APC Purge Rate & Fragmentation

Question

I have got a Magento running on Nginx, PHP-FPM, Varnish & APC setup.

APC is running high on both fragmentation and purge rate.

Fragmentation increases in 2 hours >50% and slowly goes to 100%.

Even more disturbin is the fact and Munin shows that the purge rate is almost in 100% all the time.

I tried to increase the shm size to 1G (APC normally uses about 200M of this, total RAM is 6G) which didn't help, as well tried to tweak the ttl:s without any luck.

Is there something in varnish that is causing the constant purging of APC?

Here is my APC.ini

extension = apc.so
apc.enabled=1
apc.shm_segments=1
apc.shm_size=1024M
apc.num_files_hint=10000
apc.user_entries_hint=10000
apc.max_file_size=5M
apc.stat=0
apc.enable_cli = 1
apc.optimization=0
apc.cache_by_default=1
apc.slam_defense=0
apc.include_once_override=0
apc.ttl=3600
apc.user_ttl=7200
apc.gc_ttl = 600

apc.php:

**File Cache Information**
Cached Files    1053 ( 58.1 MBytes)
Hits    40172
Misses  1038
Request Rate (hits, misses) 102.26 cache requests/second
Hit Rate    99.68 cache requests/second
Miss Rate   2.58 cache requests/second
Insert Rate 10.41 cache requests/second
Cache full count    0

**User Cache Information**
Cached Variables    74 ( 1.2 MBytes)
Hits    374
Misses  5907
Request Rate (hits, misses) 15.59 cache requests/second
Hit Rate    0.93 cache requests/second
Miss Rate   14.66 cache requests/second
Insert Rate 0.18 cache requests/second
Cache full count    0

Varnish should not affect the purges on APC. Usually the easiest way to find details about your APC is this apc.php file that comes with the apc source together. If you execute it inside your PHP-FPM it can show you details about the content of the APC, based on this information you might be able to find the reason why you have so many purges. — replay
Thanks for the answer. I have been trying to find out the reason from apc.php without any luck. I pasted the apc.php details above. — Peter Mann
I would increase your apc.ttl and lower your apc.user_ttl. If you have pm.max_requests set in php-fpm, it can affect this as well. This is because cache for that process is dropped when the process hits its max request limit and the process is terminated. This, in turn, creates fragmentation. — tacotuesday
Thanks for this advice. What would you say, switching it to apc.ttl=7200, apc.user_ttl=3600 could be close to optimal? — Peter Mann

Synchro Synchro · Accepted Answer · 2013-07-10T17:11:01

One trick I've found effective in keeping APC sane is to set gc_ttl to 600, as you have, but keep ttl at 0. This is less efficient and could theoretically cause stampedes on a very busy site, but it stops fragmentation building up. You also need to keep your purge rate down by making sure you're not placing too much stuff in the cache (i.e. allocate it enough memory).

This may sound counterintuitive, but here's what happened when I switched to this approach on a server today:

Cache hit ratio Fragmentation Purge rate

This munin plug-in is great for monitoring how APC changes over time, much better than trying to decipher the apc status page.

APC Purge Rate & Fragmentation

1 Answers