APC acting really funky

Andy Huang

Well-known member
I've actually never seen this issue before...
I've allocated plenty of space to APC (64M), and there are some fragmentations according to the detailed usage and fragmentations... But the pie chart, wow, looks crazy!
I've tried increasing the space (was 32M, it never used that much), and didn't seem to help.
I've tried restarting Apache, and the problem comes back after an hour or so.
Does anyone have any idea what may be causing this, and how can I make the data make more sense?
Voila_Capture39.webp
 
Wow, that's a mess, heh

I never had any luck with APC or Xcache, so stuck with EAccelerator for the time being.
 
The pie chart is pretty and all that but the useful information is in the panes to the left which you've cropped out.

Also APC-3.1.8 is out as of a few hours ago (has one show stopping bug that's fixed in apc_main.c in svn).
 
The madness continues. I've been updating via pecl, since I'm lazy to compile from source... But maybe I'll give that a try later...
 

Attachments

  • edited.webp
    edited.webp
    100.9 KB · Views: 24
Ours is a bit fragmented, but nowhere near as much as yours. Our version is much older (3.0.14), but since it is running well enough, I just leave it alone. I briefly had an earlier version when I first set it up, and had to disable it a few hours after since it crashed httpd several times.

Is this on a server hosting a forum? If so, how many active users at a time? Just curious to compare to ours--we have about 20 fragmentation entries in the bar chart currently. We have over 700 online around this time of day (about 40% members/60% guests).
 
64MB isn't enough, my APC burns through 192MB+ easily
I've set it to 256MB now, and it is doesn't seem to help. The chart and fragments are still all over the park.

Voila_Capture44.webp

I'm not using eAccelerator because last I checked it was not actively developed, and I'd much rather to use something that's more up-to-date. I've changed the config.php file to relocate the user cache into memcache, but that's only like a few kb's at my current configuration, majority of the opcode cache still sits in this crazy looking APC.
 
There are many other settings for APC, but I've never found a really clear reference online that told exactly what the settings were for, and what they affected.

Our default APC installation (with a few minor adjustments) cut our loads in half, and for me to tweak any further would be a case of diminishing returns.

eAccelerator doesn't look dead by any means, but it is not all that actively developed either--the last release was almost a year ago. Other releases were slow in coming.
 
Your APC looks fine on both 64M and 256M

Since your cache isnt actually filling up completely, the fragmentation is from things expiring due to your TTL.
 
Top Bottom