Ideas for improving the loadbalancer

[benediktkr]

Currently the load balancer is implemented (on the Model level) at `model.NodeList.best()'.

It works like this:

def best(cls, n=3):
    all_nodes = cls.get()
    candidates =  [a for a in all_nodes if a.alive and a.enabled]
    by_score = sorted(candidates, key=lambda a: a.score)[:n]
    random.shuffle(by_score)
    return cls(by_score)

This is then exposed through two calls, POST /node?filter=best (requiers authentication and gives full info on the nodes) and GET /lokun/loadbalancer?count=n (exposes IP and name).

• What happens to this if there are only 0, 1 or 2 nodes alive and enabled?

This has one minor problem. The best n nodes are picked and then shuffled around. I would like to pick the best nodes, even though the score varies by maybe 1-2, then shuffle them and return the n best ones. Something like this (in pythonic psuedo code):

def best(cls, n=3):
    all_nodes = cls.get()
    candidates = [a for a in all_nodes if a.alive and a.enabled]
    min_score = min(a.score for a in candidates)
    by_score = [a for a in candidates if a.score.within_logscale(min_score)]
    random.shuffle(by_score)
    return by_score

This would be less exclusive of nodes that are very close in score.