Static cookie required for zero-downtime deployments?
Closed this issue · 3 comments
It seems that if an app does not have a static cookie set in a fly.io secret, then during deployment some instances may be unable to connect to nodes in primary region until those instances are replaced. So, for example, apps using fly_postgres_elixir could lose some writes.
If so, then maybe this should be mentioned in the docs (both here and in fly_postgres_elixir)?
Something like, "Apps that require zero-downtime deployments should have a RELEASE_COOKIE secret set so that all nodes can keep talking to primary region nodes during release rollouts"?
That's a good point. Thanks for suggesting this!
Thanks for the suggestion. I added it to both projects.
- https://github.com/superfly/fly_rpc_elixir#prevent-temporary-outages-during-deployments
- https://github.com/superfly/fly_postgres_elixir/#prevent-temporary-outages-during-deployments
Do the instructions make sense? Would they have helped you avoid the problem?
Oh yeah, this looks great!