Application Load Balancer (ALB)

Pre-requisites

• Docker Toolbox 1.9
• A running Docker engine provisioned via Docker Machine (assumed to be "alb" in the rest of this documentation):

docker-machine create -d virtualbox alb

 Building the container

docker build -t tldr_alb .

How it works

TODO

Running the container

Use the TLDRLBSERVICE_NAME environment variable, which identifies the names of the services that this load balancer should work with.

Variable SERVICE_TAGS should contain tldr.type:lb to identify this component as a load balancer, and tldr.app:testapp to identify the app to which this load balancer belongs.

Also, provide the URL to the Consul server with the service information as a parameter:

docker run --name=web_lb -e SERVICE_TAGS=web -e SERVICE_NAME=web_lb --dns 172.17.0.1 -p 80:80 -p 1936:1936 --rm alb -consul=$(docker-machine ip alb):8500

When ready, visit http://$(docker-machine ip alb):1936 (someuser/password) to verify that both nginx containers are visible to haproxy. To access nginx, use http://$(docker-machine ip alb)/web-80 (the 'web-80') part is a known issue.

Test run

To load balance a service called service1_testapp-80, use the following command:

docker run --name=service1_testapp_lb -e TLDR_LB_SERVICE_NAME=service1_testapp-80 -e SERVICE_TAGS=tldr.type:lb,tldr.app:testapp -e SERVICE_NAME=service1_testapp_lb --dns 172.17.0.1 -p 80:80 -p 1936:1936 --rm pass_alb -consul=$(docker-machine ip alb):8500 -dry -once

Alternatively, use the scripts under the test folder:

1. Run tests/setup.sh to bootstrap a docker-machine host with the needed components (Consul and registrator)
2. Run tests/startService.sh 1 to start two instances for the given service id/number
3. Run tests/startAlb.sh 1 to start a load balancer for the given servie id/number. Must match with the parameter provided to startService.sh.

Known issues

• Haproxy restart after updating configuration may not correctly work, and instead multiple haproxy instances are spawned. This eventually leads to a situation where haproxy configuration data is not correct and the entire load balancer has to be restarted. This is due to issue https://github.com/hashicorp/consul-template/issues/442, which should be monitored. In the meantime, we have downgraded to consul-template 0.10.0 where the issue cannot be reproduced.
• Zero downtime Haproxy reloads: http://engineeringblog.yelp.com/2015/04/true-zero-downtime-haproxy-reloads.html