server: upstream connection rebalancing

[lhpqaq]

29

I have implemented a periodic rebalancing that executes on a schedule. If this does not meet the requirements, please guide me on the next steps to take. Additionally, it might lack integration with other modules (such as gossip). Thanks. https://github.com/andydunstall/piko/issues/29#issuecomment-2252815506

[andydunstall]

Hey @lhpqaq, thanks for this contribution! I'll try to review it early next week

[andydunstall]

Hey @lhpqaq, nice work with the PR, though if you don't mind I'll leave this for now so will close. My priority for Piko at the moment is to extend the test coverage of the existing Piko functionality and improve documentation, so I'm not sure this is needed yet

Theres a lot of good stuff here in the PR, just to provide some feedback incase its helpful:

• I think this needs a 'minimum' threshold (say 100) to avoid excess rebalancing, say you have 2 nodes with 3 upstreams, one node will always have more than 20% of the other?
• I'm always keen to keep state and methods private if possible, particularly state like Conns which requires a mutex so shouldn't be public
• It would be great to have unit tests as well as system tests (plus aiming to keep all tests fast, the currently slowest test is about half a second so would be good to avoid long sleeps)