spec for resource deleting

1. What is the problem?
During the deletion of a network which is mapped to several local networks, if
the local neutron server receives a network-get request, it will bring some
conflict operations. For example, we delete local networks before deleting
central network. If a "get_network" request comes to a local neutron server
after the local network is completely deleted in that region, if central
network is still there (assuming it takes certain time to delete all local
networks), Tricircle will also query central neutron and the deleted local
network will be recreated.

2. What is the solution to the problem?
The solution to resource deleting is covered in this specification.

3. What the features to be implemented in the Tricircle
to realize the solution?
No new features.

Change-Id: I4f29ee9f214242262bffc616dc60c8fdde76a6f6
Signed-off-by: song baisen <songbaisen@szzt.com.cn>
Co-Authored-By: CR_hui <CR_hui@126.com>

specs/queens/resource_deleting.rst

@@ -0,0 +1,66 @@
+========================================
+Reliable resource deleting in Tricircle
+========================================
+
+Background
+==========
+During the deletion of resources which are mapped to several local Neutron(s),
+it may bring some conflict operations. For example, deleting a network in
+central neutron which is also resided in several local Neutron(s). The reason
+is that network-get request will trigger local neutron to query central
+neutron and create the network, and we delete local networks before deleting
+central network. When a network-get request comes to a local neutron server
+after the local network is completely deleted in that region and at this time
+the network in central neutron still exists (assuming it takes certain time to
+delete all local networks), local neutron will still retrieve the network from
+central neutron and the deleted local network will be recreated. This issue
+also applies to the deletion cases of other resource types.
+
+Proposed Solution
+=================
+Recently, Tricircle adds a feature to distinguish the source of requests[1], so
+we can distinguish the deletion request from 'Central Neutron' or
+'Local Neutron'. In order to avoid the conflict mentioned above, we introduce a
+new table called "deleting_resource" in Tricircle database, so central plugin
+can save the resource deletion information and set the information when it
+receives a deletion request. Here is the schema of the table:
+
+.. csv-table:: Resource deleting table
+  :header: Field, Type, Nullable, pk/fk/uk, Description
+
+  resource_id, string, False, uk, resource id in central Neutron
+  resource_type, string, False, uk, resource_type denotes one of the available resource types
+  deleted_at, timestamp, False, n/a, deletion timestamp
+
+**How to delete the resource without conflict operation**
+
+Let's take network deletion as an example.
+
+At the beginning of network-delete handle, central neutron server sets the
+information of deleted network into the "deleting_resource" table.
+
+At this point, if get-request from local neutron servers comes, central
+neutron server will check the "deleting_resource" table whether the
+associated resource has been recorded and return 404 to local neutron server
+if the associated resources is being deleting.
+
+At this point, if deletion request is from central Neutron, central neutron
+server will check the "deleting_resource" table whether the associated
+resource has been recorded and it will return 204 to user if associated
+resource is being deleting.
+
+For the get-request of user, central neutron server will query the related
+network information in "deleting_resource" table and will return the deleting
+resource to user if the network information which the user queries exists in
+the table. When user re-deleting the network after something wrong happens,
+central neutron will return 204 to user.
+
+At the end of network-delete handle that all the mapped local networks have
+been deleted, central neutron server will remove the deleting resource record
+and remove this network.
+
+In addition, there is a timestamp in table that cloud administrator is able to
+delete a resource which is in deleting status over long time (too long to
+delete, or in abnormal status).
+
+[1] https://review.openstack.org/#/c/518421/