25-1: Fix db counters (#20397)

Author: CyberROFL

ydb/core/base/appdata_fwd.h

@@ -318,6 +318,10 @@ inline TAppData* AppData(NActors::TActorSystem* actorSystem) {
     return x;
 }
 
+inline bool HasAppData(NActors::TActorSystem* actorSystem) {
+    return actorSystem && AppData(actorSystem);
+}
+
 inline bool HasAppData() {
     return !!NActors::TlsActivationContext
         && NActors::TActivationContext::ActorSystem()

ydb/core/grpc_services/counters/counters.cpp

@@ -658,7 +658,7 @@ NYdbGrpc::ICounterBlockPtr TServiceCounterCB::operator()(const char* serviceName
     auto block = MakeIntrusive<TYdbCounterBlock>(Counters, serviceName, requestName, streaming);
 
     NYdbGrpc::ICounterBlockPtr res(block);
-    if (ActorSystem && HasAppData() && AppData(ActorSystem)->FeatureFlags.GetEnableDbCounters()) {
+    if (ActorSystem && HasAppData(ActorSystem) && AppData(ActorSystem)->FeatureFlags.GetEnableDbCounters()) {
         res = MakeIntrusive<TYdbCounterBlockWrapper>(block, serviceName, requestName, streaming);
     }
 

ydb/tests/functional/tenants/test_db_counters.py

@@ -23,7 +23,7 @@
 
 
 def get_db_counters(mon_port, service):
-    counters_url = f"http://localhost:{mon_port}/counters/counters%3D{service}/json"
+    counters_url = f"http://localhost:{mon_port}/counters/counters={service}/json"
     reply = requests.get(counters_url)
     if reply.status_code == 204:
         return None
@@ -144,7 +144,7 @@ def get_default_feature_flag_value(feature_flag_camel_case) -> bool:
     ids=["enable_separate_quotas", "disable_separate_quotas"],
 )
 def ydb_cluster_configuration(request):
-    extra_feature_flags = []
+    extra_feature_flags = ['enable_db_counters']
     if request.param:
         extra_feature_flags.append("enable_separate_disk_space_quotas")
     else:
@@ -275,22 +275,37 @@ def wait_for_stats(client, table, retries=10, sleep_duration=5):
 
 
 # Note: the default total sleep time is 300 seconds, because 200 seconds can sometimes be not enough
-def check_counters(mon_port, sensors_to_check, retries=60, sleep_duration=5):
+def check_counters(mon_port, sensors_to_check, service="ydb", retries=60, sleep_duration=5):
+    sensor_count = 0
+    for _, expected_value in sensors_to_check.items():
+        if isinstance(expected_value, list):
+            sensor_count += len(expected_value)
+        else:
+            sensor_count += 1
+
     for attempt in range(retries + 1):
-        counters = get_db_counters(mon_port, "ydb")
+        counters = get_db_counters(mon_port, service)
         correct_sensors = 0
         if counters:
             for sensor in counters["sensors"]:
+                labels = sensor["labels"]
                 for target_name, expected_value in sensors_to_check.items():
-                    if sensor["labels"]["name"] == target_name:
-                        logger.debug(
-                            f"sensor {target_name}: expected {expected_value}, "
-                            f'got {sensor["value"]} in {sleep_duration * attempt} seconds'
-                        )
-                        if sensor["value"] == expected_value:
-                            correct_sensors += 1
-                            if correct_sensors == len(sensors_to_check):
-                                return
+                    if isinstance(expected_value, list):
+                        if labels.get("api_service", "") != target_name:
+                            continue
+                        for x in expected_value:
+                            if len(x) == [labels.get(k, "") == v if k != "value" else sensor[k] == v for k, v in x.items()].count(True):
+                                correct_sensors += 1
+                    else:
+                        if labels["name"] == target_name:
+                            logger.debug(
+                                f"sensor {target_name}: expected {expected_value}, "
+                                f'got {sensor["value"]} in {sleep_duration * attempt} seconds'
+                            )
+                            if sensor["value"] == expected_value:
+                                correct_sensors += 1
+                    if correct_sensors == sensor_count:
+                        return
 
         logger.debug(
             f"got {correct_sensors} out of {len(sensors_to_check)} correct sensors "
@@ -403,3 +418,85 @@ def test_storage_counters(self, ydb_cluster_configuration, ydb_cluster, ydb_data
                     "resources.storage.table.used_bytes.hdd": 0,
                 },
             )
+
+
+def test_serverless_counters(ydb_cluster, ydb_endpoint, ydb_root, ydb_safe_test_name):
+    hostel_db = os.path.join(ydb_root, "hostel_db", ydb_safe_test_name)
+    ydb_cluster.create_hostel_database(
+        hostel_db,
+        storage_pool_units_count={
+            'hdd': 1,
+        },
+    )
+
+    ydb_cluster.register_and_start_slots(hostel_db, count=1)
+    ydb_cluster.wait_tenant_up(hostel_db)
+
+    serverless_db = os.path.join(ydb_root, "serverless", ydb_safe_test_name)
+    ydb_cluster.create_serverless_database(
+        serverless_db,
+        hostel_db=hostel_db,
+        attributes={
+            "cloud_id": "CLOUD_ID_VAL",
+            "folder_id": "FOLDER_ID_VAL",
+            "database_id": "DATABASE_ID_VAL",
+        },
+    )
+
+    driver_config = ydb.DriverConfig(ydb_endpoint, serverless_db, auth_token='root@builtin')
+    driver = ydb.Driver(driver_config)
+    driver.wait(120)
+
+    session = driver.table_client.session().create()
+
+    session.create_table(
+        os.path.join(serverless_db, "table") ,
+        ydb.TableDescription()
+        .with_column(ydb.Column("id", ydb.OptionalType(ydb.DataType.Uint64)))
+        .with_primary_key("id")
+    )
+
+    try:
+        session.create_table(
+            os.path.join(serverless_db, "invalid_table") ,
+            ydb.TableDescription()
+            .with_column(ydb.Column("id", ydb.OptionalType(ydb.DataType.Float)))
+            .with_primary_key("id")
+        )
+    except ydb.issues.SchemeError:
+        pass
+
+    slot_mon_port = ydb_cluster.slots[1].mon_port
+    expected_counters = {
+        "table": [
+            {
+                "method": "CreateSession",
+                "name": "api.grpc.request.count",
+                "value": 1,
+            },
+            {
+                "method": "CreateSession",
+                "name": "api.grpc.response.count",
+                "status": "SUCCESS",
+                "value": 1,
+            },
+            {
+                "method": "CreateTable",
+                "name": "api.grpc.request.count",
+                "value": 2,
+            },
+            {
+                "method": "CreateTable",
+                "name": "api.grpc.response.count",
+                "status": "SUCCESS",
+                "value": 1,
+            },
+            {
+                "method": "CreateTable",
+                "name": "api.grpc.response.count",
+                "status": "SCHEME_ERROR",
+                "value": 1,
+            },
+        ],
+    }
+    check_counters(slot_mon_port, expected_counters, "ydb_serverless")