The cluster discovery process aims to keep administration and management overheads at a minimum and take advantage of the highly dynamic platforms and tools available today (“The Cloud”).
So how does cluster discovery work?
DiscoverySettingobjects though the API.
discoverycomponent periodically runs at fixed intervals. The interval should be short as it determines the delay between discoveries needing to run and them being scheduled.
discoveryrun looks for any discoveries with an expected next schedule time in the past. If no cluster discovery needs to run the
discoveryrun does nothing.
discoveryrun schedules a discovery task for each discovery that needs to be performed.
now() + discovery interval.
When clusters and nodes are automatically discovered they can also automatically go away.
This feature is not currently available and is yet to be designed in full.
The primary use case for Replicante is part of an automated, distributed, dynamic infrastructure that scales from a small number of small clusters to a large number of large clusters.
It is assumed that managing a list of nodes is at best impractical, but may even be impossible in combination with tools such as auto scaling groups and automated instance provisioners.
The idea of cluster discovery was inspired by Prometheus. Cluster discovery has several advantages: