7. Shared machines
On page 5, each node belonged to one person. A shared node runs the turns of every member. It must keep the turns of one person away from the files of another person. fleet calls this isolation.
| Node | Runs | Code |
|---|---|---|
| Personal | Only the agents of its owner | fleet/internal/sched/eligible.go |
Shared, uid isolation |
The agents of every member, but not kubejob. Each member has a unix account fleet-<hash>. |
fleet/internal/backend/process/uid.go |
Shared, pod isolation |
Only kubejob agents, one pod for each turn |
fleet/internal/backend/kubejob/ |
| Shared, no isolation | Nothing. The server gives it no turns. | fleet/internal/sched/eligible.go |
- An admin makes the join token of a shared node:
fleet node token --shared. The node starts with--isolation uid. - When a node claims, the server takes turns from each member in turn (round robin). It never
uses the profile of one member for the agent of another member
(
fleet/internal/node/coordinator.go). - A turn gets an access token for the profile of its owner. The refresh token stays on the server.
- When an admin removes a member, the server gives the personal nodes of that member no more turns. There is no drain state, and the node token stays valid. See add and remove a person.
Details: shared nodes.
Next: 8. Kubernetes and some clusters. A pod gives the strongest isolation.