feat(telemetry): save uuids to files

In order for the node and cluster UUIDs used by telemetry to survive
database purges, we'll save the generated UUIDs to files in EMQX's
data directory as well.

apps/emqx_modules/src/emqx_telemetry.erl

@@ -93,6 +93,9 @@
 
 -define(TELEMETRY_SHARD, emqx_telemetry_shard).
 
+-define(NODE_UUID_FILENAME, "node.uuid").
+-define(CLUSTER_UUID_FILENAME, "cluster.uuid").
+
 %%--------------------------------------------------------------------
 %% API
 %%--------------------------------------------------------------------
@@ -535,7 +538,11 @@ ensure_uuids() ->
         NodeUUID =
             case mnesia:wread({?TELEMETRY, node()}) of
                 [] ->
-                    NodeUUID0 = generate_uuid(),
+                    NodeUUID0 =
+                        case get_uuid_from_file(node) of
+                            {ok, NUUID} -> NUUID;
+                            undefined -> generate_uuid()
+                        end,
                     mnesia:write(
                         ?TELEMETRY,
                         #telemetry{
@@ -551,7 +558,11 @@ ensure_uuids() ->
         ClusterUUID =
             case mnesia:wread({?TELEMETRY, ?CLUSTER_UUID_KEY}) of
                 [] ->
-                    ClusterUUID0 = generate_uuid(),
+                    ClusterUUID0 =
+                        case get_uuid_from_file(cluster) of
+                            {ok, CUUID} -> CUUID;
+                            undefined -> generate_uuid()
+                        end,
                     mnesia:write(
                         ?TELEMETRY,
                         #telemetry{
@@ -566,8 +577,35 @@ ensure_uuids() ->
             end,
         {NodeUUID, ClusterUUID}
     end,
-    {atomic, UUIDs} = mria:transaction(?TELEMETRY_SHARD, Txn),
-    UUIDs.
+    {atomic, {NodeUUID, ClusterUUID}} = mria:transaction(?TELEMETRY_SHARD, Txn),
+    save_uuid_to_file(NodeUUID, node),
+    save_uuid_to_file(ClusterUUID, cluster),
+    {NodeUUID, ClusterUUID}.
+
+get_uuid_from_file(Type) ->
+    Path = uuid_file_path(Type),
+    case file:read_file(Path) of
+        {ok,
+            UUID =
+                <<_:8/binary, "-", _:4/binary, "-", _:4/binary, "-", _:4/binary, "-", _:12/binary>>} ->
+            {ok, UUID};
+        _ ->
+            undefined
+    end.
+
+save_uuid_to_file(UUID, Type) when is_binary(UUID) ->
+    Path = uuid_file_path(Type),
+    ok = filelib:ensure_dir(Path),
+    ok = file:write_file(Path, UUID).
+
+uuid_file_path(Type) ->
+    DataDir = emqx:data_dir(),
+    Filename =
+        case Type of
+            node -> ?NODE_UUID_FILENAME;
+            cluster -> ?CLUSTER_UUID_FILENAME
+        end,
+    filename:join(DataDir, Filename).
 
 empty_state() ->
     #state{

apps/emqx_modules/test/emqx_telemetry_SUITE.erl

@@ -151,6 +151,41 @@ init_per_testcase(t_cluster_uuid, Config) ->
     Node = start_slave(n1),
     ok = setup_slave(Node),
     [{n1, Node} | Config];
+init_per_testcase(t_uuid_restored_from_file, Config) ->
+    mock_httpc(),
+    NodeUUID = <<"AAAAAAAA-BBBB-CCCC-DDDD-EEEEEEEEEEEE">>,
+    ClusterUUID = <<"FFFFFFFF-GGGG-HHHH-IIII-JJJJJJJJJJJJ">>,
+    DataDir = emqx:data_dir(),
+    NodeUUIDFile = filename:join(DataDir, "node.uuid"),
+    ClusterUUIDFile = filename:join(DataDir, "cluster.uuid"),
+    file:delete(NodeUUIDFile),
+    file:delete(ClusterUUIDFile),
+    ok = file:write_file(NodeUUIDFile, NodeUUID),
+    ok = file:write_file(ClusterUUIDFile, ClusterUUID),
+
+    %% clear the UUIDs in the DB
+    {atomic, ok} = mria:clear_table(emqx_telemetry),
+    emqx_common_test_helpers:stop_apps([emqx_conf, emqx_authn, emqx_authz, emqx_modules]),
+    emqx_common_test_helpers:start_apps(
+        [emqx_conf, emqx_authn, emqx_authz, emqx_modules],
+        fun set_special_configs/1
+    ),
+    Node = start_slave(n1),
+    ok = setup_slave(Node),
+    [
+        {n1, Node},
+        {node_uuid, NodeUUID},
+        {cluster_uuid, ClusterUUID}
+        | Config
+    ];
+init_per_testcase(t_uuid_saved_to_file, Config) ->
+    mock_httpc(),
+    DataDir = emqx:data_dir(),
+    NodeUUIDFile = filename:join(DataDir, "node.uuid"),
+    ClusterUUIDFile = filename:join(DataDir, "cluster.uuid"),
+    file:delete(NodeUUIDFile),
+    file:delete(ClusterUUIDFile),
+    Config;
 init_per_testcase(t_num_clients, Config) ->
     mock_httpc(),
     ok = snabbkaffe:start_trace(),
@@ -213,6 +248,14 @@ end_per_testcase(t_num_clients, Config) ->
     meck:unload([httpc]),
     ok = snabbkaffe:stop(),
     Config;
+end_per_testcase(t_uuid_restored_from_file, _Config) ->
+    DataDir = emqx:data_dir(),
+    NodeUUIDFile = filename:join(DataDir, "node.uuid"),
+    ClusterUUIDFile = filename:join(DataDir, "cluster.uuid"),
+    ok = file:delete(NodeUUIDFile),
+    ok = file:delete(ClusterUUIDFile),
+    meck:unload([httpc]),
+    ok;
 end_per_testcase(_Testcase, _Config) ->
     meck:unload([httpc]),
     ok.
@@ -248,6 +291,48 @@ t_cluster_uuid(Config) ->
     ?assertNotEqual(NodeUUID0, NodeUUID1),
     ok.
 
+%% should attempt read UUID from file in data dir to keep UUIDs
+%% unique, in the event of a database purge.
+t_uuid_restored_from_file(Config) ->
+    ExpectedNodeUUID = ?config(node_uuid, Config),
+    ExpectedClusterUUID = ?config(cluster_uuid, Config),
+    ?assertEqual(
+        {ok, ExpectedNodeUUID},
+        emqx_telemetry:get_node_uuid()
+    ),
+    ?assertEqual(
+        {ok, ExpectedClusterUUID},
+        emqx_telemetry:get_cluster_uuid()
+    ),
+    ok.
+
+t_uuid_saved_to_file(_Config) ->
+    DataDir = emqx:data_dir(),
+    NodeUUIDFile = filename:join(DataDir, "node.uuid"),
+    ClusterUUIDFile = filename:join(DataDir, "cluster.uuid"),
+    %% preconditions
+    ?assertEqual({error, enoent}, file:read_file(NodeUUIDFile)),
+    ?assertEqual({error, enoent}, file:read_file(ClusterUUIDFile)),
+
+    %% clear the UUIDs in the DB
+    {atomic, ok} = mria:clear_table(emqx_telemetry),
+    emqx_common_test_helpers:stop_apps([emqx_conf, emqx_authn, emqx_authz, emqx_modules]),
+    emqx_common_test_helpers:start_apps(
+        [emqx_conf, emqx_authn, emqx_authz, emqx_modules],
+        fun set_special_configs/1
+    ),
+    {ok, NodeUUID} = emqx_telemetry:get_node_uuid(),
+    {ok, ClusterUUID} = emqx_telemetry:get_cluster_uuid(),
+    ?assertEqual(
+        {ok, NodeUUID},
+        file:read_file(NodeUUIDFile)
+    ),
+    ?assertEqual(
+        {ok, ClusterUUID},
+        file:read_file(ClusterUUIDFile)
+    ),
+    ok.
+
 t_official_version(_) ->
     true = emqx_telemetry:official_version("0.0.0"),
     true = emqx_telemetry:official_version("1.1.1"),