2

we are using capistrano to deploy our PHP application. and php-apc is enabled on the production server with apc.stat = 0. In capistrano config there is a rule to clear apc cache after deploy:update :

after "deploy:update", "clear_apc_cache"

the cache is successfuly cleared: apc.php 'View Host Stats' shows an uptime of 0 minute, numbers of cached files & hits is close to 0 (not 0 because there is quite a lot of traffic and some files are cached before we can check apc.php)

the problem is, sometime an old file is still cached in apc. I think this is due to an HTTP request starting before capistrano update the 'current' symlink and ending after we clear apc cache. in that case, file included after apc clear in this request are still from the old revision folder (we are using relative path in includes and requires) and are saved in apc cache

Is there a way to permanently fix this? without updating all includes and requires?

for now I will try to delay the "clear_apc_cache" for one or two second but I'm afraid it will cause others problems during the delay (and some requests could still be longer than 2 seconds)

Improve this question

1 Answer 1

Reset to default
2
+25

This answer states you need to pass an argument to clear_apc_cache and that only with apc.enable_cli=1 will the APC cache clearing work through CLI calls.

You might want to clear_apc_cache again with the after:restart hook to ensure the previous release is fully cleared.

Improve this answer
4
  • well, clear_apc_cache is not call directly with a PHP-CLI call, the capistrano receipe use curl to call a script to clear the cache (and the cache is cleared) the problem is that current request fill the empty cache with opcode from the previous release
    – Mathieu
    Aug 7, 2013 at 20:27
  • @Mathieu see the after:restart note
    – parhamr
    Aug 7, 2013 at 20:43
  • nothing changes by clearing the cache a second time with after:restart. Unfortunately I think the two only solutions are to either fix the relatives includes* and requires* or to use a long enough delay
    – Mathieu
    Aug 11, 2013 at 21:00
  • @Mathieu is the PHP service ever restarted during the deployment? I have also used that as a fallback “nuke everything” operation.
    – parhamr
    Aug 14, 2013 at 18:53

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .