📎 Webclip
Why Hot Deploys are Cool
Greg Vaughn says he values hot deploys in Elixir because most of his releases only change plain modules, not supervision trees or long-running process state. In that setup, Distillery can generate relup files for most deploys, while full restarts handle the harder cases.
Reading notes#
- He links hot code loading to Erlang’s original zero-downtime requirement and treats it as part of the platform’s history.
- He distinguishes between relying on hot deploys and simply benefiting from them.
- Relying on them means designing the system to stay up continuously and handling appup, relup, and code_changed cases carefully.
- He says most of his deploys do not need changes to the supervision tree or long-running GenServer state.
- Distillery lets him hot deploy plain modules for more than 90% of cases without extra work.
- For the remaining cases, he uses a full restart release and depends on durable storage and cache refresh at startup.
- He says hot deploys let background tasks keep running without interruption.
- He notes that Docker and orchestration can make hot deploys less relevant in some organizations.
- He says his current startup context fits Elixir and BEAM well because the system is mostly periodic background work with a smaller Phoenix UI.
- He advises newcomers not to dismiss hot deploys outright and to test them in a side project to understand where they fit.
