eisenberg-deploy

name	eisenberg-deploy
description	Full deploy pipeline — test, typecheck, deploy to HAOS, smoke test

Run the full deployment pipeline. Abort on any failure — never continue past a red gate.

Why this is tar-based, not per-file SSH

Earlier versions of this skill did for f in *.py; do ssh ... < $f; done. That pattern silently dropped files: when ssh's ControlMaster reuses a single TCP connection across iterations, stdin pipes can interleave or close early, and some docker exec -i writes complete with truncated content. The loop exits 0 on every iteration so the failure is invisible until HA throws ImportError on restart.

The fix: stream the whole tree through a single SSH connection with tar, then verify every deployed file's sha256 against the local copy. One connection, atomic extract, explicit verification — no silent drops.

Why we restart twice

The HA Supervisor's component loader calls pip install pyeisenberg>=X.Y.Z on every integration startup, even when the requirement is already satisfied — pip re-writes the package files from its wheel cache (PyPI's published version), overwriting our manually-deployed development code. So a single restart leaves the container running the previous PyPI build, not the dev copy.

The robust dance: restart once (let HA finish its pip-install settle), re-deploy the library on top (now nothing else will touch it), then restart again so HA picks up the dev code. Two restarts, one source of truth.

Steps

1. Test suite: pytest tests/ -x -q — abort if any test fails.

2. Type check: pyright eisenberg/ custom_components/ — abort on any new error.

3. Lint: ruff check eisenberg/ tests/ custom_components/ — abort on errors.

4. Resolve container library path (Python version changes across HA updates):

   PY_DIR=$(ssh root@homeassistant -p 22222 \
     "docker exec homeassistant python3 -c 'import eisenberg, os; print(os.path.dirname(eisenberg.__file__))'")
   echo "Container eisenberg path: $PY_DIR"
   

   If this fails because the package doesn't exist yet (first deploy), create it:

   ssh root@homeassistant -p 22222 \
     "docker exec homeassistant sh -c 'mkdir -p /usr/local/lib/python*/site-packages/eisenberg'"
   

   Then re-run the resolve step.

5. Deploy integration files (HAOS host filesystem — safe from HA's pip, so we do these once before the first restart):

   ssh root@homeassistant -p 22222 \
     "mkdir -p /mnt/data/supervisor/homeassistant/custom_components/eisenberg/translations"
   tar -cf - -C custom_components/eisenberg . | \
     ssh root@homeassistant -p 22222 \
       "tar -xf - -C /mnt/data/supervisor/homeassistant/custom_components/eisenberg/"
   

6. First restart so HA picks up the new manifest and lets pip-install settle on its own copy of the library before we overwrite it:

   ssh root@homeassistant -p 22222 "ha core restart"
   until ssh root@homeassistant -p 22222 "ha core info" >/dev/null 2>&1; do sleep 3; done
   

7. Deploy client library (single SSH, tar pipe — HA's pip step is done so nothing else will touch these files now):

   tar -cf - -C eisenberg . | \
     ssh root@homeassistant -p 22222 \
       "docker exec -i homeassistant tar -xf - -C $PY_DIR/"
   

8. Verify checksums — catches both tar-pipe corruption and any race with pip's reinstall:

   set -e
   # Integration
   for f in custom_components/eisenberg/*.py custom_components/eisenberg/manifest.json; do
     name=$(basename "$f")
     local=$(sha256sum "$f" | awk '{print $1}')
     remote=$(ssh root@homeassistant -p 22222 \
       "sha256sum /mnt/data/supervisor/homeassistant/custom_components/eisenberg/$name | awk '{print \$1}'")
     [ "$local" = "$remote" ] || { echo "MISMATCH: $name"; exit 1; }
   done
   # Library
   for f in eisenberg/*.py; do
     name=$(basename "$f")
     local=$(sha256sum "$f" | awk '{print $1}')
     remote=$(ssh root@homeassistant -p 22222 \
       "docker exec homeassistant sha256sum $PY_DIR/$name | awk '{print \$1}'")
     [ "$local" = "$remote" ] || { echo "MISMATCH: $name"; exit 1; }
   done
   echo "All checksums match"
   

   If any mismatch: abort and investigate. A mismatch on the library after the first restart usually means HA's pip step hadn't actually finished when ha core info returned ready — wait longer and re-deploy.

9. Second restart so HA loads the dev library now sitting on disk:

   ssh root@homeassistant -p 22222 "ha core restart"
   until ssh root@homeassistant -p 22222 "ha core info" >/dev/null 2>&1; do sleep 3; done
   

10. Log check — explicitly grep for ERROR lines AFTER the second restart:

    ssh root@homeassistant -p 22222 \
      "docker logs homeassistant --since 2m 2>&1 | grep -iE 'eisenberg.*(error|traceback|importerror)' | tail -30"
    

    Empty output = clean. Any line shown = investigate before declaring success.

On failure at any step: stop, show full error, do NOT proceed or auto-fix. On success: report all gates passed.

Deploy does NOT release. Releasing (PyPI + GitHub tag) is a separate manual step.