feat(samever): step back to older base when last-green canonical == head version

resolve_upgrade_base now reads the head's published version (abra.head_compose_version,
the coop-cloud.<stack>.version label) and, when the last-green warm-canonical version
equals it, steps back to the newest published version strictly older than head instead
of deploying a same-version no-op. warm_reconcile gains version_key + newest_older_version
(single coop-cloud ordering source; sort_versions refactored onto version_key, no behavior
change). Skip only when no older published predecessor exists. Step-back returns kind=version
so it inherits F1d-2 pinned-tag checkout. Extends tests/unit/test_upgrade_base.py (13 pass).

runner/harness/abra.py

@@ -11,6 +11,7 @@ from __future__ import annotations
 
 import json
 import os
+import re
 import subprocess
 
 ABRA = "abra"
@@ -34,6 +35,29 @@ def recipe_dir(recipe: str) -> str:
     return os.path.join(abra_dir(), "recipes", recipe)
 
 
+_VERSION_LABEL_RE = re.compile(r"coop-cloud\.[^.\s]*\.version=([^\s\"']+)")
+
+
+def head_compose_version(recipe: str) -> str | None:
+    """The published version of the recipe's on-disk compose.yml (the head checkout under test):
+    the value of the `coop-cloud.<stack>.version` label, e.g. "1.0.0+3.5.3". None if the file is
+    unreadable or carries no version label.
+
+    Used by the upgrade-base resolver (phase samever) to compare the head's declared version against
+    the last-green warm-canonical version: when they are equal, deploying the canonical as the base
+    would be a vacuous same-version no-op, so the resolver steps back to an older published version.
+    `${STACK_NAME}` is a literal in the file (abra interpolates it at deploy time, not on disk), so
+    the regex matches it as the stack segment without needing interpolation."""
+    path = os.path.join(recipe_dir(recipe), "compose.yml")
+    try:
+        with open(path) as f:
+            text = f.read()
+    except OSError:
+        return None
+    m = _VERSION_LABEL_RE.search(text)
+    return m.group(1).strip() if m else None
+
+
 def _run_pty(
     args: list[str], timeout: int = 900, check: bool = True
 ) -> subprocess.CompletedProcess:

runner/run_recipe_ci.py

@@ -42,6 +42,7 @@ from typing import NamedTuple
 
 ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 sys.path.insert(0, os.path.join(ROOT, "runner"))
+import warm_reconcile  # noqa: E402  (runner/ is on sys.path; owns coop-cloud version ordering)
 from harness import (  # noqa: E402
     abra,
     canonical,
@@ -108,11 +109,21 @@ class BasePlan(NamedTuple):
         return self.kind in ("version", "ref")
 
 
-def resolve_upgrade_base(stages, meta, recipe: str, head_ref: str | None = None) -> BasePlan:
+def resolve_upgrade_base(
+    stages, meta, recipe: str, head_ref: str | None = None, head_version: str | None = None
+) -> BasePlan:
     """Dynamic upgrade-base resolution (phase prevb, replaces the static `recipe_versions[-2]`
     default). Order: explicit override → last-green (warm canonical) → target-branch (main) tip →
     skip. EXPECTED_NA[upgrade] / upgrade∉stages short-circuit to a declared skip first.
 
+    `head_version` is the head checkout's published version (the `coop-cloud.<stack>.version` label;
+    see abra.head_compose_version). When the last-green warm-canonical version EQUALS it, deploying
+    the canonical as the base would be a vacuous same-version no-op, so the resolver STEPS BACK to the
+    newest published version strictly older than the head (phase samever) — the upgrade tier always
+    crosses a real version delta. This is the nightly STEADY STATE: a green cold-on-latest run promotes
+    canonical→latest, so the next night finds canonical == head and must step back. Skip only when no
+    older published predecessor exists.
+
     last-green is the PRIMARY base — the version cc-ci last recorded green for this recipe (the
     warm-canonical registry record). main-tip is the FALLBACK: the recipe repo's `main` HEAD, the
     real predecessor the PR merges on top of, used when there is no last-green. Else the tier is
@@ -135,11 +146,37 @@ def resolve_upgrade_base(stages, meta, recipe: str, head_ref: str | None = None)
         return BasePlan("version", override, None, "explicit UPGRADE_BASE_VERSION override")
     rec = canonical.read_registry(recipe)
     if rec and rec.get("version"):
+        canon = rec["version"]
+        same = head_version is not None and warm_reconcile.version_key(
+            canon
+        ) == warm_reconcile.version_key(head_version)
+        if not same:
+            # canonical ≠ head version (the common version-bump PR / nightly-with-new-version case):
+            # the green-verified primary base, unchanged from prevb.
+            return BasePlan(
+                "version",
+                canon,
+                None,
+                f"last-green (warm canonical, status={rec.get('status')})",
+            )
+        # canonical == head version → deploying it would be a same-version no-op. Step back to the
+        # newest published version strictly older than the head (phase samever).
+        older = warm_reconcile.newest_older_version(
+            warm_reconcile.recipe_tags(recipe), head_version
+        )
+        if older:
+            return BasePlan(
+                "version",
+                older,
+                None,
+                f"step-back: last-green canonical ({canon}) == head version {head_version}; "
+                f"newest older published base",
+            )
         return BasePlan(
-            "version",
+            "skip",
-            rec["version"],
             None,
-            f"last-green (warm canonical, status={rec.get('status')})",
+            None,
+            f"base == head ({head_version}) and no older published predecessor",
         )
     main_tip = lifecycle.recipe_branch_commit(recipe, "main")
     if main_tip and main_tip != head_ref:
@@ -983,7 +1020,10 @@ def main() -> int:
 
     domain = naming.app_domain(recipe, os.environ.get("PR", "0"), ref)
 
-    base_plan = resolve_upgrade_base(stages, meta, recipe, head_ref=head_ref)
+    head_version = abra.head_compose_version(recipe)
+    base_plan = resolve_upgrade_base(
+        stages, meta, recipe, head_ref=head_ref, head_version=head_version
+    )
     prev = base_plan.runs  # gates the upgrade tier
     # base deploy target: a pinned published version (kind=version) or main-tip commit (kind=ref);
     # on skip fall back to the run's VERSION/head (target=None → chaos head deploy, as before).

runner/warm_reconcile.py

@@ -145,14 +145,31 @@ def is_version_tag(tag: str) -> bool:
     return bool(_VER_RE.match(tag.strip()))
 
 
+def version_key(tag: str):
+    """Ordering key for a coop-cloud version tag: (recipe-semver tuple, app-version tuple). The single
+    source of version ordering — sort_versions / newest_older_version both key on this so the
+    upgrade-base step-back (phase samever) never hand-rolls semver comparison."""
+    recipe, _, app = tag.partition("+")
+    return (_numtuple(recipe), _numtuple(app))
+
+
 def sort_versions(tags) -> list[str]:
     """Sort coop-cloud version tags ascending by (recipe-semver tuple, app-version tuple)."""
+    return sorted([t for t in tags if is_version_tag(t)], key=version_key)
 
-    def key(t: str):
-        recipe, _, app = t.partition("+")
-        return (_numtuple(recipe), _numtuple(app))
 
-    return sorted([t for t in tags if is_version_tag(t)], key=key)
+def newest_older_version(tags, version: str | None) -> str | None:
+    """The newest published version tag STRICTLY OLDER than `version` (coop-cloud ordering), or None
+    when no such tag exists. Used by the upgrade-base resolver (phase samever): when the last-green
+    warm-canonical version equals the head version, the resolver steps back to this older base so the
+    upgrade tier always crosses a real version delta instead of deploying a same-version no-op.
+    "Strictly older" excludes any tag whose ordering key equals the head's (so a re-published or
+    differently-formatted equal version is never chosen)."""
+    if not version:
+        return None
+    target = version_key(version)
+    older = [t for t in sort_versions(tags) if version_key(t) < target]
+    return older[-1] if older else None
 
 
 def _numtuple(s: str) -> tuple:

tests/unit/test_upgrade_base.py

@@ -14,6 +14,7 @@ from types import SimpleNamespace
 
 sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "runner"))
 import run_recipe_ci  # noqa: E402
+import warm_reconcile  # noqa: E402
 from harness import canonical, lifecycle  # noqa: E402
 
 ALL = {"install", "upgrade", "backup", "restore", "custom"}
@@ -105,6 +106,90 @@ def test_no_canonical_no_main_skip(monkeypatch):
     assert plan.kind == "skip" and "no predecessor" in plan.reason
 
 
+# --- phase samever: step back to an older base when canonical == head version ---
+
+# A realistic keycloak tag list (coop-cloud "<recipe-semver>+<app-version>"), out of order on purpose
+# so the test exercises the ordering, not list position.
+KC_TAGS = ["10.6.0+26.5.0", "10.8.0+26.6.3", "10.7.1+26.6.2", "10.7.0+26.6.0", "not-a-version"]
+
+
+def test_canonical_equals_head_steps_back_to_newest_older(monkeypatch):
+    # nightly steady state: a green cold-on-latest run promoted canonical→latest, so the next night
+    # finds canonical == head version. The resolver must NOT deploy the same version — it steps back
+    # to the newest published version STRICTLY OLDER than the head.
+    head_v = "10.8.0+26.6.3"  # == canonical == latest tag
+    monkeypatch.setattr(canonical, "read_registry", lambda r: {"version": head_v, "status": "warm"})
+    monkeypatch.setattr(warm_reconcile, "recipe_tags", lambda r: KC_TAGS)
+    # main must never be consulted on the step-back (canonical path owns the decision)
+    monkeypatch.setattr(
+        lifecycle,
+        "recipe_branch_commit",
+        lambda r, b="main": (_ for _ in ()).throw(AssertionError("not consulted")),
+    )
+    plan = run_recipe_ci.resolve_upgrade_base(
+        ALL, _meta(), "keycloak", head_ref=HEAD, head_version=head_v
+    )
+    assert plan.kind == "version" and plan.runs
+    assert plan.version == "10.7.1+26.6.2"  # newest tag strictly older than 10.8.0+26.6.3
+    # strictly older than head (the load-bearing invariant)
+    assert warm_reconcile.version_key(plan.version) < warm_reconcile.version_key(head_v)
+    assert "step-back" in plan.reason
+
+
+def test_canonical_differs_from_head_uses_canonical_unchanged(monkeypatch):
+    # the common version-bump case: canonical (older) ≠ head version → primary base unchanged; the
+    # step-back never triggers, recipe_tags never consulted.
+    monkeypatch.setattr(
+        canonical, "read_registry", lambda r: {"version": "10.7.1+26.6.2", "status": "warm"}
+    )
+    monkeypatch.setattr(
+        warm_reconcile,
+        "recipe_tags",
+        lambda r: (_ for _ in ()).throw(AssertionError("step-back not taken")),
+    )
+    plan = run_recipe_ci.resolve_upgrade_base(
+        ALL, _meta(), "keycloak", head_ref=HEAD, head_version="10.8.0+26.6.3"
+    )
+    assert plan.kind == "version" and plan.version == "10.7.1+26.6.2"
+    assert "last-green" in plan.reason
+
+
+def test_canonical_equals_head_no_older_published_skips(monkeypatch):
+    # canonical == head and it is the ONLY (oldest) published version → genuinely no predecessor →
+    # declared skip with the samever reason (never a same-version no-op).
+    head_v = "1.0.0+3.5.3"
+    monkeypatch.setattr(canonical, "read_registry", lambda r: {"version": head_v, "status": "warm"})
+    monkeypatch.setattr(warm_reconcile, "recipe_tags", lambda r: [head_v])  # only the head version
+    plan = run_recipe_ci.resolve_upgrade_base(
+        ALL, _meta(), "discourse", head_ref=HEAD, head_version=head_v
+    )
+    assert plan.kind == "skip" and not plan.runs
+    assert "no older published predecessor" in plan.reason
+
+
+def test_no_head_version_preserves_canonical_primary(monkeypatch):
+    # head_version unreadable (None) → cannot compare → preserve prevb behavior: canonical is primary,
+    # never a step-back (no regression for callers that don't pass head_version).
+    monkeypatch.setattr(
+        canonical, "read_registry", lambda r: {"version": "10.8.0+26.6.3", "status": "warm"}
+    )
+    monkeypatch.setattr(
+        warm_reconcile,
+        "recipe_tags",
+        lambda r: (_ for _ in ()).throw(AssertionError("step-back not taken")),
+    )
+    plan = run_recipe_ci.resolve_upgrade_base(ALL, _meta(), "keycloak", head_ref=HEAD)
+    assert plan.kind == "version" and plan.version == "10.8.0+26.6.3"
+
+
+def test_newest_older_version_ordering():
+    # the ordering helper picks the correct strictly-older tag and excludes the equal one.
+    assert warm_reconcile.newest_older_version(KC_TAGS, "10.8.0+26.6.3") == "10.7.1+26.6.2"
+    assert warm_reconcile.newest_older_version(KC_TAGS, "10.7.0+26.6.0") == "10.6.0+26.5.0"
+    assert warm_reconcile.newest_older_version(KC_TAGS, "10.6.0+26.5.0") is None  # oldest → none
+    assert warm_reconcile.newest_older_version(KC_TAGS, None) is None
+
+
 def test_expected_na_other_rung_does_not_suppress_upgrade(monkeypatch):
     # an EXPECTED_NA for a DIFFERENT rung (backup_restore) must NOT short-circuit the upgrade base —
     # resolution proceeds to last-green/main-tip (custom-html-tiny shape).