feat postgres: Add disabled_replicas to topology_settings

Tests: testsuite
commit_hash:c9cb7f67d8fd15e5642dde89d8f53fe672e077e5

postgresql/include/userver/storages/postgres/options.hpp

@@ -12,6 +12,7 @@
 #include <userver/congestion_control/controllers/linear.hpp>
 #include <userver/storages/postgres/postgres_fwd.hpp>
 #include <userver/utils/impl/transparent_hash.hpp>
+#include <userver/utils/str_icase.hpp>
 
 USERVER_NAMESPACE_BEGIN
 
@@ -150,8 +151,16 @@ inline constexpr std::size_t kDefaultPoolMaxQueueSize = 200;
 /// Default limit for concurrent establishing connections number
 inline constexpr std::size_t kDefaultConnectingLimit = 0;
 
+/// @brief PostgreSQL topology options
+///
+/// Dynamic option @ref POSTGRES_TOPOLOGY_SETTINGS
 struct TopologySettings {
+    /// Maximum replication lag. Once the replica lag exceeds this value it will be automatically disabled.
     std::chrono::milliseconds max_replication_lag{kDefaultMaxReplicationLag};
+
+    /// List of manually disabled replicas (FQDNs).
+    std::unordered_set<std::string, USERVER_NAMESPACE::utils::StrIcaseHash, USERVER_NAMESPACE::utils::StrIcaseEqual>
+        disabled_replicas{};
 };
 
 /// @brief PostgreSQL connection pool options

postgresql/src/storages/postgres/detail/topology/hot_standby.cpp

@@ -98,7 +98,7 @@ size_t ParseSize(std::string_view token) {
 }  // namespace
 
 struct HotStandby::HostState {
-    explicit HostState(const Dsn& dsn) : app_name{EscapeHostName(OptionsFromDsn(dsn).host)} {}
+    explicit HostState(const Dsn& dsn) : host_name{OptionsFromDsn(dsn).host}, app_name(EscapeHostName(host_name)) {}
 
     ~HostState() {
         // close connections synchronously
@@ -117,7 +117,9 @@ struct HotStandby::HostState {
 
     std::unique_ptr<Connection> connection;
 
-    const std::string host_port;
+    // Used to check against disabled_replicas
+    std::string host_name;
+
     // In pg_stat_replication slaves' host names are escaped and the column is
     // called `application_name`
     std::string app_name;
@@ -225,11 +227,16 @@ void HotStandby::RunDiscovery() {
             std::chrono::duration_cast<std::chrono::milliseconds>(slave_lag).count()
         );
 
-        auto& max_replication_lag = GetTopologySettings().max_replication_lag;
-        if (max_replication_lag > std::chrono::milliseconds{0} && slave_lag > max_replication_lag) {
+        const auto& topology_settings = GetTopologySettings();
+        if (topology_settings.max_replication_lag > std::chrono::milliseconds{0} &&
+            slave_lag > topology_settings.max_replication_lag) {
             // Demote lagged slave
             LOG_INFO() << "Disabling slave " << slave.app_name << " due to replication lag of " << slave_lag.count()
-                       << " ms (max " << GetTopologySettings().max_replication_lag.count() << " ms)";
+                       << " ms (max " << topology_settings.max_replication_lag.count() << " ms)";
+            slave.role = ClusterHostType::kNone;
+        } else if (topology_settings.disabled_replicas.count(slave.host_name)) {
+            // Manually disable slave
+            LOG_INFO() << "Disabling slave " << slave.app_name << " due to manual setting in config";
             slave.role = ClusterHostType::kNone;
         } else if (master) {
             // Check for sync slave

postgresql/src/storages/postgres/postgres_config.cpp

@@ -123,6 +123,7 @@ TopologySettings Parse(const formats::json::Value& config, formats::parse::To<To
 
     result.max_replication_lag =
         config["max_replication_lag_ms"].template As<std::chrono::milliseconds>(result.max_replication_lag);
+    result.disabled_replicas = config["disabled_replicas"].template As<decltype(result.disabled_replicas)>({});
 
     if (result.max_replication_lag < std::chrono::milliseconds{0})
         throw InvalidConfig{"max_replication_lag cannot be less than 0"};

scripts/docs/en/schemas/dynamic_configs.md

@@ -467,6 +467,11 @@ properties:
       minimum: 0
       description: maximum allowed replication lag. If equals 0 no replication 
       lag checks are performed
+    disabled_replicas:
+      type: array
+      description: List of manually disabled replicas (FQDNs).
+      items:
+        type: string
 required:
   - max_replication_lag_ms
 ```
@@ -475,7 +480,8 @@ required:
 ```json
 {
   "__default__": {
-    "max_replication_lag_ms": 60000
+    "max_replication_lag_ms": 60000,
+    "disabled_replicas": ["replica-01.example.com", "replica-02.example.com"]
   }
 }
 ```