gNMI server based SONiC Configuration Set HLD

doc/SONiC-Configuration-Set-HLD.md

@@ -0,0 +1,67 @@
+# Terminology
+CS: Configuration SET
+
+# Summary
+This design propose an secure and light weight service that will provide a transparent and automated configuration query/subscription on given device set.
+
+# Scope
+This design only works on devices having gNMI server equipped.
+
+# Config DB Extention
+For example if we want to query/subscribe aggregate address on given device set:
+```json
+{
+    ...
+    "CONFIGURATION_SET" : {
+        "BGP_AGGREGATE_ADDRESS": {
+            "set_members" : [ "ip1", "ip2", "ip3", "ip4" ], # It contains all devices' ip address in the same set
+            "query_per_second" : "n", # it means it will query every peer n times per second
+        },
+        "SOME_CONFIG_KEY": {
+            ...
+        },
+        ...
+    }
+    ...
+}
+```
+
+# State DB Extention
+```json
+{
+    ...
+    "CONFIGURATION_SET": {
+        "BGP_AGGREGATE_ADDRESS|Prefix-1": {
+            "state": "entire"
+        },
+        "BGP_AGGREGATE_ADDRESS|Prefix-2": {
+            "state": "partial"
+        },
+        "BGP_AGGREGATE_ADDRESS|Prefix-3": {
+            "state": "outage"
+        },
+        ...
+    }
+    ...
+}
+```
+
+# Server Logic
+1. Track target configuration on all devices in set
+    - every device will query target configuration on peers periodically
+    - once confirmed that all devices have same config, it will notify subscriber and update state db with "entire" state.
+    - once any device lose target configuration, it will notify subscriber and update state db with "partial" state.
+2. Outage Scenarios
+   - Server only works when all peers is reachable via gNMi. If any device in CS is unreachable via gNMi, all configuration will be update to "outage" state.
+    - If the CS members config is not identical on every devices, all configuration will be update to "outage" state.
+
+# Usage Scenario
+1. Aggregate Address Configuration on T1 set:
+   - Firstly, we make T1s as a CS and register aggregate address config key in CS.
+   - Secondly, we let the bgp container subscribe aggregate address config key in CS.
+   - Thirdly, we can add aggregate address config on T1s one by one.
+   - Then, CS will notify the bgp container the state of aggregate address in CS, and the bgp container can take action accordingly.
+
+# Possible Enhancement
+1. Network isolation awareness: if some devices in configuration set is in planned maintenance, we can ignore those devices.
+2. Configuration setting: if there is config diff between devices, we can align them.