Daemon mode

rk daemon turns a host into a worker for Runik Control, which schedules jobs across many workers and keeps their reports. The worker connects out to Runik Control, runs one job at a time, and streams events and the report back. It keeps no state of its own between jobs.

Daemon mode needs the licence feature fleet.worker (Enterprise).

Shell

Flags

Flag	Default	What it does
`-control <host:port>`		required: Runik Control's address
`-token-file <path>`	`$RK_TOKEN_FILE`	the join token, its first line. `rk` warns if other users can read the file
`-token <t>`	`$RK_TOKEN`	the token on the command line; `rk` warns, because other users can see it in the process list
`-tls`		TLS with the system's trusted CAs
`-ca <pem>`		TLS with this CA
`-cert`, `-key`		mutual TLS. The certificate's CN or a DNS SAN must equal `-id`
`-server-name`		the TLS server name to verify
`-insecure`		allow cleartext. Without it, cleartext to anything but loopback is refused
`-id`	the hostname	the worker's identity
`-labels k=v,...`		labels for scheduling. Empty, duplicate or valueless labels are an error
`-drain-timeout`	20s	how long a shutdown waits for held reports to be delivered
`-max-held-bytes`	256 MiB	memory for reports waiting to be delivered (at least 64 MiB)
`-metrics-port`, `-metrics-addr`	off, 127.0.0.1	Prometheus metrics (also needs `obs.prometheus`); see Metrics
`-allow-cmd-actions`	off	allow `cmd` actions; a job can't turn this on
`-debug`, `-trace`, `-json`, `-logfile`, `-log-caller`		logging

How a worker behaves

Capabilities. It advertises what it can run: ngap, diameter, sbi, rest, pfcp, gtpu, fuzz, suite, narrowed to what its licence grants.
One job at a time. A job offered while the worker is busy or draining is answered with an error.
Every job is checked against the licence: its protocols, environment, rate, starts and subscribers.
Lost connection. A job keeps running when the connection drops. Its report is held until Runik Control acknowledges it: up to 16 reports and 8 suite cycles, within -max-held-bytes. When the budget runs out, the worker trims reports first, then drops the oldest.
Reconnect:
- it retries with backoff from 1s up to 30s, capped at 5s while it holds reports;
- if Runik Control refuses the worker (bad token or certificate), it exits 1, unless it holds reports. Then it keeps trying for up to 10 minutes.

Signals

First SIGINT or SIGTERM: the worker drains:
- it takes no new job and cancels the running one;
- it waits until its held reports are delivered, or -drain-timeout passes;
- it exits 0 after a clean drain, 1 after a timeout.
A second signal: it exits 1 at once.
SIGHUP: it re-reads the licence. A licence that doesn't load leaves the old one in place. Changed capabilities are re-announced.

Exit codes

2: no -control, or a bad flag.
5: the licence doesn't grant fleet.worker, or obs.prometheus when -metrics-port is set.
1:
- an unsafe transport (cleartext to a remote address without -insecure);
- a token problem;
- a metrics port that can't be bound;
- a refusal from Runik Control.