werkregie/doc/prs/2026-10-14-PR#70-deploy-launcher-check.md

PR#70: The Deploy Test Runs the Bundle's Own Launcher

WARNING: This document describes only the change applied in this PR. It may already be outdated once the next PR is merged. Historic PR documentation is not maintained along with new PRs; treat it as a snapshot.

Related Links

• werkregie PR#69: Werkator deploys a green main to mih09; merged with werkrevisor's ORANGE finding "Launcher version handling untested" still open.

Problem

The receiver of PR#69 asks the new launcher for its version with bin/werkregie-dashboard version. Its tests used launcher stubs only, so no test showed that the bundle's own launcher answers this command. A launcher that ignores its argument would start the dashboard in the mode local and never end, so the deploy would hang.

Non-Goals

• No change to what the deploy does when the launcher answers.

Specification

Feature: the receiver checks the bundle's own launcher

Scenario#70.01: the bundle's own launcher prints its version

… so that a green main is deployed and not refused by the launcher check.

• Given a bundle with the launcher of src/bundle/bin
• When the receiver installs it
• Then the launcher prints the version of Main, and the service runs with it

Verified by

• DeployReceiveTest: a bundle replaces the runtime and the service answers with its version

Scenario#70.02: a launcher that prints no version leaves the service running

… so that a broken launcher costs no downtime.

• Given a bundle whose launcher prints no version, or does not end within 60 s
• When the receiver checks it
• Then the service keeps running unchanged, and the build deploy fails

Verified by

• DeployReceiveTest: a bundle whose launcher prints no version leaves the service untouched
• DeployReceiveTest: a bundle whose launcher does not end leaves the service untouched

Solution

• The bundle of the successful deploy in DeployReceiveTest has the real launcher of src/bundle/bin and runs the real Main, with the test's JVM and classpath in place of the jlinked JRE and the jars. Without "$@" in the launcher, this test fails.
• New tests give a launcher that prints no version and one that does not end.
• The receiver gives the launcher 60 s, WERKREGIE_LAUNCHER_SECONDS; the test of the launcher that does not end sets 2 s.
• The test isolates a dashboard that a broken launcher starts by mistake: its configuration names a stateDir of its own, since the JVM takes user.home from passwd and not from HOME. Before that, a red run of the test wrote to the journal of the user on think for some minutes.
• The README names the two failures instead of "both cases".
• Version 0.43.2; PR#71 took 0.43.1 first.

Open Questions

None.

Additional Changes

• werkregie-deploy calls ssh -T: the first deploy, 0.43.1 by main@deploy, logged "Pseudo-terminal will not be allocated". That deploy installed 0.43.1 and reported it running, so the plan sets #ops.deploy to [^].