test(haos): retry initial Supervisor readiness

Author: kingpanther13

tests/haos_image_build/build_image.py

@@ -1800,30 +1800,16 @@ def _wait_supervisor_ready(
     operation. Beta image configuration uses the same predicate after channel
     selection and update.
     """
-    info = ws.supervisor_api("/supervisor/info", method="get", timeout=30.0)
-    LOG.info(
-        "Supervisor ready: version=%s version_latest=%s arch=%s",
-        info.get("version"),
-        info.get("version_latest"),
-        info.get("arch"),
-    )
-    if _supervisor_info_ready(
-        info,
-        expected_channel=expected_channel,
-        minimum_version=minimum_version,
-    ):
-        return info
-
-    LOG.info(
-        "Supervisor self-update pending (%s -> %s); waiting before store ops...",
-        info.get("version"),
-        info.get("version_latest"),
-    )
     deadline = time.monotonic() + update_timeout
-    last_version = info.get("version")
+    info: dict[str, Any] = {}
+    last_version: object = None
     last_error: BaseException | None = None
-    while time.monotonic() < deadline:
-        time.sleep(10.0)
+    first_probe = True
+    first_success = True
+    while first_probe or time.monotonic() < deadline:
+        if not first_probe:
+            time.sleep(10.0)
+        first_probe = False
         try:
             info = ws.supervisor_api("/supervisor/info", method="get", timeout=30.0)
         except _SUPERVISOR_WAIT_TRANSIENT_ERRORS as e:
@@ -1845,14 +1831,30 @@ def _wait_supervisor_ready(
                 last_error = reconnect_err
             continue
         version = info.get("version")
-        if version != last_version:
-            LOG.info("Supervisor version changed: %s -> %s", last_version, version)
-            last_version = version
-        if _supervisor_info_ready(
+        ready = _supervisor_info_ready(
             info,
             expected_channel=expected_channel,
             minimum_version=minimum_version,
-        ):
+        )
+        if first_success:
+            LOG.info(
+                "Supervisor ready: version=%s version_latest=%s arch=%s",
+                info.get("version"),
+                info.get("version_latest"),
+                info.get("arch"),
+            )
+            first_success = False
+            if ready:
+                return info
+            LOG.info(
+                "Supervisor self-update pending (%s -> %s); waiting before store ops...",
+                info.get("version"),
+                info.get("version_latest"),
+            )
+        elif version != last_version:
+            LOG.info("Supervisor version changed: %s -> %s", last_version, version)
+        last_version = version
+        if ready:
             LOG.info("Supervisor self-update complete: version=%s", version)
             return info
     last_err_suffix = f"; last error: {last_error!r}" if last_error else ""

tests/src/unit/test_haos_supervisor_wait.py

@@ -263,6 +263,27 @@ def test_returns_immediately_when_up_to_date() -> None:
     sleep.assert_not_called()
 
 
+def test_tolerates_setup_state_from_initial_probe() -> None:
+    """A setup-state error on the first readiness probe is retried."""
+    ws = Mock()
+    ws.supervisor_api.side_effect = [
+        WSCommandError(
+            "not ready",
+            code="unknown_error",
+            supervisor_message="System is not ready with state: setup",
+        ),
+        _info(update_available=False),
+    ]
+
+    with patch("tests.haos_image_build.build_image.time.sleep") as sleep:
+        result = _wait_supervisor_ready(ws)
+
+    assert result["version"] == "2026.06.1"
+    assert ws.supervisor_api.call_count == 2
+    sleep.assert_called_once_with(10.0)
+    ws.reconnect.assert_called_once_with()
+
+
 def test_waits_until_update_clears() -> None:
     """Polls /supervisor/info until update_available flips False."""
     ws = Mock()
@@ -318,9 +339,8 @@ def test_raises_on_update_timeout() -> None:
 def test_persistent_error_surfaced_in_timeout() -> None:
     """Persistent WSCommandError -> timeout message includes last error."""
     ws = Mock()
-    # Initial read sees a pending update; the poll then hits a persistent
-    # WSCommandError until the deadline (the initial read is outside the
-    # tolerant loop, so it must succeed for the loop to be exercised).
+    # Initial read sees a pending update; later probes hit a transient
+    # WSCommandError until the deadline.
     ws.supervisor_api.side_effect = [
         _info(update_available=True, version="2026.05.1"),
         WSCommandError(