Skip to content

feat(LinuxTentacleE2E): Phase 12.M.L.A.1 — Section A bootstrap + bogus-version exit-1

Placeholder ppxd requested to merge phase12.M.L.A.1-install-script-bogus-version into main

Summary

PIVOT to Section A install-tentacle.sh after the upgrade-flow phase closed (J.L.E.7–19, 13 PRs, 20 E2E tests, 1 production bug fixed). The user's original goal is "from install to deploy to lifecycle to upgrade — all functions confident" — install was the largest unfilled gap.

This is the smallest viable Section A scope: bogus version → exit 1. No successful install means no symlinks, no service install, no sudoers — minimal pollution surface, minimal cleanup, but real coverage of the .sh's argument parsing + URL construction + curl error handling.

Why bogus-version first (not happy path)

Concern Bogus version Happy path
Pollution surface INSTALL_DIR only (failed mkdir, no other side effects) /etc/squid-tentacle, /var/lib/squid-tentacle, /usr/local/bin, sudoers, systemd unit, system user, apt sources
Cleanup complexity Trivial Significant matrix
Caught regression class arg parsing, URL construction, curl errors, exit ordering All of bogus-version + extract + chmod + symlink + service install + ownership + APT repo persistence
Risk on first PR Low Higher (each cleanup gap is a leak between tests)

The fixture (LinuxInstallScriptContext) already implements full cleanup defensively (covering every production path the script writes to), so happy-path tests in J.M.L.A.2+ drop in additively without re-engineering.

Test mechanism

Configure LocalReleaseMirror to 404 BOTH the plain-version AND the v-prefixed download URLs (.sh tries both per lines 234-241). No tarball staged → both URL forms 404. .sh's download_ok wrapper retries 3× per URL; both URLs fail → exit 1 with "Could not download" error.

Assertions

Assertion Catches
exitCode == 1 .sh exit-on-failure regression
stdout contains "Could not download" Operator-actionable diagnostic regression
stdout echoes the constructed URL Operator can verify --version typo by reading the log
INSTALL_DIR does NOT exist .sh's mkdir -p runs only after successful curl; presence = exit ordering regressed
/etc/squid-tentacle does NOT exist Post-install block never ran
/var/lib/squid-tentacle does NOT exist Same
/usr/local/bin/squid-tentacle symlink does NOT exist Symlink creation is post-extract

Infrastructure additions

  • LinuxInstallScriptContext fixture (sibling of LinuxLifecycleContext; structurally different env vars + cleanup matrix so kept separate per ISP / Rule 7)
  • LinuxTentacleE2ECategories.InstallScript constant
  • tentacle-linux-e2e.yml workflow filter updated to include the new category in default (per CLAUDE.md trait-filter convention)

Cleanup matrix (forward-compatible with happy path)

Dispose() runs sudo rm -rf against:

  • INSTALL_DIR (per-test temp dir)
  • /etc/squid-tentacle/
  • /var/lib/squid-tentacle/
  • /usr/local/bin/squid-tentacle
  • /etc/apt/sources.list.d/squid.list
  • /etc/apt/keyrings/squid.gpg
  • /etc/apt/apt.conf.d/99-squid-direct.conf
  • /etc/sudoers.d/squid-tentacle-upgrade
  • /etc/systemd/system/squid-tentacle.service

Each is best-effort; missing paths are no-ops.

Fidelity tier

🟢 High (Rule 12.4): drives the real production .sh against real bash + real curl + real LocalReleaseMirror returning real HTTP 404. No mocks at OS-resource layer.

Expected runtime: ~10s (curl retries 3× per URL × 2 URLs ~6s + apt-get update on first run ~3s; libicu typically pre-installed on ubuntu-latest so install_runtime_deps short-circuits).

Test plan

  • Linux E2E workflow runs Squid.LinuxTentacleE2ETests (manual workflow_dispatch after merge)
  • A2u1_BogusVersion_ExitsOneWithCleanErrorAndNoPartialInstall passes within ~15s
  • No regression on existing 20 Linux E2E tests
  • New trait Category=LinuxTentacleInstallScriptE2E in workflow filter picks up the new test class

🤖 Generated with Claude Code

Merge request reports