twolfson
commented
4 years ago Using a load balancer seems like the more intuitive way to go. There are some tricks for slow boot time programs (e.g. using special signal to tell parent process to restart children) but that's not our scenario
- https://dzone.com/articles/zero-downtime-deployment
- https://coderwall.com/p/fmuira/zero-downtime-deployments-w-unicorn-supervisord
And setting up a load balancer/running multiple servers to transition to is more costly and requires even more maintenance
(not to mention figuring out how to automate more private provisioning content like SOPS (though this prob means share it or set up another secrets solution) and updating all our SSL logic to live in the load balancer itself
Going to close this as a wontfix for now
This is an issue that was recorded a while ago (2013-12-07) but is still incomplete.
When we deploy, we can run into issues where we take down the server to restart it.
One approach is run 2 servers on different ports and then load balance from one to the other.
We can do this via vanilla servers or Vagrant.
Another option would be to use UNIX sockets.