fix: wait for cloud-init.service to fully activate

lib/daemon.py

@@ -19,8 +19,19 @@ def _wait_for_cloud_config():
     LOG.debug("waiting for cloud-config.service to finish")
     for i in range(WAIT_FOR_CLOUD_CONFIG_POLL_TIMES + 1):
         state = system.get_systemd_unit_active_state("cloud-config.service")
+        ci_state = system.get_systemd_unit_active_state("cloud-init.service")
         LOG.debug("cloud-config.service state: %r", state)
-        if state is not None and state == "activating":
+        LOG.debug("cloud-init.service state: %r", ci_state)
+        # if cloud-config.service is not yet activating but cloud-init is
+        # running, wait for cloud-config to start
+        if state is not None and (
+            state == "activating"
+            or (
+                state == "inactive"
+                and ci_state is not None
+                and (ci_state == "activating" or ci_state == "active")
+            )
+        ):
             if i < WAIT_FOR_CLOUD_CONFIG_POLL_TIMES:
                 LOG.debug(
                     "cloud-config.service is activating. "

uaclient/tests/test_lib_daemon.py

@@ -17,24 +17,41 @@ class TestWaitForCloudConfig:
         (
             # not activating
             (
-                ["active"],
+                ["active"] * 2,
                 [],
             ),
+            # inactive (all cloud-init)
             (
-                ["inactive"],
+                ["inactive"] * 2,
                 [],
             ),
             (
-                [None],
+                [None] * 2,
                 [],
             ),
-            # activating, then finishes
+            # cloud-config activating, then finishes
+            # cloud-init is active
             (
-                (["activating"] * 11) + ["active"],
+                (["activating", "active"] * 11) + ["active"] * 2,
                 [mock.call(WAIT_FOR_CLOUD_CONFIG_SLEEP_TIME)] * 11,
             ),
             (
-                (["activating"] * 11) + ["failed"],
+                (["activating", "active"] * 11) + ["failed"] + ["active"],
+                [mock.call(WAIT_FOR_CLOUD_CONFIG_SLEEP_TIME)] * 11,
+            ),
+            # inactive cloud-config, active cloud-init
+            (
+                (["inactive", "active"] * 6)
+                + (["activating", "active"] * 5)
+                + ["active"] * 2,
+                [mock.call(WAIT_FOR_CLOUD_CONFIG_SLEEP_TIME)] * 11,
+            ),
+            # inactive cloud-config, activating cloud-init
+            (
+                (["inactive", "activating"] * 2)
+                + (["inactive", "active"] * 4)
+                + (["activating", "active"] * 5)
+                + ["active"] * 2,
                 [mock.call(WAIT_FOR_CLOUD_CONFIG_SLEEP_TIME)] * 11,
             ),
             # still activating after polling maximum times