[#130] Add service healthcheck and allow to skip version check for some binaries

Signed-off-by: Andrey Berezin <a.berezin@yadro.com>

src/frostfs_testlib/healthcheck/basic_healthcheck.py

@@ -6,8 +6,9 @@ from frostfs_testlib.reporter import get_reporter
 from frostfs_testlib.resources.cli import FROSTFS_CLI_EXEC
 from frostfs_testlib.shell import CommandOptions
 from frostfs_testlib.steps.node_management import storage_node_healthcheck
-from frostfs_testlib.storage.cluster import ClusterNode
+from frostfs_testlib.storage.cluster import ClusterNode, ServiceClass
 from frostfs_testlib.testing.test_control import wait_for_success
+from frostfs_testlib.utils.failover_utils import check_services_status
 
 reporter = get_reporter()
 
@@ -51,6 +52,26 @@ class BasicHealthcheck(Healthcheck):
         with reporter.step(f"Perform storage healthcheck on {cluster_node}"):
             self._perform(cluster_node, checks)
 
+    @wait_for_success(120, 5)
+    def services_healthcheck(self, cluster_node: ClusterNode):
+        svcs_to_check = cluster_node.services
+        checks = {
+            check_services_status: {
+                "service_list": svcs_to_check,
+                "expected_status": "active",
+            },
+            self._check_services: {"services": svcs_to_check},
+        }
+
+        with reporter.step(f"Perform service healthcheck on {cluster_node}"):
+            self._perform(cluster_node, checks)
+
+    def _check_services(self, cluster_node: ClusterNode, services: list[ServiceClass]):
+        for svc in services:
+            result = svc.service_healthcheck()
+            if result == False:
+                return f"Service {svc.get_service_systemctl_name()} healthcheck failed on node {cluster_node}."
+
     @reporter.step_deco("Storage healthcheck on {cluster_node}")
     def _storage_healthcheck(self, cluster_node: ClusterNode) -> str | None:
         result = storage_node_healthcheck(cluster_node.storage_node)

src/frostfs_testlib/healthcheck/interfaces.py

@@ -15,3 +15,7 @@ class Healthcheck(ABC):
     @abstractmethod
     def storage_healthcheck(self, cluster_node: ClusterNode):
         """Perform storage service healthcheck on target cluster node"""
+
+    @abstractmethod
+    def services_healthcheck(self, cluster_node: ClusterNode):
+        """Perform service status check on target cluster node"""

src/frostfs_testlib/storage/cluster.py

@@ -116,8 +116,24 @@ class ClusterNode:
             self.host,
         )
 
-    def get_list_of_services(self) -> list[str]:
-        return [config.attributes[ConfigAttributes.SERVICE_NAME] for config in self.host.config.services]
+    @property
+    def services(self) -> list[NodeBase]:
+        svcs: list[NodeBase] = []
+        svcs_names_on_node = [svc.name for svc in self.host.config.services]
+        for entry in self.class_registry._class_mapping.values():
+            hosting_svc_name = entry["hosting_service_name"]
+            pattern = f"{hosting_svc_name}{self.id:02}"
+            if pattern in svcs_names_on_node:
+                config = self.host.get_service_config(pattern)
+                svcs.append(
+                    entry["cls"](
+                        self.id,
+                        config.name,
+                        self.host,
+                    )
+                )
+
+        return svcs
 
     def get_all_interfaces(self) -> dict[str, str]:
         return self.host.config.interfaces

src/frostfs_testlib/utils/failover_utils.py

@@ -11,6 +11,7 @@ from frostfs_testlib.steps.node_management import storage_node_healthcheck
 from frostfs_testlib.steps.storage_policy import get_nodes_with_object
 from frostfs_testlib.storage.cluster import Cluster, ClusterNode, NodeBase, StorageNode
 from frostfs_testlib.storage.dataclasses.frostfs_services import MorphChain
+from frostfs_testlib.storage.dataclasses.node_base import ServiceClass
 from frostfs_testlib.testing.test_control import wait_for_success
 from frostfs_testlib.utils.datetime_utils import parse_time
 
@@ -96,11 +97,11 @@ def multiple_restart(
 
 @reporter.step_deco("Get status of list of services and check expected status")
 @wait_for_success(60, 5)
-def check_services_status(service_list: list[str], expected_status: str, shell: Shell):
+def check_services_status(cluster_node: ClusterNode, service_list: list[ServiceClass], expected_status: str):
     cmd = ""
     for service in service_list:
-        cmd += f' sudo systemctl status {service} --lines=0 | grep "Active:";'
-    result = shell.exec(cmd).stdout.rstrip()
+        cmd += f' sudo systemctl status {service.get_service_systemctl_name()} --lines=0 | grep "Active:";'
+    result = cluster_node.host.get_shell().exec(cmd).stdout.rstrip()
     statuses = list()
     for line in result.split("\n"):
         status_substring = line.split()

src/frostfs_testlib/utils/version_utils.py

@@ -3,12 +3,7 @@ import re
 
 from frostfs_testlib.cli import FrostfsAdm, FrostfsCli
 from frostfs_testlib.hosting import Hosting
-from frostfs_testlib.resources.cli import (
-    FROSTFS_ADM_EXEC,
-    FROSTFS_AUTHMATE_EXEC,
-    FROSTFS_CLI_EXEC,
-    NEOGO_EXECUTABLE,
-)
+from frostfs_testlib.resources.cli import FROSTFS_ADM_EXEC, FROSTFS_AUTHMATE_EXEC, FROSTFS_CLI_EXEC, NEOGO_EXECUTABLE
 from frostfs_testlib.resources.common import DEFAULT_WALLET_CONFIG
 from frostfs_testlib.shell import Shell
 
@@ -44,36 +39,44 @@ def get_remote_binaries_versions(hosting: Hosting) -> dict[str, str]:
         binary_path_by_name = {}  # Maps binary name to executable path
         for service_config in host.config.services:
             exec_path = service_config.attributes.get("exec_path")
+            requires_check = service_config.attributes.get("requires_version_check", "true")
             if exec_path:
-                binary_path_by_name[service_config.name] = exec_path
+                binary_path_by_name[service_config.name] = {
+                    "exec_path": exec_path,
+                    "check": requires_check.lower() == "true",
+                }
         for cli_config in host.config.clis:
-            binary_path_by_name[cli_config.name] = cli_config.exec_path
+            requires_check = cli_config.attributes.get("requires_version_check", "true")
+            binary_path_by_name[cli_config.name] = {
+                "exec_path": cli_config.exec_path,
+                "check": requires_check.lower() == "true",
+            }
 
         shell = host.get_shell()
         versions_at_host = {}
-        for binary_name, binary_path in binary_path_by_name.items():
+        for binary_name, binary in binary_path_by_name.items():
             try:
+                binary_path = binary["exec_path"]
                 result = shell.exec(f"{binary_path} --version")
-                versions_at_host[binary_name] = _parse_version(result.stdout)
+                versions_at_host[binary_name] = {"version": _parse_version(result.stdout), "check": binary["check"]}
             except Exception as exc:
                 logger.error(f"Cannot get version for {binary_path} because of\n{exc}")
-                versions_at_host[binary_name] = "Unknown"
+                versions_at_host[binary_name] = {"version": "Unknown", "check": binary["check"]}
         versions_by_host[host.config.address] = versions_at_host
 
     # Consolidate versions across all hosts
     versions = {}
     for host, binary_versions in versions_by_host.items():
-        for name, version in binary_versions.items():
-            captured_version = versions.get(name)
+        for name, binary in binary_versions.items():
+            captured_version = versions.get(name, {}).get("version")
+            version = binary["version"]
             if captured_version:
-                assert (
-                    captured_version == version
-                ), f"Binary {name} has inconsistent version on host {host}"
+                assert captured_version == version, f"Binary {name} has inconsistent version on host {host}"
             else:
-                versions[name] = version
+                versions[name] = {"version": version, "check": binary["check"]}
     return versions
 
 
 def _parse_version(version_output: str) -> str:
     version = re.search(r"version[:\s]*v?(.+)", version_output, re.IGNORECASE)
-    return version.group(1).strip() if version else "Unknown"
+    return version.group(1).strip() if version else version_output