Forum: NGINX Strange problem with cache

Hi,

i've come accross a strange problem of caching. My site is run by nginx
1.2.1, on two different servers. The config are exactly the same (i use
puppet to mirror configs). The contents are exactly the same, everything
that i can think of is exactly identical on both servers.

In the config, i set "expires max" for the /static location.

The two nginx are behind a haproxy instance for loadbalancing and 
failover.
The setup is in active-active mode.

When i load a page with no browser cache, each item in the static 
location
is served by nginx with the correct response regarding the max-age and
expire directives. Then i reload the page : some of these ressources are
served with a 304 response (that is the expected behaviour), some with a 
200
code (hence, a fresh copy of the file). If i reload the page again and
again, the ressources served with 304 answer are varying.

Now, i've put the server into active-backup mode, and only one instance 
of
nginx is serving the files. In this setup, the ressources are always 
served
with a 304 answer.

Did i miss something? I assume that my servers setup is quite common, 
and i
d'ont believe that the behaviour i observe is "normal".

Any idea?

thanks

Florian

Posted at Nginx Forum: 
http://forum.nginx.org/read.php?2,232235,232235#msg-232235

Hello!

On Thu, Oct 25, 2012 at 04:34:14PM -0400, florian.iragne wrote:

> The setup is in active-active mode.
> with a 304 answer.
>
> Did i miss something? I assume that my servers setup is quite common, and i
> d'ont believe that the behaviour i observe is "normal".
>
> Any idea?

Symptoms suggest last modification times are different for the
same static file on different servers.

You have to sync files between servers with modification time
preserved for conditional requests to work correctly ("rsync -a"
is usually a good starting point).

--
Maxim Dounin
http://nginx.com/support.html

Ok, now i understand. The browser sends a request withe the
if-not-modified-since corresponding to one server to the other server, 
and
this one reply with 200 since the mtime is in the future of the
if-not-modified-since

is there any simple solution to handle this? I don't want to rsync 
assets
between the servers

I use django in my project and its static/collectstatic process. So, i 
force
the mtime of each static file collected, afet each collecstatic 
operation. I
give the command line if someone fall in the same trap:
./manage.py collectstatic --noinput && find static -type f -exec touch 
-t
$(date +%m%d%H)00 '{}' \;

Now, the cache is working as intended and my servers are in 
active-active
setup.

Thanks for your answer that leads me to the solution

Posted at Nginx Forum: 
http://forum.nginx.org/read.php?2,232235,232254#msg-232254

Yesterday Oct 26, 2012 at 07:57 florian.iragne wrote:

> Ok, now i understand. The browser sends a request withe the
> if-not-modified-since corresponding to one server to the other server, and
> this one reply with 200 since the mtime is in the future of the
> if-not-modified-since
>
> is there any simple solution to handle this? I don't want to rsync assets
> between the servers

http://nginx.org/r/if_modified_since
if_modified_since before;

--
WNGS-RIPE

Thanks for your answer. However, it's not secure enough for me and i 
prefer
adding one more command line to the deployment process to ensure cache 
is
working.

thanks

Posted at Nginx Forum: 
http://forum.nginx.org/read.php?2,232235,232297#msg-232297

Hello!

On Sat, Oct 27, 2012 at 03:35:51PM +0300, Oleksandr V. Typlyns'kyi 
wrote:

> http://nginx.org/r/if_modified_since
> if_modified_since before;

1) This will work only if you'll switch off ETag's as well.

2) This isn't a solution, but a bandaid.  It won't fix the real
problem (mtimes mismatch), but will make response with greatest
mtime eventually win.  Even with only 2 servers used this
will still result in about 1.5x more 200 responses than needed.

--
Maxim Dounin
http://nginx.com/support.html