NGINX Work Distributor

This project was created to be a "jumping-off point" to have NGINX configured as a load-balancer and reverse-proxy for a number of micro-services under high load. Don't take my work as gospel, read the docs and make adjustments based on your needs.

Also, note that you will need to tune your Linux configurations separately. There are several great posts on this topic floating around the internet including this one from the NGINX blog which also covers a lot of the configurations in this repository.

Setup

The config files in this project assume that the default NGINX file structure is left in-tact and that this repo is cloned into a sub-directory of the nginx/conf directory named distributor. It is also assumed that your main nginx.conf is empty except for an include to the distributor/global.conf file:

cd /path/to/your/nginx.conf/directory
git clone [email protected]:dgarlitt/nginx-work-distributor.git distributor
mv nginx.conf nginx.conf.original
echo "include distributor/global.conf;" > nginx.conf

Logging

In global.conf the access_log is turned off, which disables the writing of access logs for all routes. Though having the access log disabled definitely reduces disk I/O, it is likely that you will want to turn it back on for monitoring. Below are some of the possible configurations.

Alternative Access Log Configurations

Access logs are often an important source of information and, when enabled, can be tuned to perform better. Listed below are some additional options for getting better performance from the access_log directive. See the NGINX docs for more info

Buffer and Flush

The access log can be configured to only write when the buffer is full or the flush timeout is reached. By default, NGINX doesn't buffer writes to the access log.

access_log /path/to/log.file buffer=128K flush=30s;

In the above case, the access log will be written to when the 128K buffer fills or when the flush timeout is reached, whichever comes first. In place of buffer, gzip=[compression-level] can be specified which will automatically buffer.

Conditional Logging

Logging could be adjusted to use a condition for more granular control over what is logged by following the the advice of this stackoverflow post.

Disabled Per-Server

Alternatively, it could also be changed so that the access log is enabled by default in the global.config:

access_log /some/file/path/to/some.log

and then turned off using access_log off on a per-server basis inside of each server block:

server {
  access_log off
  more server configs...
}

Note, however, that this would be useless in the default configuration for this repo, as there is only one server block which is located in the route.config file.

Send Logs to Syslog

Additionally, access or error log output can be directed to syslog or a unix-domain socket, see here for more info.

Log Rotation and More...

DigitalOcean posted this great article with more logging options including how to configure log-rotation in NGINX.

Keepalive

Make an effort to tune the keepalive settings to make sense for your stack. Moreover, optimize the keepalive property on a per-upstream basis. More info from NGINX.

Disclaimer

Please be mindful that the files in this repo make use of some advanced configuration options available in NGINX. As such, before using this in your own projects, it would be wise to go through the files and make sure you understand each of the settings and adjust them according to your needs. I cannot guarantee that these settings will work for everyone, so use at your own risk. Other than that, best of luck!

Contributing

Contributions are always welcome. If you would like to contribute to this project, follow these easy steps:

• Fork this repository
• Create a feature branch for your changes
• If you have a lot of commits, rebase from master and squash your commits to a single commit with a meaningful message
• Make sure that your configurations are passing

sudo nginx -t

• Make sure your commit messages follow this guide
• Open a pull request