On 01/14/2014 04:16 AM, S. Dale Morrey wrote:
I'm building an online service. I expect that this may have to scale to
tens of thousands of users. For the sake of having a drop dead simple
deployment. I decided to build the website front end on top of Drupal (the
service itself is linked to from the site, but is actually delivered by
node.js and thus wasn't part of this particular test).
I have had all kinds of fun in the past with PHP/MySQL and even Drupal
vulnerabilities, so I decided to sign the site up for regular security
scanning.
The scanner just fired up for the first time and when it was complete my
site had essentially become non-responsive. (Strangely the security
scanner marked the site as having passed though)
Logging in via SSH showed quite literally hundreds and hundreds of
<defunct> apache processes. I'm guessing that they were stress testing the
site and obviously the site fell down.
What bothers me is that there were hundreds of apache processes just
sitting there dead. Since the box is only a single core with the ability
to use up to 2 in a burst type situation (Amazon EC2 t1.micro to be
exact). I don't really see the advantage of spinning up a whole new
process just to deal with a new connection. Seems like death by a thousand
paper cuts to me.
I'm aware that node spins a new thread but not a new process. I would
think that apache would have a mode to do that as well. Is there any
advantage to 1 process per connection? Is there possibly a better
configuration I could try that would allow it handle the load better?
I'm aware that if the website is going to be under regular heavy load that
there are some really important steps I can take such as serving static
content from a CDN, putting the site behind an autoscaling load balancer
etc. Those are in the works, but before I do all of that I want to make
sure I have the best apache config I can considering the hardware
limitations. This way we only scale once we've made the most effective use
of hardware.
FYI the server is running Ubuntu 12.04 with all the latest updates
applied. Apache config is stock I'm even serving content right out of
/var/www. Same with PHP. The DB server is on it's own seperate instance
and is MySQL 5.6 managed by Amazon Simple RDS.
Thanks in advance!
Apache2 has multiple methods of doing concurrency called
Multi-Processing Modules. You'll might want to look at changing apache
to use the worker mpm module instead of the default prefork mpm module.
http://httpd.apache.org/docs/2.2/mpm.html is the starting place for the
docs related to the various apache mpm modules. Prefork is going to use
a new child process per connection where worker will spawn multiple
processes but then also have multiple threads in those children to
handle multiple connections. Though I'm not entirely sure how well
worker handles various other plugins, such as mod_wsgi. So you'll
probably want to do some research about worker with whatever
infrastructure you're using for your drupal installation.
It looks like there is a separate package for apache with prefork vs
apache with worker in debian and ubuntu. To get the worker version,
you'll need to install the apache2-mpm-worker package which will
probably replace apache2-mpm-prefork.
Hope that helps.
mike
/*
PLUG: http://plug.org, #utah on irc.freenode.net
Unsubscribe: http://plug.org/mailman/options/plug
Don't fear the penguin.
*/
