[WIP] Update stacks when needed #123
Conversation
| if s == nil { | ||
| return false | ||
| } | ||
| return isComplete(s.status) |
There was a problem hiding this comment.
I think it would be better to move the whole isComplete(string) body to this function, because it's only used if you have a Stack and this will reduce number of functions spread.
| model[certificateARN+"/"+ingress.Scheme()] = &managedItem{ | ||
| ingresses: []*kubernetes.Ingress{ingress}, | ||
| Update: true, | ||
| } |
There was a problem hiding this comment.
Update is always true here. This is wrong..
| @@ -175,8 +157,12 @@ func buildManagedModel(certsProvider certs.CertificatesProvider, ingresses []*ku | |||
| } | |||
| if item, ok := model[certificateARN+"/"+ingress.Scheme()]; ok { | |||
| item.ingresses = append(item.ingresses, ingress) | |||
| item.Update = true | |||
There was a problem hiding this comment.
This Update = true is in both cases set. We have to lookup it somewhere to decide if there was a change that we now want to react on. Maybe we have a currentKnownState and check the new data to this one, but this would mean we are becoming stateful and are not stateless anymore. I don't see another way of doing it or trigger updates by time.Duration and additionally in the initial run.
There was a problem hiding this comment.
The state is in the cloudformation stack as it has the certificate arn and scheme defined. We just need to check against that an decide if there was a change.
|
Closing this in favor of #111 where these changes are now included and make more sense in that context. |
This PR aims to address two issues.
Stacks were scheduled to update by sending them to a channel and having a separate go routine triggering the updates every hour (by default). Instead of waiting one hour to do the update we can just check if an update is needed and schedule it right away. This is important for the SNI support (Add sni support #111) as we want to add new certificates to existing ALBs and not wait an hour for it to propagate.
FindManagedStackswas only returning stacks in a completed state, this resulted in more stacks being created in cases where a stack update was triggered and some stacks were not returned from theFindManagedStackscall because they were in an update state. Instead it just returns all stacks and checks the state when deciding if they should be updated or not. I.e. stacks already being updated will not be updated again until they reach a complete state first.I marked this WIP because I need to validate that this works correctly, but initial review is welcome :)