Background

Today ClusterPools immediately hibernate new clusters. This means a ClusterClaim is guaranteed to suffer at least Resuming time before the cluster is available for use.

User Story

As a consumer of a (properly tuned) ClusterPool,

I want ClusterClaim to give me an active cluster immediately in the common case,

So that I can get my work done more quickly.

(Ideally without sacrificing the cost savings of having some clusters in the pool hibernating.)

Done Criteria

• It is possible to tune a ClusterPool such that some number of clusters  (0 <= N <= all) become/remain Active for some amount of time before hibernating (0 <= T <= never).
• A ClusterClaim first attempts to claim an Active cluster before Resuming a Hibernating one.
• Stretch: If there are no Active clusters, but >1 Resuming clusters, ClusterClaim attempts to claim the one closest to Active (least recently asked to resume).

Implementation

To be discussed in an enhancement doc.