Skip to content

semaphore.md

Latest commit

 

History

History
136 lines (109 loc) · 5.7 KB

semaphore.md

File metadata and controls

136 lines (109 loc) · 5.7 KB
Raw

Semaphore

Read when:

  • choosing provider: semaphore;
  • configuring Semaphore host, project, machine, or auth;
  • changing Semaphore lease creation, SSH resolution, or cleanup.

provider: semaphore provisions a standalone Semaphore CI job and hands it back to Crabbox as a plain SSH target. Semaphore owns the CI job, project, machine image, and API auth; Crabbox owns local config, the lease slug and repo claim, repo sync, remote command execution, and normalized list/status output. The integration uses Semaphore's REST API directly — no sem agent or self-hosted runner is involved.

How a lease works

  1. warmup/run resolve the configured project name to a project ID, then create a standalone job named crabbox testbox. The job runs a keepalive command that prepares the work root, prints a ready marker, and sleeps for the idle-timeout duration.
  2. Crabbox polls the job until it reaches RUNNING and exposes an agent IP plus an ssh port, then fetches the job's debug SSH key.
  3. The SSH key is written under the lease's testbox key directory and the lease is exposed as SSH user semaphore at the agent IP/port.

Lease IDs are sem_<job-id>. Crabbox only manages jobs whose name begins with crabbox testbox, so list/status/stop ignore unrelated Semaphore jobs.

Auth

Provide the host, project, and an API token. Prefer environment variables or user config; do not commit tokens to repo config.

export CRABBOX_SEMAPHORE_HOST=example-org.semaphoreci.com
export CRABBOX_SEMAPHORE_PROJECT=my-app
export CRABBOX_SEMAPHORE_TOKEN=...

Semaphore-native variable names are also accepted as fallbacks:

export SEMAPHORE_HOST=example-org.semaphoreci.com
export SEMAPHORE_PROJECT=my-app
export SEMAPHORE_API_TOKEN=...

Create an API token at https://<host>/me/api-tokens. The host must be a bare host name (for example example-org.semaphoreci.com), not a full API URL.

Config

provider: semaphore
target: linux
semaphore:
  host: example-org.semaphoreci.com
  project: my-app
  machine: f1-standard-2
  osImage: ubuntu2204
  idleTimeout: 30m

Defaults when unset: machine: f1-standard-2, osImage: ubuntu2204, idleTimeout: 30m. host, project, and token are required.

Setting Config key Flag Env (and fallback)
Host semaphore.host --semaphore-host CRABBOX_SEMAPHORE_HOST (SEMAPHORE_HOST)
Project semaphore.project --semaphore-project CRABBOX_SEMAPHORE_PROJECT (SEMAPHORE_PROJECT)
Token semaphore.token — CRABBOX_SEMAPHORE_TOKEN (SEMAPHORE_API_TOKEN)
Machine semaphore.machine --semaphore-machine CRABBOX_SEMAPHORE_MACHINE
OS image semaphore.osImage --semaphore-os-image CRABBOX_SEMAPHORE_OS_IMAGE
Idle timeout semaphore.idleTimeout --semaphore-idle-timeout CRABBOX_SEMAPHORE_IDLE_TIMEOUT

The token has no flag; pass it via env or config only. Semaphore API error bodies redact the configured token before they reach CLI diagnostics.

Equivalent one-off invocations:

crabbox warmup --provider semaphore --semaphore-host example-org.semaphoreci.com --semaphore-project my-app
crabbox run --provider semaphore --semaphore-machine f1-standard-4 -- pnpm test
crabbox ssh --provider semaphore --id <slug>
crabbox status --provider semaphore --id <slug>
crabbox stop --provider semaphore <slug>

Behavior

  • warmup creates a job, waits until it is reachable over SSH, and records a local Crabbox claim.
  • run acquires (or resolves) a job, syncs the current Git manifest over SSH, and runs the command through Crabbox's standard SSH executor.
  • ssh prints the SSH command for the resolved job.
  • status/list report Crabbox-managed running jobs, mapped to local claims where available.
  • stop posts the Semaphore job-stop request, then removes the local claim and stored SSH key.
  • The idle timeout is enforced by the job's keepalive sleep, not by Crabbox; Touch/heartbeat is a no-op for this provider.

Boundaries

  • Linux only.
  • No Crabbox coordinator (broker); Semaphore API auth is local/provider-native.
  • No VNC, desktop, browser, code-server, or Actions hydration.
  • --class does not map cleanly to Semaphore machines — set the machine with --semaphore-machine or semaphore.machine.
  • --type is ignored; use the Semaphore-specific machine field instead.
  • --checksum works, since Semaphore exposes a real SSH target.

Troubleshooting

  • semaphore provider requires semaphore.host ... and semaphore.token: set the host and token via config or the env vars above.
  • semaphore.project is required: set semaphore.project, --semaphore-project, or CRABBOX_SEMAPHORE_PROJECT.
  • semaphore host ... must be a host name, not an API URL: pass the bare host, not a https://.../api/... URL.
  • project "..." not found: the project name does not match any project the token can access.
  • returned 401: the token is wrong for the host, expired, or lacks access to the project.
  • job ... did not reach RUNNING state within timeout / no SSH endpoint: the Semaphore agent has not attached debug SSH metadata yet; retry once capacity is available.
  • invalid semaphore idle timeout: use Go duration syntax such as 30m, 1h, or 90m.

Related docs: