Hot Code Loading experiences for APIs

I’ve been happy using Caddy’s reverse_proxy with Phoenix with very little configuration. I haven’t done blue/green with it but it seems like the cookie lb_policy would work for sticky sessions, though I guess not for API clients.

But do you actually need sessions to stick? If you’re bringing up a new server don’t you actually want new connections to go to the new server?

What’s your current infrastructure like? Deployment strategy and tooling?

You do not need hot code reload, you need better deployment strategy and connection draining (so requests “in progress” will be finished).

There are different options how to achieve seamless deployment and from my experience - if you can avoid hot code reload, then it is better to avoid it.

github acdtions → deploys to single instance elixir/phoenix server

i can setup 2 elixir servers but how will the session sharing work with elixir phoenix if i don’t use sticky sessions? do you have expeirence with this?

the sessions draining in blue/green deployment isn’t an issue

it’s also the exlir/phoenix webinterface app that will need sticky sessions or i will have to store the sessions if they are shared

By default sessions are cookie based. The only shared knowledge the servers need is the secret_key configured on the endpoint. Everything is sent to the client and back. That’s at least the case unless you switched to a different session implementation.

ok thanks so auth on server1→ cookie session is set on client, we use server2 all of a sudden but we sill use that same session and it’s accepted by server1 and server2 because they use the same secret_key?

i remember my elixir phoenix app locking the postgres, how does this work wihen 2 servers connect to the same postgres? arent they both try to lock the tables they work with? thanks!!