Autoscaling functionality

[ETisREAL]

How many people would be interested on an autoscaling functionality?
It would works as follow: the manifest will allow to specify a threshold of max connections a DB is able to handle. A cron job will be scheduled to periodically count the connection number in each DB (to create a sort of view) then every x seconds the operator will go check whether any of the clusters that have the autoscaling functionality enabled need to scale.

The parameters to specify in the postgres manifest are these:

  clusterAutoscaler:
    metricsUpdateIntervalSecs: 5
    trigger: 15
    leeway: 3

metricsUpdateIntervalSecs: how often to run the cron job query that registers how many connections are opened to a DB (the more frequent, the more precise but also the more demanding)
trigger: the value of connections above which an upscale should be triggered (only the highest number across all the DBs will be considered)
leeway: the standard deviation of how many points the value should actually differ from the trigger to cause a scaling event (in order to prevent continuous scaling for outliers or transitory events)

Currently known limitations:
Of course spinning up and down the cluster requires time, and the more data there is in the DB, the longer it would take. However I have personally implemented this on a semi-production environment (we are currently in the midst of switching over) and to give you an idea, for a DB that holds 1.5 GB of data in a VM with 4 cores and 16 GB or RAM is actually handling the scaling more than decently. Scaling up and down in about a minute or so (of course network is also a main factor).

This solution is definitely intended for small/medium sized applications, but I found it actually has a lot of potential. If a decent amount of people are interested in it, I will continue the development and as soon as possible at least publish a branch in the main repo.