Group controller always calling instance plugin with "properties=true"

[kaufers]

Group controller always calling instance plugin with "properties=true"

[chungers]

How about this..

We can create a wrapper instance plugin that implements the DescribeInstance method by returning cached result up a TTL. If the TTL expires then it does a real query by delegating to the wrapped instance plugin's DescribeInstances. We'd always query with properties=true. Go can support this type of mixin behavior via composition / embedding pretty easily and we would only have to implement one function. This could be a quick fix I'd try to see if that improve things.

[kaufers]

@chungers Sorry, I don't understand how this will help reduce the load that is incurred by including properties=true?

Are you suggesting that we implement a cache layer for every instance plugin? If so, we need to be careful about how'd we invalidate that cache. For example, a Describeinstances right after a Provision will need to return that newly provisioned instance -- same requirement for any Label or Destroy operation.

[chungers]

Just realized we'd have to invalidate the cached value whenever state changes. This means we'd have to implement the mutation methods like provision and destroy to forward to the delegated real plugin. When the delegate returns successfully, we need to invalidate the cache result and let the next DescribeInstances do the real query. What do you think?

[kaufers]

@chungers Yes, that's what I'm thinking. But we also need to think about filtering. The describe API accepts a tag map that it uses to filter. If we do this solution wouldn't we want to cache "all" instances in the wrapper, and then apply the tag filter to the cached instances? If so, we need to think about how the instance plugin would implement a describe without a tag -- we don't want to return all instances in the cloud provider account. Maybe we'd use something like the swarm ID since that'd be unique and apply to all groups?