bug(schema): only write network-config if instance dir present (#4635)

In Local boot stage (pre-systemnetworking) only write
/var/lib/cloud/instance/network-config.json if the symlink
/v/l/c/instance already exits. This instance symlink is created only
once a datasource is discovered in either Local or Network boot
stage.

This bug affects any environment where the detected datasource is
only discovered in the Network boot stage.
If no viable datasource is detected during the Local boot stage,
cloud-init calls apply_network_config to minimally setup a basic
fallback network config with DHCP on the primary NIC.

In this case, cloud-init will no longer attempt to write the
fallback network-config.json during Local boot. Defer writing the
networ-config.json to Network boot stage once
/var/lib/cloud/instance exists for a detected datasource.

Fixes GH-4630

cloudinit/stages.py

@@ -936,8 +936,9 @@ def _find_networking_config(self):
     def _apply_netcfg_names(self, netcfg):
         ncfg_instance_path = self.paths.get_ipath_cur("network_config")
         network_link = self.paths.get_runpath("network_config")
-        if not self._network_already_configured():
+        if os.path.islink(self.paths.instance_link):
             if os.path.exists(ncfg_instance_path):
+                # Compare write on delta of current network-config
                 if netcfg != util.load_json(
                     util.load_file(ncfg_instance_path)
                 ):
@@ -948,8 +949,8 @@ def _apply_netcfg_names(self, netcfg):
                 atomic_helper.write_json(
                     ncfg_instance_path, netcfg, mode=0o600
                 )
-        if not os.path.islink(network_link):
-            util.sym_link(ncfg_instance_path, network_link)
+            if not os.path.islink(network_link):
+                util.sym_link(ncfg_instance_path, network_link)
         try:
             LOG.debug("applying net config names for %s", netcfg)
             self.distro.networking.apply_network_config_names(netcfg)

tests/unittests/test_stages.py

@@ -10,7 +10,7 @@
 from cloudinit import sources, stages
 from cloudinit.event import EventScope, EventType
 from cloudinit.sources import NetworkConfigSource
-from cloudinit.util import write_file
+from cloudinit.util import sym_link, write_file
 from tests.unittests.helpers import mock
 from tests.unittests.util import TEST_INSTANCE_ID, FakeDataSource
 
@@ -28,7 +28,8 @@ def setup(self, tmpdir):
                 "paths": {"cloud_dir": self.tmpdir, "run_dir": self.tmpdir},
             }
         }
-        tmpdir.mkdir("instance")
+        tmpdir.mkdir("instance-uuid")
+        sym_link(tmpdir.join("instance-uuid"), tmpdir.join("instance"))
         self.init.datasource = FakeDataSource(paths=self.init.paths)
         self._real_is_new_instance = self.init.is_new_instance
         self.init.is_new_instance = mock.Mock(return_value=True)
@@ -393,9 +394,12 @@ def fake_network_config():
         assert caplog.records[0].levelname == "INFO"
         assert f"network config is disabled by {disable_file}" in caplog.text
 
+    @pytest.mark.parametrize("instance_dir_present", (True, False))
     @mock.patch("cloudinit.net.get_interfaces_by_mac")
     @mock.patch("cloudinit.distros.ubuntu.Distro")
-    def test_apply_network_on_new_instance(self, m_ubuntu, m_macs):
+    def test_apply_network_on_new_instance(
+        self, m_ubuntu, m_macs, instance_dir_present
+    ):
         """Call distro apply_network_config methods on is_new_instance."""
         net_cfg = {
             "version": 1,
@@ -415,20 +419,30 @@ def fake_network_config():
         m_macs.return_value = {"42:42:42:42:42:42": "eth9"}
 
         self.init._find_networking_config = fake_network_config
+        if not instance_dir_present:
+            self.tmpdir.join("instance").remove()
+            self.tmpdir.join("instance-uuid").remove()
 
         self.init.apply_network_config(True)
         networking = self.init.distro.networking
         networking.apply_network_config_names.assert_called_with(net_cfg)
         self.init.distro.apply_network_config.assert_called_with(
             net_cfg, bring_up=True
         )
-        assert net_cfg == json.loads(
-            self.tmpdir.join("instance/network-config.json").read()
-        )
-        assert net_cfg == json.loads(
-            self.tmpdir.join("network-config.json").read()
-        )
-        assert os.path.islink(self.tmpdir.join("network-config.json"))
+        if instance_dir_present:
+            assert net_cfg == json.loads(
+                self.tmpdir.join("instance/network-config.json").read()
+            )
+            assert net_cfg == json.loads(
+                self.tmpdir.join("network-config.json").read()
+            )
+            assert os.path.islink(self.tmpdir.join("network-config.json"))
+        else:
+            for path in (
+                "instance/network-config.json",
+                "network-config.json",
+            ):
+                assert not self.tmpdir.join(path).exists()
 
     @mock.patch("cloudinit.distros.ubuntu.Distro")
     def test_apply_network_on_same_instance_id(self, m_ubuntu, caplog):