Add state provider status GraphQL API (#2199)

CHANGELOG.rst

@@ -18,6 +18,8 @@ Added
 
 - Twophase commit timeouts now can be set with env.
 
+- New GraphQL API ``failover_state_provider_status`` to ping state provider connection.
+
 -------------------------------------------------------------------------------
 [2.9.0] - 2024-03-06
 -------------------------------------------------------------------------------

cartridge/lua-api/failover.lua

@@ -5,6 +5,9 @@
 local checks = require('checks')
 local errors = require('errors')
 local fun = require('fun')
+local net_box = require('net.box')
+local httpc = require('http.client')
+local digest = require('digest')
 
 local twophase = require('cartridge.twophase')
 local topology = require('cartridge.topology')
@@ -351,12 +354,57 @@ local function resume()
     return true
 end
 
+local --[[const]] PING_TIMEOUT = 3 -- seconds
+--- Gets status of the state provider if stateful failover is enabled.
+--
+-- @function get_state_provider_status
+--
+-- @treturn table {"state-provider-url": bool status}
+--   or empty table if there are no state provider
+local function get_state_provider_status()
+    local failover_params = topology.get_failover_params(
+        confapplier.get_readonly('topology')
+    )
+    if failover_params.mode == 'stateful' then
+        if failover_params.state_provider == 'etcd2' then
+            local result = {}
+            local etcd_params = failover_params.etcd2_params
+            local etcd_uris = etcd_params.endpoints
+            local http_auth
+            if failover_params.etcd2_params.username ~= '' then
+                local credentials = etcd_params.username .. ":" .. etcd_params.password
+                http_auth = "Basic " .. digest.base64_encode(credentials)
+            end
+            for _, uri in ipairs(etcd_uris) do
+                local resp = httpc.head(uri, {
+                    timeout = PING_TIMEOUT,
+                    headers = {
+                        ['Authorization'] = http_auth,
+                    },
+                })
+                result[uri] = resp.headers ~= nil
+            end
+            return result
+        elseif failover_params.state_provider == 'tarantool' then
+            local state_provider_uri = failover_params.tarantool_params.uri
+            local conn = net_box.connect(state_provider_uri, {
+                user = 'client',
+                password = failover_params.tarantool_params.password,
+            })
+            return {[state_provider_uri] = conn:ping({timeout = PING_TIMEOUT})}
+        end
+    end
+
+    return {}
+end
+
 return {
     get_params = get_params,
     set_params = set_params,
     promote = promote,
     pause = pause,
     resume = resume,
+    get_state_provider_status = get_state_provider_status,
     get_failover_enabled = get_failover_enabled, -- deprecated
     set_failover_enabled = set_failover_enabled, -- deprecated
 }

cartridge/webui/api-failover.lua

@@ -73,6 +73,21 @@ local gql_type_userapi = gql_types.object({
     }
 })
 
+local gql_type_state_provider_status = gql_types.object({
+    name = 'StateProviderStatus',
+    description = 'Failover state provider status',
+    fields = {
+        uri = {
+            kind = gql_types.string.nonNull,
+            description = 'State provider uri',
+        },
+        status = {
+            kind = gql_types.boolean.nonNull,
+            description = 'State provider status',
+        },
+    }
+})
+
 local function get_failover_params(_, _)
     local failover_params = lua_api_failover.get_params()
     local masked_pwd = '******'
@@ -116,6 +131,14 @@ local function promote(_, args)
     return lua_api_failover.promote({[replicaset_uuid] = instance_uuid}, opts)
 end
 
+local function get_state_provider_status(_, _)
+    local result = {}
+    for uri, status in pairs(lua_api_failover.get_state_provider_status()) do
+        table.insert(result, {uri = uri, status = status})
+    end
+    return result
+end
+
 local function init(graphql)
 
     graphql.add_callback({
@@ -150,6 +173,15 @@ local function init(graphql)
         callback = module_name .. '.get_failover_params',
     })
 
+    graphql.add_callback({
+        prefix = 'cluster',
+        name = 'failover_state_provider_status',
+        doc = 'Get state provider status.',
+        args = {},
+        kind = gql_types.list(gql_type_state_provider_status).nonNull,
+        callback = module_name .. '.get_state_provider_status',
+    })
+
     graphql.add_mutation({
         prefix = 'cluster',
         name = 'failover_params',
@@ -209,6 +241,7 @@ return {
     set_failover_enabled = set_failover_enabled,
     get_failover_params = get_failover_params,
     set_failover_params = set_failover_params,
+    get_state_provider_status = get_state_provider_status,
     promote = promote,
     pause = pause,
     resume = resume,

doc/schema.graphql

@@ -31,6 +31,9 @@ type Apicluster {
   """Show suggestions to resolve operation problems"""
   suggestions: Suggestions
 
+  """Get state provider status."""
+  failover_state_provider_status: [StateProviderStatus]!
+
   """List authorized users"""
   users(
     """
@@ -41,24 +44,24 @@ type Apicluster {
     username: String
   ): [User!]
 
-  """Virtual buckets count in cluster"""
-  vshard_bucket_count: Int!
+  """compression info about cluster"""
+  cluster_compression: ClusterCompressionInfo!
 
   """Get list of all registered roles and their dependencies."""
   known_roles: [Role!]!
 
+  """Virtual buckets count in cluster"""
+  vshard_bucket_count: Int!
+
   """Get automatic failover configuration."""
   failover_params: FailoverAPI!
-
-  """Get list of known vshard storage groups."""
-  vshard_known_groups: [String!]!
   vshard_groups: [VshardGroup!]!
 
   """Validate config"""
   validate_config(sections: [ConfigSectionInput]): ValidateConfigResult!
 
-  """compression info about cluster"""
-  cluster_compression: ClusterCompressionInfo!
+  """Get list of known vshard storage groups."""
+  vshard_known_groups: [String!]!
 
   """Get cluster config sections"""
   config(sections: [String!]): [ConfigSection]!
@@ -1069,6 +1072,15 @@ type SpaceCompressionInfo {
   space_name: String!
 }
 
+"""Failover state provider status"""
+type StateProviderStatus {
+  """State provider uri"""
+  uri: String!
+
+  """State provider status"""
+  status: Boolean!
+}
+
 type Suggestions {
   force_apply: [ForceApplySuggestion!]
   restart_replication: [RestartReplicationSuggestion!]

test/integration/failover_stateful_test.lua

@@ -776,3 +776,42 @@ add('test_invalid_params', function(g)
         return require('cartridge.confapplier').get_state()
     end), 'RolesConfigured')
 end)
+
+local function get_status()
+    return S1:graphql({
+        query = [[{
+            cluster {
+                failover_state_provider_status {
+                    uri
+                    status
+                }
+            }
+        }]],
+    })
+end
+
+local test_cases = {
+    [g_stateboard] = "localhost:14401",
+    [g_etcd2] = "http://127.0.0.1:14001",
+}
+for group, uri in pairs(test_cases) do
+    group.test_get_state_provider_status = function(g)
+        helpers.retrying({}, function()
+            t.assert(g.client:get_session():get_coordinator())
+        end)
+
+        local resp = get_status()
+        t.assert_equals(resp['data']['cluster']['failover_state_provider_status'][1]['uri'], uri)
+        t.assert_equals(resp['data']['cluster']['failover_state_provider_status'][1]['status'], true)
+
+        -- stop state provider
+        g.state_provider:stop()
+
+        local resp = get_status()
+        t.assert_equals(resp['data']['cluster']['failover_state_provider_status'][1]['uri'], uri)
+        t.assert_equals(resp['data']['cluster']['failover_state_provider_status'][1]['status'], false)
+        g.state_provider:start()
+    end
+end
+
+