More nodes, from more independent providers, let your engine survive more failures, and every node adds its own monthly price.
How fault tolerance works:
- An engine of n nodes tolerates f = floor((n - 1) / 3) failed or misbehaving nodes. That steps up only at 4, 7, 10 and 13 nodes: 4 to 6 nodes tolerate 1, 7 to 9 tolerate 2, 10 to 12 tolerate 3.
- To keep finalizing blocks, at least n - f nodes must be online: 3 of 4, 4 of 5, 5 of 6, 5 of 7. Below that the engine stops and writes fail.
- An engine needs at least 4 nodes.
Independence matters as much as count:
- Nodes from one provider can fail together, so on Add a node the Provider fault tolerance row counts providers, not nodes. It reads None below 4 providers, and then says how many providers the next level needs, or Best possible for this provider count.
- Catalog rows say New provider or In engine. Tick New providers only to see just the providers your engine doesn't use yet.
- When you add or replace a node, nodes on a physical machine your engine already uses can't be picked, and a Shared host badge on Replica nodes flags nodes that share one.
How it affects price:
- Each node has its own monthly price, set by its class, provider and data center.
- On Add a node, Monthly run-rate shows the total before and after, and Funded for how long the balance lasts before and after.
- Nothing is charged up front: the balance pays each node day by day, and every new node is paid for at least its first 7 days.
- Removing a node lowers the rent from the next daily check, so the balance lasts longer; nothing is refunded.
Good to know
- A 5th or 6th node doesn't raise the number of failures tolerated, and it raises how many nodes must stay online. The next step up in tolerance is 7 nodes.