src/frostfs_testlib/hosting/docker_host.py

@@ -11,7 +11,7 @@ import docker
 from requests import HTTPError
 
 from frostfs_testlib.hosting.config import ParsedAttributes
-from frostfs_testlib.hosting.interfaces import DiskInfo, Host
+from frostfs_testlib.hosting.interfaces import DiskInfo, Host, HostStatus
 from frostfs_testlib.shell import LocalShell, Shell, SSHShell
 from frostfs_testlib.shell.command_inspectors import SudoInspector
 
@@ -87,6 +87,15 @@ class DockerHost(Host):
         for service_config in self._config.services:
             self.start_service(service_config.name)
 
+    def get_host_status(self) -> HostStatus:
+        # We emulate host status by checking all services.
+        for service_config in self._config.services:
+            state = self._get_container_state(service_config.name)
+            if state != "running":
+                return HostStatus.OFFLINE
+
+        return HostStatus.ONLINE
+
     def stop_host(self) -> None:
         # We emulate stopping machine by stopping all services
         # As an alternative we can probably try to stop docker service...
@@ -293,11 +302,16 @@ class DockerHost(Host):
         # To speed things up, we break timeout in smaller iterations and check container state
         # several times. This way waiting stops as soon as container reaches the expected state
         for _ in range(iterations):
-            container = self._get_container_by_name(container_name)
+            state = self._get_container_state(container_name)
-            logger.debug(f"Current container state\n:{json.dumps(container, indent=2)}")
 
-            if container and container["State"] == expected_state:
+            if state == expected_state:
                 return
             time.sleep(iteration_wait_time)
 
         raise RuntimeError(f"Container {container_name} is not in {expected_state} state.")
+
+    def _get_container_state(self, container_name: str) -> str:
+        container = self._get_container_by_name(container_name)
+        logger.debug(f"Current container state\n:{json.dumps(container, indent=2)}")
+
+        return container.get("State", None)

src/frostfs_testlib/hosting/interfaces.py

@@ -4,6 +4,13 @@ from typing import Optional
 
 from frostfs_testlib.hosting.config import CLIConfig, HostConfig, ServiceConfig
 from frostfs_testlib.shell.interfaces import Shell
+from frostfs_testlib.testing.readable import HumanReadableEnum
+
+
+class HostStatus(HumanReadableEnum):
+    ONLINE = "Online"
+    OFFLINE = "Offline"
+    UNKNOWN = "Unknown"
 
 
 class DiskInfo(dict):
@@ -79,6 +86,10 @@ class Host(ABC):
     def start_host(self) -> None:
         """Starts the host machine."""
 
+    @abstractmethod
+    def get_host_status(self) -> HostStatus:
+        """Check host status."""
+
     @abstractmethod
     def stop_host(self, mode: str) -> None:
         """Stops the host machine.

src/frostfs_testlib/storage/constants.py

@@ -10,6 +10,7 @@ class ConfigAttributes:
     ENDPOINT_DATA_0 = "endpoint_data0"
     ENDPOINT_DATA_1 = "endpoint_data1"
     ENDPOINT_INTERNAL = "endpoint_internal0"
+    ENDPOINT_PROMETHEUS = "endpoint_prometheus"
     CONTROL_ENDPOINT = "control_endpoint"
     UN_LOCODE = "un_locode"
     HTTP_HOSTNAME = "http_hostname"

src/frostfs_testlib/storage/dataclasses/node_base.py

@@ -64,6 +64,9 @@ class NodeBase(HumanReadableABC):
     def service_healthcheck(self) -> bool:
         """Service healthcheck."""
 
+    def get_metrics_endpoint(self) -> str:
+        return self._get_attribute(ConfigAttributes.ENDPOINT_PROMETHEUS)
+
     def stop_service(self):
         with reporter.step(f"Stop {self.name} service on {self.host.config.address}"):
             self.host.stop_service(self.name)