APC Server Cache Fragmentation

Question

I'm running a Wordpress site with APC and W3TC on CentOS 5.5 which gets around 10000 pageviews a day (including ajax and WP-Admin). The server is a low-level VPS and has a max of 376MB RAM.

Recently I've started getting a few server crashes which I can now attribute to APC (I disabled it for a little while and the crashes stopped).

After reinstalling and looking at the APC graphs I can see that we're getting a lot of fragmentation, just a few seconds after launching PHP-FPM. In some cases 100% fragmentation can occur with a few minutes.

Current cache size is 40MB and I'm slightly hesitant to raise it due to the memory limit on the server.

I've set a ttl on the cache at 2 hours, but I'm afraid that it may be too long considering the amount of fragmentation that's occurring, however if I set it too short, it would defeat the entire purpose of the cache.

I've also noticed that when setting userttl above 0 I get loads of errors when certain require/include/require_once/include_once functions are called.

Any help would be greatly appreciated.

Answer 1

If your running APC 3.0.6 like is suggested in W3 Total Cache installation instructions your probably getting the potential cache slam avoided errors that fill up your logs very quickly.

Uninstall APC and rebuild using version 3.1.7 (3.1.9 is current stable) which fixes the cache slam bug and numerous other issues.

You will also get better results running latest PHP 5.3.6 every version gets better at interacting with APC (they will be completely integrated when PHP 6 comes out).

Also take a look at your config file and the apc.slam_defense setting and raise it to around 30 or 40.

For a real performance boost set apc.stat to 0 and it removes the process of checking for a new version of the file. If you make a lot of changes to your site they won't update till the cache expires. (My users get to frustrated so I have to keep it set at 1)

setting apc.slam_defense to 75 would mean that there is a 75% chance that the process will not cache an uncached file. So, the higher the setting the greater the defense against cache slams. Setting this to 0 disables this feature.

Lower the apc.user_ttl which is what W3 Total Cache uses and it the more dynamic stuff. Keep the apc.ttl high because this will be your WordPress core files and theme files that you don't really change.

I assume your setting object and db cache in W3 to APC. For your server size I would set page caching to disk enhanced. Also remember you can set the ttl for the dynamic stuff in W3.

Here's mine running a pretty large WordPress Multisite on a dedicated server with 12G Ram. Don't be afraid to increase the cache size even with limited memory. You can always go back down if needed. Mine is set at 512M but it never uses more than around 120M.

; configuration for php apc module
extension = apc.so
apc.enabled = 1
apc.shm_segments = 1
apc.shm_size = 512M
apc.optimization = 0
apc.num_files_hint = 2700
apc.user_entries_hint = 2700
apc.ttl = 7200
apc.user_ttl = 3600
apc.gc_ttl = 600
apc.cache_by_default = 1
apc.slam_defense = 1
apc.use_request_time = 1
apc.mmap_file_mask = /dev/zero
apc.file_update_protection = 2
apc.enable_cli = 0
apc.max_file_size = 2M
apc.stat = 1
apc.write_lock = 1
apc.report_autofilter = 0
apc.include_once_override = 0
apc.rfc1867 = 0
apc.rfc1867_prefix = "upload_"
apc.rfc1867_name = "APC_UPLOAD_PROGRESS"
apc.rfc1867_freq = 0
apc.localcache = 1
apc.localcache.size = 1350
apc.coredump_unmap = 0
apc.stat_ctime = 0

Answer 2

Do you have a high "Cache full count"? Sounds like you need to increase the size of your apc.shm_size in your php.ini to reduce cached items being removed from the cache. Still, it is disappointing that this is causing stability issues...