fix(jobs): a recovered migration restores autostart and removes the source NVRAM file

migrate_instance sets autostart on the destination (it is per host and
the migration does not carry it) and deletes the NVRAM file libvirt
keeps on the source. A worker that stopped after libvirt moved the VM
skipped both: recovery only recorded the move, so an autostart VM no
longer started after the destination rebooted, and a later VM of the
same name on the source started with these UEFI variables. The job now
records both before the move, and recovery does them after a proven
move; a step that fails is a warning of the job's result.

instances/jobs.py

@@ -930,11 +930,15 @@ def migration_target(instance, target_pk):
     return target
 
 
+def _connect(compute_pk):
+    c = Compute.objects.get(pk=compute_pk)
+    return wvmConnect(c.hostname, c.login, c.password, c.type)
+
+
 def _domain(compute_pk, uuid):
     """The VM on the compute, or None when it has none there."""
-    c = Compute.objects.get(pk=compute_pk)
     try:
-        return wvmConnect(c.hostname, c.login, c.password, c.type).wvm.lookupByUUIDString(uuid)
+        return _connect(compute_pk).wvm.lookupByUUIDString(uuid)
     except libvirtError as err:
         if err.get_error_code() != VIR_ERR_NO_DOMAIN:
             raise  # unreachable: recovery waits and retries
@@ -947,6 +951,25 @@ def _shut_off_again(op, dom):
         dom.destroy()
 
 
+def _moved(op, dest):
+    """What migrate_instance does once libvirt moved the VM, for a job that
+    stopped before it: autostart is per host, and the source keeps the
+    NVRAM file. The job's result, with a warning for a step that failed."""
+    notes = []
+    if op.intent.get("autostart"):
+        try:
+            dest.setAutostart(1)
+        except libvirtError as err:
+            notes += say(_("The VM migrated, but autostart could not be set on the destination: %(error)s"),
+                         error=err)
+    if nvram := op.intent.get("nvram"):
+        try:
+            _connect(op.intent["compute"]).remove_nvram(nvram, _connect(op.params["target"]))
+        except libvirtError as err:
+            notes += say(_("The VM migrated, but its NVRAM file stays on the source: %(error)s"), error=err)
+    return {"warning": notes} if notes else {}
+
+
 @register
 class MigrateInstance(Kind):
     name = "instance.migrate"
@@ -993,8 +1016,10 @@ class MigrateInstance(Kind):
     def run(self, op, ctx):
         p = op.params
         instance = _instance(p["instance"])
-        # a shut-off VM may migrate started paused (migrate_running): recovery shuts it off again
-        ctx.record_intent(shut_off=instance.proxy.get_status() == 5)
+        # a shut-off VM may migrate started paused (migrate_running): recovery shuts it off again;
+        # it also does what migrate_instance does after the move (_moved)
+        proxy = instance.proxy
+        ctx.record_intent(shut_off=proxy.get_status() == 5, autostart=proxy.get_autostart(), nvram=proxy.get_nvram())
         ctx.set_phase(gettext_noop("Migrating the VM"))
         try:
             utils.migrate_instance(Compute.objects.get(pk=p["target"]), instance, op.user,
@@ -1048,7 +1073,7 @@ class MigrateInstance(Kind):
             if instance is not None:  # the job stopped before it moved the record
                 utils.record_migration(instance, Compute.objects.get(pk=op.params["target"]))
             _shut_off_again(op, dest)
-            return Outcome(Operation.SUCCEEDED)
+            return Outcome(Operation.SUCCEEDED, result=_moved(op, dest))
         if source is not None and dest is None:
             _shut_off_again(op, source)
             return Outcome(Operation.FAILED, error=say(_("The VM stayed on the source host")))

instances/tests/test_migrate_job.py

@@ -100,6 +100,27 @@ class MigrateJobTestCase(TransactionTestCase):
             MigrateInstance().reconcile(op)
         dom.destroy.assert_not_called()
 
+    def test_recovery_of_a_moved_vm_sets_its_autostart_and_removes_the_source_nvram(self):
+        self.vm.proxy.get_autostart.return_value = True
+        self.vm.proxy.get_nvram.return_value = "/var/lib/libvirt/qemu/nvram/mig-vm_VARS.fd"
+        op = self.queue()
+        ctx = self.start(op)
+        self.dest_conn.moveto.side_effect = OSError("the worker stops here")
+        with self.assertRaises(OSError):
+            get_kind(op.kind).run(op, ctx)
+        dom = MagicMock()
+        with patch("instances.jobs._domain", side_effect=[None, dom]), patch("instances.jobs._connect") as connect:
+            outcome = MigrateInstance().reconcile(op)
+        self.assertEqual((outcome.state, outcome.result), (Operation.SUCCEEDED, {}))
+        dom.setAutostart.assert_called_once_with(1)
+        connect.return_value.remove_nvram.assert_called_once_with(
+            "/var/lib/libvirt/qemu/nvram/mig-vm_VARS.fd", connect.return_value)
+        dom.setAutostart.side_effect = libvirtError("refused")
+        with patch("instances.jobs._domain", side_effect=[None, dom]), patch("instances.jobs._connect"):
+            outcome = MigrateInstance().reconcile(op)
+        self.assertEqual(outcome.state, Operation.SUCCEEDED)
+        self.assertIn("autostart could not be set", outcome.result["warning"][0]["text"])
+
     def test_a_copy_claims_its_volume_on_both_computes(self):
         Instance.proxy.get_disk_devices.return_value = [
             {"dev": "vda", "storage": "default", "image": "mig-vm.qcow2", "path": "/i/mig-vm.qcow2"},

locale/tr/LC_MESSAGES/django.po

@@ -3018,53 +3018,65 @@ msgstr "%(compute)s sunucusuna ulaşılamıyor"
 msgid "%(compute)s is the VM's own host under another name"
 msgstr "%(compute)s, sanal makinenin kendi sunucusunun başka bir adı"
 
-#: instances/jobs.py:947
+#: instances/jobs.py:963
+#, python-format
+msgid ""
+"The VM migrated, but autostart could not be set on the destination: %(error)s"
+msgstr ""
+"Sanal makine taşındı, ancak hedefte otomatik başlatma ayarlanamadı: %(error)s"
+
+#: instances/jobs.py:969
+#, python-format
+msgid "The VM migrated, but its NVRAM file stays on the source: %(error)s"
+msgstr "Sanal makine taşındı, ancak NVRAM dosyası kaynakta kaldı: %(error)s"
+
+#: instances/jobs.py:976
 msgid "Migrate VM"
 msgstr "Sanal makineyi taşı"
 
-#: instances/jobs.py:991
+#: instances/jobs.py:1023
 msgid "Migrating the VM"
 msgstr "Sanal makine taşınıyor"
 
-#: instances/jobs.py:1038
+#: instances/jobs.py:1070
 msgid "libvirt is still migrating the VM"
 msgstr "libvirt sanal makineyi hâlâ taşıyor"
 
-#: instances/jobs.py:1045
+#: instances/jobs.py:1079
 msgid "The VM stayed on the source host"
 msgstr "Sanal makine kaynak sunucuda kaldı"
 
-#: instances/jobs.py:1047
+#: instances/jobs.py:1081
 msgid ""
 "The VM is on both computes or on neither; check them, then resolve this job."
 msgstr ""
 "Sanal makine ya iki sunucuda da var ya da hiçbirinde yok; kontrol edin, sonra"
 " bu işi çözün."
 
-#: instances/jobs.py:1095
+#: instances/jobs.py:1129
 msgid "Move disk"
 msgstr "Diski taşı"
 
-#: instances/jobs.py:1102
+#: instances/jobs.py:1136
 msgid "Choose a disk of the VM"
 msgstr "Sanal makinenin bir diskini seçin"
 
-#: instances/jobs.py:1126
+#: instances/jobs.py:1160
 msgid "Copying the disk"
 msgstr "Disk kopyalanıyor"
 
-#: instances/jobs.py:1169
+#: instances/jobs.py:1203
 #, python-format
 msgid "Another block job runs on %(dev)s; check the VM"
 msgstr ""
 "%(dev)s üzerinde başka bir blok işi çalışıyor; sanal makineyi kontrol edin"
 
-#: instances/jobs.py:1175
+#: instances/jobs.py:1209
 #, python-format
 msgid "The disk stayed in %(path)s; remove %(dest)s if it is there"
 msgstr "Disk %(path)s içinde kaldı; varsa %(dest)s dosyasını silin"
 
-#: instances/jobs.py:1177
+#: instances/jobs.py:1211
 #, python-format
 msgid "The definitions of %(dev)s name %(paths)s; check the VM"
 msgstr "%(dev)s tanımları %(paths)s gösteriyor; sanal makineyi kontrol edin"

locale/zh/LC_MESSAGES/django.po

@@ -2898,50 +2898,61 @@ msgstr "无法连接 %(compute)s"
 msgid "%(compute)s is the VM's own host under another name"
 msgstr "%(compute)s 是该虚拟机所在主机的另一个名称"
 
-#: instances/jobs.py:947
+#: instances/jobs.py:963
+#, python-format
+msgid ""
+"The VM migrated, but autostart could not be set on the destination: %(error)s"
+msgstr "虚拟机已迁移，但无法在目标主机上设置自动启动：%(error)s"
+
+#: instances/jobs.py:969
+#, python-format
+msgid "The VM migrated, but its NVRAM file stays on the source: %(error)s"
+msgstr "虚拟机已迁移，但其 NVRAM 文件仍留在源主机上：%(error)s"
+
+#: instances/jobs.py:976
 msgid "Migrate VM"
 msgstr "迁移虚拟机"
 
-#: instances/jobs.py:991
+#: instances/jobs.py:1023
 msgid "Migrating the VM"
 msgstr "正在迁移虚拟机"
 
-#: instances/jobs.py:1038
+#: instances/jobs.py:1070
 msgid "libvirt is still migrating the VM"
 msgstr "libvirt 仍在迁移虚拟机"
 
-#: instances/jobs.py:1045
+#: instances/jobs.py:1079
 msgid "The VM stayed on the source host"
 msgstr "虚拟机留在了源主机上"
 
-#: instances/jobs.py:1047
+#: instances/jobs.py:1081
 msgid ""
 "The VM is on both computes or on neither; check them, then resolve this job."
 msgstr "虚拟机同时位于两个计算节点上，或都不在；请检查后处理此任务。"
 
-#: instances/jobs.py:1095
+#: instances/jobs.py:1129
 msgid "Move disk"
 msgstr "移动磁盘"
 
-#: instances/jobs.py:1102
+#: instances/jobs.py:1136
 msgid "Choose a disk of the VM"
 msgstr "请选择虚拟机的一个磁盘"
 
-#: instances/jobs.py:1126
+#: instances/jobs.py:1160
 msgid "Copying the disk"
 msgstr "正在复制磁盘"
 
-#: instances/jobs.py:1169
+#: instances/jobs.py:1203
 #, python-format
 msgid "Another block job runs on %(dev)s; check the VM"
 msgstr "%(dev)s 上正在运行另一个块任务；请检查虚拟机"
 
-#: instances/jobs.py:1175
+#: instances/jobs.py:1209
 #, python-format
 msgid "The disk stayed in %(path)s; remove %(dest)s if it is there"
 msgstr "磁盘仍在 %(path)s；如果 %(dest)s 存在，请将其删除"
 
-#: instances/jobs.py:1177
+#: instances/jobs.py:1211
 #, python-format
 msgid "The definitions of %(dev)s name %(paths)s; check the VM"
 msgstr "%(dev)s 的定义指向 %(paths)s；请检查虚拟机"