Reloading conf

Server NGINX
#1

Hi! How reloading my conf can affect my running nginx instances that are
under load?
Just made a small test and I saw that some requests to the upstream are
being lost. I am reloading the same configuration I had, didn’t change
anything.

Is there any best practice on how to do it? how often? if I should avoid
something in particular? etc…

Thanks for your help!

Jonathan

#2

Thanks! I will check it again. In general my nginx instances have very
low cpu and memory usage, way beneath 50%, probably around 10% or even
less.
And request in our app are avg 200 or 300 ms, so no long running
requests.
We are serving around 15k req/s on 5 nginx instances
(instance=machine), so maybe we have something wrong in the conf or
something.
Anyway, now I understand how it works, so it shouldn’t affect anything.

Thanks again!

Jonathan

#3

Hello!

On Mon, May 02, 2011 at 07:00:18PM -0300, Jonathan L. wrote:

Hi! How reloading my conf can affect my running nginx instances that are
under load?

While reloading nginx master process parses new configuration,
starts new worker processes and instructs old worker processes to
gracefully terminate (i.e. finish processing of all requests and
then exit).

This shouldn’t affect anything but may require additional
resources (mostly memory). If you don’t have enough memory reload
may be problematic operation, and it’s generally good idea to make
sure your running nginx doesn’t consume more than half of your
server memory under normal load (to make sure it will be possible
to start additional worker processes on reload).

Just made a small test and I saw that some requests to the upstream are
being lost. I am reloading the same configuration I had, didn’t change
anything.

It’s strange you see something being lost. Are you sure you
actually did reload (i.e. sent SIGHUP to nginx master process)?

Is there any best practice on how to do it? how often? if I should avoid
something in particular? etc…

General practice is to reload config when you’ve changed it.

If you are doing reloads often - you may want to make sure
previous reload was completed (i.e. old workers terminated) before
doing another one (there is nothing fatal if it’s not, but see
above about resources). Please note that with some use cases
(long-running requests, downloads and so on) reload may take
several hours or even more.

Maxim D.