frostfs-testcases/pytest_tests/testsuites/metrics/test_grpc_metrics.py

import random
import re

import allure
import pytest
from frostfs_testlib import reporter
from frostfs_testlib.healthcheck.interfaces import Healthcheck
from frostfs_testlib.steps.cli.container import create_container, delete_container, get_container, list_containers
from frostfs_testlib.steps.cli.object import get_object, head_object, put_object, search_object
from frostfs_testlib.steps.cli.tree import get_tree_list
from frostfs_testlib.storage.cluster import Cluster, ClusterNode
from frostfs_testlib.storage.controllers.cluster_state_controller import ClusterStateController
from frostfs_testlib.storage.controllers.state_managers.config_state_manager import ConfigStateManager
from frostfs_testlib.storage.dataclasses.frostfs_services import StorageNode
from frostfs_testlib.storage.dataclasses.object_size import ObjectSize
from frostfs_testlib.storage.dataclasses.wallet import WalletInfo
from frostfs_testlib.testing.cluster_test_base import ClusterTestBase
from frostfs_testlib.testing.test_control import wait_for_success
from frostfs_testlib.utils.file_utils import generate_file


class TestGRPCMetrics(ClusterTestBase):
    @pytest.fixture
    def disable_policer(self, cluster_state_controller: ClusterStateController):
        config_manager = cluster_state_controller.manager(ConfigStateManager)
        config_manager.set_on_all_nodes(StorageNode, {"policer:unsafe_disable": "true"})
        yield
        cluster_state_controller.manager(ConfigStateManager).revert_all()

    @wait_for_success(interval=10)
    def check_metrics_in_node(self, cluster_node: ClusterNode, counter_exp: int, **metrics_greps: str):
        counter_act = self.get_metrics_value(cluster_node, **metrics_greps)
        assert counter_act == counter_exp, f"Expected: {counter_exp}, Actual: {counter_act} in node: {cluster_node}"

    def get_metrics_value(self, node: ClusterNode, **metrics_greps: str):
        try:
            command_result = node.metrics.storage.get_metrics_search_by_greps(**metrics_greps)
            metrics_counter = self.calc_metrics_count_from_stdout(command_result.stdout)
        except RuntimeError as e:
            metrics_counter = 0

        return metrics_counter

    @staticmethod
    def calc_metrics_count_from_stdout(metric_result_stdout: str):
        result = re.findall(r"}\s(\d+)", metric_result_stdout)
        return sum(map(int, result))

    @allure.title("GRPC metrics container operations")
    def test_grpc_metrics_container_operations(self, default_wallet: WalletInfo, cluster: Cluster):
        placement_policy = "REP 2 IN X CBF 1 SELECT 4 FROM * AS X"

        with reporter.step("Select random node"):
            node = random.choice(cluster.cluster_nodes)

        with reporter.step("Get current gRPC metrics for method 'Put'"):
            metrics_counter_put = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ContainerService", method="Put"
            )

        with reporter.step(f"Create container with policy {placement_policy}"):
            cid = create_container(default_wallet, self.shell, node.storage_node.get_rpc_endpoint(), placement_policy)

        with reporter.step(f"Check gRPC metrics method 'Put', 'the counter should increase by 1'"):
            metrics_counter_put += 1
            self.check_metrics_in_node(
                node, metrics_counter_put, command="grpc_server_handled_total", service="ContainerService", method="Put"
            )

        with reporter.step("Get current gRPC metrics for method 'Get'"):
            metrics_counter_get = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ContainerService", method="Get"
            )

        with reporter.step(f"Get container"):
            get_container(default_wallet, cid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method=Get, 'the counter should increase by 1'"):
            metrics_counter_get += 1
            self.check_metrics_in_node(
                node, metrics_counter_get, command="grpc_server_handled_total", service="ContainerService", method="Get"
            )

        with reporter.step("Get current gRPC metrics for method 'List'"):
            metrics_counter_list = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ContainerService", method="List"
            )

        with reporter.step(f"Get container list"):
            list_containers(default_wallet, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method=List, 'the counter should increase by 1'"):
            metrics_counter_list += 1
            self.check_metrics_in_node(
                node,
                metrics_counter_list,
                command="grpc_server_handled_total",
                service="ContainerService",
                method="List",
            )

        with reporter.step("Delete container"):
            delete_container(default_wallet, cid, self.shell, self.cluster.default_rpc_endpoint)

    @allure.title("GRPC metrics object operations")
    def test_grpc_metrics_object_operations(
        self, simple_object_size: ObjectSize, default_wallet: WalletInfo, cluster: Cluster, disable_policer
    ):
        file_path = generate_file(simple_object_size.value)
        placement_policy = "REP 2 IN X CBF 1 SELECT 4 FROM * AS X"

        with reporter.step("Select random node"):
            node = random.choice(cluster.cluster_nodes)

        with reporter.step(f"Create container with policy {placement_policy}"):
            cid = create_container(default_wallet, self.shell, node.storage_node.get_rpc_endpoint(), placement_policy)

        with reporter.step("Get current gRPC metrics for method 'Put'"):
            metrics_counter_put = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ObjectService", method="Put"
            )

        with reporter.step("Put object to selected node"):
            oid = put_object(default_wallet, file_path, cid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method 'Put', 'the counter should increase by 1'"):
            metrics_counter_put += 1
            self.check_metrics_in_node(
                node, metrics_counter_put, command="grpc_server_handled_total", service="ObjectService", method="Put"
            )

        with reporter.step("Get current gRPC metrics for method 'Get'"):
            metrics_counter_get = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ObjectService", method="Get"
            )

        with reporter.step(f"Get object"):
            get_object(default_wallet, cid, oid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method=Get, 'the counter should increase by 1'"):
            metrics_counter_get += 1
            self.check_metrics_in_node(
                node, metrics_counter_get, command="grpc_server_handled_total", service="ObjectService", method="Get"
            )

        with reporter.step("Get current gRPC metrics for method 'Search'"):
            metrics_counter_search = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ObjectService", method="Search"
            )

        with reporter.step(f"Search object"):
            search_object(default_wallet, cid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method=Search, 'the counter should increase by 1'"):
            metrics_counter_search += 1
            self.check_metrics_in_node(
                node,
                metrics_counter_search,
                command="grpc_server_handled_total",
                service="ObjectService",
                method="Search",
            )

        with reporter.step("Get current gRPC metrics for method 'Head'"):
            metrics_counter_head = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="ObjectService", method="Head"
            )

        with reporter.step(f"Head object"):
            head_object(default_wallet, cid, oid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics method=Head, 'the counter should increase by 1'"):
            metrics_counter_head += 1
            self.check_metrics_in_node(
                node, metrics_counter_head, command="grpc_server_handled_total", service="ObjectService", method="Head"
            )

        with reporter.step("Delete container"):
            delete_container(default_wallet, cid, self.shell, self.cluster.default_rpc_endpoint)

    @allure.title("GRPC metrics Tree healthcheck")
    def test_grpc_metrics_tree_service(self, cluster: Cluster, healthcheck: Healthcheck):
        with reporter.step("Select random node"):
            node = random.choice(cluster.cluster_nodes)

        with reporter.step("Get current gRPC metrics for Healthcheck"):
            metrics_counter = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="TreeService", method="Healthcheck"
            )

        with reporter.step("Query Tree healthcheck status"):
            healthcheck.tree_healthcheck(node)

        with reporter.step(f"Check gRPC metrics for Healthcheck, 'the counter should increase'"):
            metrics_counter_new = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="TreeService", method="Healthcheck"
            )
            assert metrics_counter_new > metrics_counter, "the metrics has not increased"

    @allure.title("GRPC metrics Tree list")
    def test_grpc_metrics_tree_list(self, default_wallet: WalletInfo, cluster: Cluster):
        placement_policy = "REP 2 IN X CBF 1 SELECT 4 FROM * AS X"

        with reporter.step("Select random node"):
            node = random.choice(cluster.cluster_nodes)

        with reporter.step(f"Create container with policy {placement_policy}"):
            cid = create_container(default_wallet, self.shell, node.storage_node.get_rpc_endpoint(), placement_policy)

        with reporter.step("Get current gRPC metrics for Tree List"):
            metrics_counter = self.get_metrics_value(
                node, command="grpc_server_handled_total", service="TreeService", method="TreeList"
            )

        with reporter.step("Query Tree List"):
            get_tree_list(default_wallet, cid, self.shell, node.storage_node.get_rpc_endpoint())

        with reporter.step(f"Check gRPC metrics for Tree List, 'the counter should increase by 1'"):
            metrics_counter += 1
            self.check_metrics_in_node(
                node, metrics_counter, command="grpc_server_handled_total", service="TreeService", method="TreeList"
            )

        with reporter.step("Delete container"):
            delete_container(default_wallet, cid, self.shell, self.cluster.default_rpc_endpoint)