orca.pool_bootHow a machine the pool autoscaler made becomes a worker (#267): its startup script, and the UI's answer to its /api/worker/boot.
The script (boot_script, the machine's cloud-init user data) holds one
secret, the machine's own worker token, which orca revokes when it removes
the machine. With it the machine asks the UI for the rest, over TLS:
boot_answer gives the worker image (ORCA_WORKER_IMAGE, the server's
own), the login to pull it (orca.registry), and the variables a worker
needs to reach the bucket (ORCA_WAREHOUSE, ORCA_LOGS, the AWS_*
ones), as the UI's own environment has them. It answers only a token the
autoscaler made for a machine that's live, so a token someone made with
orca worker-token create gets nothing from it.
The machine then installs Docker if it hasn't it, logs in, and runs the
image as orca worker --pool <pool> --name <machine>, with run cgroups
(docs/hetzner.md, "A cgroup per run"). With ORCA_POOL_WORKER_COMMAND
set on the scheduler, the script runs that command as the worker instead
(tests, without Docker).
boot_scriptdef boot_script(remote: String, pool: String, name: String, token: String, command: String = "") -> StringThe startup script of machine name of pool: ask remote (the UI) for what it needs with token, then run the worker, in Docker or as command when given.
boot_answerdef boot_answer(mut db: State, worker: Int) -> StringThe UI's answer to a machine's /api/worker/boot, for worker token worker: what its script needs, as JSON. Raises (the UI answers 403) for a token the autoscaler didn't make, or whose machine isn't live.
comptime BOOT_PATH = "/api/worker/boot"comptime CREATED_BY = "pool:"orca.workers.created_by of a token the autoscaler made: pool:<pool>.
comptime WORKER_IMAGE = "ORCA_WORKER_IMAGE"