slapos_crm: Move SLA check into SoftwareInstance_getReportedErrorDict

   In general the context of the report should be the instance tree. The calculation is cheap to assert the SLA, and this script already has specific checks to inform the user about bad approaches.

master/bt5/slapos_crm/SkinTemplateItem/portal_skins/slapos_crm_monitoring/ComputePartition_checkAllocationConsistencyState.py

@@ -9,60 +9,12 @@ instance_list = compute_partition.getAggregateRelatedValueList(portal_type=[
   'Software Instance', 'Slave Instance'])
 
 for instance in instance_list:
-  instance_sla_error_list = []
   if instance.getValidationState() != 'validated' or \
       instance.getSlapState() == 'destroy_requested':
     # Outdated catalog or instance under garbage collection,
     # we skip for now.
     continue
 
-  sla_dict = instance.getSlaXmlAsDict()
-  if sla_dict:
-    # Simple check of instance SLAs
-    if "computer_guid" in sla_dict:
-      computer_guid = sla_dict.pop("computer_guid")
-      if compute_node.getReference() != computer_guid:
-        instance_sla_error_list.append('computer_guid do not match on: %s (%s != %s)' % (
-          instance.getTitle(), computer_guid, compute_node.getReference()))
-
-    if "instance_guid" in sla_dict:
-      if instance.getPortalType() != 'Slave Instance':
-        instance_sla_error_list.append('instance_guid is provided to a Software Instance: %s' % instance.getTitle())
-      else:
-        instance_guid = sla_dict.pop("instance_guid")
-        software_instance = compute_partition.getAggregateRelatedValue(portal_type='Software Instance')
-        if software_instance is None:
-          instance_sla_error_list.append('instance_guid provided on %s but no Software Instance was found' % instance.getTitle())
-        else:
-          if software_instance.getReference() != instance_guid:
-            instance_sla_error_list.append('instance_guid do not match on: %s (%s != %s)' % (
-              instance.getTitle(), instance_guid, software_instance.getReference()))
-
-    if 'network_guid' in sla_dict:
-      network_guid = sla_dict.pop('network_guid')
-      network_reference = compute_node.getSubordinationReference()
-      if network_reference != network_guid:
-        instance_sla_error_list.append('network_guid do not match on: %s (%s != %s)' % (
-          instance.getTitle(), network_guid, network_reference))
-
-    project_reference = compute_node.getFollowUpReference()
-    if 'project_guid' in sla_dict:
-      project_guid = sla_dict.pop("project_guid")
-      if project_reference != project_guid:
-        instance_sla_error_list.append('project_guid do not match on: %s (%s != %s)' % (
-          instance.getTitle(), project_guid, project_reference))
-
-    instance_project_reference = instance.getFollowUpReference()
-    if project_reference != instance_project_reference:
-      instance_sla_error_list.append("Instance and Compute node project don't match on: %s (%s != %s)" % (
-        instance.getTitle(), project_reference, instance_project_reference))
-
-    if instance_sla_error_list:
-      error_dict[instance.getRelativeUrl()] = {
-        'instance': instance,
-        'sla_error_list': instance_sla_error_list
-      }
-
   # Now check allocation supply consistency
   instance_tree = instance.getSpecialiseValue(portal_type="Instance Tree")
 

master/bt5/slapos_crm/SkinTemplateItem/portal_skins/slapos_crm_monitoring/SoftwareInstance_getReportedErrorDict.py

@@ -48,6 +48,67 @@ if context.getPortalType() == 'Slave Instance' and compute_node.getPortalType()
     error_dict['ticket_description'] = error_dict['message']
     return error_dict
 
+
+sla_dict = context.getSlaXmlAsDict()
+if sla_dict:
+  instance_sla_error_list = []
+  instance_title = context.getTitle()
+  # Simple check of instance SLAs
+  if "computer_guid" in sla_dict:
+    computer_guid = sla_dict.pop("computer_guid")
+    if compute_node.getReference() != computer_guid:
+      instance_sla_error_list.append('computer_guid do not match on: %s (%s != %s)' % (
+        instance_title, computer_guid, compute_node.getReference()))
+
+    if "instance_guid" in sla_dict:
+      instance_guid = sla_dict.pop("instance_guid")
+      if context.getPortalType() != 'Slave Instance':
+        instance_sla_error_list.append('instance_guid is provided to a Software Instance: %s' % instance_title)
+      else:
+        if compute_node.getPortalType() == "Remote Node":
+          instance_sla_error_list.append('instance_guid provided on %s and it is allocated on a REMOTE NODE' % instance_title)
+        else:
+          software_instance = compute_partition.getAggregateRelatedValue(portal_type='Software Instance')
+          if software_instance is None:
+            instance_sla_error_list.append('instance_guid provided on %s but no Software Instance was found' % instance_title)
+          else:
+            if software_instance.getReference() != instance_guid:
+              instance_sla_error_list.append('instance_guid do not match on: %s (%s != %s)' % (
+                instance_title, instance_guid, software_instance.getReference()))
+
+  if 'network_guid' in sla_dict:
+    network_guid = sla_dict.pop('network_guid')
+    network_reference = compute_node.getSubordinationReference()
+    if network_reference != network_guid:
+      instance_sla_error_list.append('network_guid do not match on: %s (%s != %s)' % (
+        instance_title, network_guid, network_reference))
+
+  project_reference = compute_node.getFollowUpReference()
+  if 'project_guid' in sla_dict:
+    project_guid = sla_dict.pop("project_guid")
+    if project_reference != project_guid:
+      instance_sla_error_list.append('project_guid do not match on: %s (%s != %s)' % (
+        instance_title, project_guid, project_reference))
+
+  instance_project_reference = context.getFollowUpReference()
+  if project_reference != instance_project_reference:
+    instance_sla_error_list.append("Instance and Compute node project don't match on: %s (%s != %s)" % (
+      instance_title, project_reference, instance_project_reference))
+
+  if instance_sla_error_list:
+    # Slave instance is allocated but the software instance was already destroyed
+    error_dict['notification_message_reference'] = 'slapos-crm-instance-tree-has-invalid-sla.notification'
+    error_dict = updateErrorDictWithError(error_dict)
+    error_text = ""
+    for _e in instance_sla_error_list:
+      error_text += "  * %s\n" % _e
+
+    error_dict['error_text'] = error_text
+    error_dict['ticket_description'] = "%s has invalid Service Level Aggrement." % instance_title
+    error_dict['message'] = "%s Detected inconsistencies:\n%s" % (
+      error_text, error_dict['ticket_description'])
+    return error_dict
+
 # Skip to check if monitor disabled on the compute node.
 # Remote node has no state.
 if (compute_node.getPortalType() == "Compute Node") and (compute_node.getMonitorScope() != "enabled"):