you'll have to test is_file($file) || is_link($file).

2) Performance definitely degrades for all methods if the file is missing and becomes roughly equal.

3) Biggest caveat. All the methods cache the file statistics to speed lookup, so if the file is changing regularly or quickly, deleted, reappears, deletes, then clearstatcache(); has to be run to insure that the correct file existence information is in the cache. So I tested those. I left out all the filenames and such. The important thing is that almost all the times converge, except stream_resolve_include, which is 4x as fast. Again, this server has eaccelerator on it, so YMMV.

time ratio (1000x)
Array
(
    [7."stream_resolve_inclu...;clearstatcache();"] => 1.00x    (0.0066831111907959)
    [1."file_exists(...........;clearstatcache();"] => 4.39x    (0.029333114624023)
    [3."is_file(................;clearstatcache();] => 4.55x    (0.030423402786255)
    [5."is_link(................;clearstatcache();] => 4.61x    (0.030798196792603)
    [4."is_file(................;clearstatcache();] => 4.89x    (0.032709360122681)
    [8."stream_resolve_inclu...;clearstatcache();"] => 4.90x    (0.032740354537964)
    [2."file_exists(...........;clearstatcache();"] => 4.92x    (0.032855272293091)
    [6."is_link(...............;clearstatcache();"] => 5.11x    (0.034154653549194)
)

Basically, the idea is, if you're 100% sure that it is a file, not a symlink or a directory, and in all probability, it will exist, then use is_file(). You'll see a definite gain. If the file could be a file or a symlink at any moment, then the failed is_file() 14x + is_link() 14x (is_file() || is_link()), and will end up being 2x slower overall. If the file's existence changes A LOT, then use stream_resolve_include_path().

So it depends on your usage scenario.