I just want to clarify to everyone here how the node selection process works:
This is because our offline window for node selection is 4 hours. In other words, nodes that we have seen within the past 4 hours are considered “online” when we select nodes for a new upload, regardless of whether they are actually online in that moment. So when you took one node offline, the satellite still considers both of them to be healthy, and selects each one with a 50% chance. After 4 hours pass, the satellite considers the one you took down to be offline/unhealthy, so 100% of the selections for that subnet go to your online node.
This is not true. If a node is not online to receive an order for an upload/download, it does not then receive that traffic when it comes back online. The satellite will select it with the same probability as before when it is online, regardless of how much traffic it missed out on.
Nope. As far as the satellite is concerned, a node is “healthy” or “unhealthy”. That is the only thing it looks at when selecting nodes for a new upload. If your node is disqualified, suspended, or offline for more than 4 hours, it is “unhealthy”. Everyone else is “healthy”. The satellite randomly selects from the pool of healthy nodes, and ensures that there are no duplicate node IDs or subnets. So you have the same chance of ingress/egress regardless of how much rejected traffic you had, as long as you are considered “healthy”.
didn’t littleskunk do an overhaul on that system a few months back, because it was like 3-4 months ago i was seeing this effect…
not that it really matters lol… maybe my memory of what exactly happened is a bit off… was a more or less irrelevant detail that popped up a few times… would always have been on downtimes longer than 4 hours tho… more in the 12-16 hour ranges… had a lot of hardware issues, so had some long down periods… haven’t seen it in a while, but haven’t really had any meaningful downtime in a good while…
so never had a chance to investigate it further… i suppose it could just be random chance… just seems so repeatable…
Node selection cache was implemented for efficiency, which made some significant changes to the code, but not to the criteria for node selection. I first looked into this code long before that and it has always been basically a random selection of healthy nodes.
If this was a question, of course you should keep it running. Ingress was never paid, only stored data and egress. So, what changes for a SNO when the node becomes full? Nothing.