fix(profiles): a condition-gated START is not an apply failure (no spurious rollback)

A unit whose systemd start-condition fails (e.g. hexo) can never go active, so
enable --now succeeds but wait_state timed out → apply of any profile listing such
a module rolled back. condition_failed() detects ConditionResult=no and _do_change/
_rollback treat that START as reached (the module is deliberately off here). Found
applying the full profile from the panel.

Co-Authored-By: Gerald KERMA <devel@cybermind.fr>

packages/secubox-profiles/api/actuate.py

@@ -190,3 +190,19 @@ def wait_state(m: Manifest, want_on: bool, *, observe=_observe,
         if now() - start >= timeout:
             return False
         sleep(poll)
+
+
+def condition_failed(m: Manifest, run) -> bool:
+    """True si l'unité systemd du module a une Condition= de démarrage qui
+    ÉCHOUE : `systemctl enable --now` réussit, mais systemd laisse alors l'unité
+    inactive VOLONTAIREMENT (le module ne doit pas tourner dans cet
+    environnement — arch, présence d'un chemin, etc.). Ce n'est PAS un échec
+    d'actionnement : attendre `active` sur une telle unité timeout à l'infini.
+    Ne concerne que les modules natifs (LXC/portail n'ont pas de Condition=)."""
+    if m.runtime != "native":
+        return False
+    for u in m.units:
+        rc, out = run(["systemctl", "show", u, "-p", "ConditionResult", "--value"])
+        if rc == 0 and out.strip() == "no":
+            return True
+    return False

packages/secubox-profiles/api/apply.py

@@ -17,7 +17,7 @@ from dataclasses import dataclass, field
 
 from . import audit as _audit
 from . import snapshot as _snapshot
-from .actuate import ActuationError, actuate, wait_state
+from .actuate import ActuationError, actuate, condition_failed, wait_state
 from .diff import START, STOP, Change
 from .manifest import Manifest
 
@@ -41,6 +41,12 @@ def _want_on(action: str) -> bool:
 def _do_change(c: Change, m: Manifest, *, run, observe, routes_value, sleep, now,
                wait_timeout) -> None:
     actuate(c, m, run=run, route_value=routes_value)
+    # Un START d'une unité dont la Condition= de démarrage échoue ne deviendra
+    # JAMAIS active (enable --now a réussi, systemd la laisse inactive par
+    # design). On ne l'attend pas (ça timeout) et on ne la traite PAS comme un
+    # échec : le module est délibérément off ici.
+    if c.action == START and condition_failed(m, run):
+        return
     if not wait_state(m, _want_on(c.action), observe=observe, sleep=sleep, now=now,
                       timeout=wait_timeout):
         raise ActuationError(f"{c.id}: état non atteint (timeout)")
@@ -109,8 +115,12 @@ def _rollback_applied(applied, manifests, snap, *, run, observe, sleep, clock,
             continue
         try:
             actuate(rev, m, run=run, route_value=rv)
-            if wait_state(m, want_on, observe=observe, sleep=sleep, now=clock,
-                         timeout=wait_timeout):
+            # Même règle que le forward : re-démarrer une unité condition-gated
+            # ne la rend jamais active — c'est un succès, pas un timeout.
+            reached = (rev.action == START and condition_failed(m, run)) or \
+                wait_state(m, want_on, observe=observe, sleep=sleep, now=clock,
+                           timeout=wait_timeout)
+            if reached:
                 _audit.record({"ts": now, "module": c.id, "action": rev.action,
                                "result": "rollback"}, path=audit_path)
                 rolled.append(c.id)

packages/secubox-profiles/debian/changelog

@@ -1,3 +1,13 @@
+secubox-profiles (0.6.2-1~bookworm1) bookworm; urgency=medium
+
+  * Fix apply on condition-gated modules: a unit whose systemd start-condition
+    fails (e.g. hexo — ConditionResult=no) can never become active, so `enable
+    --now` succeeds but wait_state timed out → the whole apply spuriously
+    rolled back. Now a START of a condition-failed native unit is treated as
+    reached (the module is deliberately off in this environment), not a failure.
+
+ -- Gerald KERMA <devel@cybermind.fr>  Sun, 19 Jul 2026 14:00:00 +0200
+
 secubox-profiles (0.6.1-1~bookworm1) bookworm; urgency=medium
 
   * Fix apply/rollback from the panel: the service runs ProtectSystem=strict, so a

packages/secubox-profiles/tests/test_actuate.py

@@ -140,3 +140,26 @@ def test_wait_state_times_out():
                     sleep=lambda s: None, now=iter([0, 10, 20, 31]).__next__,
                     timeout=30, poll=10)
     assert ok is False
+
+
+def test_condition_failed_detects_gated_native_unit():
+    from api.actuate import condition_failed
+
+    def run(argv):
+        if argv[:2] == ["systemctl", "show"] and "ConditionResult" in argv:
+            return 0, "no\n"   # systemd start-condition failed
+        return 0, ""
+    assert condition_failed(_m("hexo"), run) is True
+
+
+def test_condition_failed_false_when_condition_ok():
+    from api.actuate import condition_failed
+    assert condition_failed(_m("x"), lambda a: (0, "yes\n")) is False
+
+
+def test_condition_failed_false_for_lxc():
+    from api.actuate import condition_failed
+    # LXC/portal modules have no systemd start-condition; never gated.
+    called = []
+    condition_failed(_m("l", runtime="lxc", lxc="l"), lambda a: (called.append(a), (0, "no"))[1])
+    assert not called  # short-circuits on runtime != native

packages/secubox-profiles/tests/test_apply.py

@@ -176,3 +176,24 @@ def test_rollback_to_restores_snapshot_state(tmp_path):
                           audit_path=tmp_path / "audit-dry.log", apply=False)
     assert dry_rep.status == "planned"
     assert dry_calls == []
+
+
+def test_start_of_condition_gated_native_module_is_not_a_failure(tmp_path):
+    # hexo-like: START issued, but its systemd start-condition fails so it never
+    # goes active. enable --now succeeded → not an actuation failure → no rollback.
+    calls = []
+    manifests = {"hexo": _m("hexo")}
+    plan = [Change("hexo", START, "", 50)]
+
+    def run(argv):
+        calls.append(argv)
+        if argv[:2] == ["systemctl", "show"] and "ConditionResult" in argv:
+            return 0, "no\n"
+        return 0, ""
+
+    rep = apply_plan(plan, manifests, {"hexo": Actual(enabled=False, active=False)},
+                     run=run, observe=lambda m: Actual(enabled=True, active=False),
+                     now="t", routes={}, snap_root=tmp_path,
+                     audit_path=tmp_path / "audit.log", apply=True, wait_timeout=0)
+    assert rep.status == "applied"          # NOT rolled_back
+    assert rep.changed == ["hexo"] and rep.failed == []