APC cache fragmentation problem

Question

I'm running APC (php cache) on a medium website (13 000 visits a day) on a CentOS server 5 running php 5.3.3 with APC 3.1.4 with 4Gb of RAM. In the last weeks I'm experiencing many "unable to allocate memory for pool" errors in logs, and often the site goes down.

I think the problem is in APC. From the stats I see that the cache is full after an hour or two, and the fragmentation always jumps to 100%. This are my config settings of apc.ini:

apc.cache_by_default    1
apc.canonicalize    0
apc.coredump_unmap  0
apc.enable_cli  0
apc.enabled 1
apc.file_md5    0
apc.file_update_protection  2
apc.filters 
apc.gc_ttl  3600
apc.include_once_override   0
apc.lazy_classes    00
apc.lazy_functions  0
apc.max_file_size   1M
apc.mmap_file_mask  /dev/zero
apc.num_files_hint  0
apc.preload_path    
apc.report_autofilter   0
apc.rfc1867 0
apc.rfc1867_freq    0
apc.rfc1867_name    APC_UPLOAD_PROGRESS
apc.rfc1867_prefix  upload_
apc.rfc1867_ttl 3600
apc.shm_segments    1
apc.shm_size    512M
apc.slam_defense    1
apc.stat    1
apc.stat_ctime  0
apc.ttl 7200
apc.use_request_time    1
apc.user_entries_hint   0
apc.user_ttl    7200
apc.write_lock  1

From APC stats I can see that the number of cached files is very high (40 000), and this is due to the plethora of cache files of phpbb and mediawiki. Do I have to prevent APC caching this files using apc.filters? Are there solutions to the problem?

Answer 1

You really should set apc.stat=0 on your production server and it will prevent APC from actually going to the IO to check if the file has been changed.

also set apc.slam_defense=0 as it's deprecated

Check out documentation for other options: http://php.net/manual/en/apc.configuration.php