Several VMs, one fleet
The nodes reach the server over WireGuard (10.90.0.0/24). Each node sends a heartbeat and
claims work every 10 s.
- Each node asks the server for work. The server never opens a connection to a node
(
fleet/internal/node/agent.go). - Each node is a peer of the hub only. One node cannot reach another node.
- A node that sends no heartbeat for more than 30 s is lost. Its working agents become
lost. After 15 minutes, the server makes them again on another node (fleet/internal/node/coordinator.go). - The working directory is on the node. The registry, the ledger and the blobs are on the server.
- Install: Several VMs.
Elastic nodes
fleet autoscale, on any box with ssh to the hub, starts cloud servers as nodes of this mesh when agents of kind
elastic wait for 60 s, and deletes them after 20 minutes with no turn. Each one is a peer of
the hub at 10.90.0.10 and up; it takes only elastic agents, and a node with no kinds does not
take them. Install and revert: fleet/deploy/elastic.