Finally managed to replace two VMs with a single VM.



This gives me a service discovery layer (Consul), moves the scheduler Sidekiq job off of the www node, and provides a platform for various background jobs to run automatically.

Now I can move www to an autoscale group.

though I think I'll upgrade redis/elasticsearch and postgres first, so they're wired into the service discovery plane

also I have to do major postgres version upgrades, so there'll be a Significant Downtime when I do that

so I guess Postgres is next on the list of "things to upgrade"

@aurynn ooh is your architecture documented at all? Mine is very simple but I might like to alter it a bit.

@daedalus I'll be documenting it after I replace it all 🙂

@daedalus Complex. Overkill for what I need, maybe, but it's not too complex.

@daedalus I'm extensively using cloud tech. It's going to be complex. 🙂 Can't really avoid that.

@aurynn complex and cloud kinda go together. Overkill is fine because it often turns out you needed it.

@daedalus This is true.

Unfortunately, cloud-overkill also means that runtime costs are higher than people expect.

@aurynn Ah. Yes there is that. I was just looking at some of mine yesterday. But I get much better uptimes and bandwidth so the expense it worth it, I feel.

@daedalus also we end up with systems engineered for resilience and self-repair, which is nice

Sign in to participate in the conversation
Cloud Island

A paid, early access, strongly moderated Mastodon instance hosted entirely in Aotearoa New Zealand.