Move check_sbd to python from ruby

Assisted-by: ClaudeCode

Author: ApetorSkol

pcs/Makefile.am

@@ -198,6 +198,7 @@ EXTRA_DIST		= \
 			  common/services/drivers/openrc.py \
 			  common/services/drivers/systemd.py \
 			  common/services/drivers/sysvinit_rhel.py \
+			  common/sbd_dto.py \
 			  common/services_dto.py \
 			  common/services/errors.py \
 			  common/services/__init__.py \

pcs/cli/routing/stonith.py

@@ -1,10 +1,6 @@
 import pcs.cli.stonith.command as stonith_cli
 import pcs.cli.stonith.levels.command as levels_cli
-from pcs import (
-    resource,
-    stonith,
-    usage,
-)
+from pcs import resource, stonith, usage
 from pcs.cli.cib.element import command as cib_element_cmd
 from pcs.cli.common.routing import create_router
 
@@ -79,8 +75,6 @@
                     {
                         "list": stonith.sbd_watchdog_list,
                         "test": stonith.sbd_watchdog_test,
-                        # internal use only
-                        "list_json": stonith.sbd_watchdog_list_json,
                     },
                     ["stonith", "sbd", "watchdog"],
                 ),

pcs/common/sbd_dto.py

@@ -0,0 +1,26 @@
+from dataclasses import dataclass
+from typing import Optional
+
+from pcs.common.interface.dto import DataTransferObject
+from pcs.common.services_dto import ServiceStatusDto
+
+
+@dataclass(frozen=True)
+class SbdWatchdogStatusDto(DataTransferObject):
+    path: str
+    exists: bool
+    is_supported: bool
+
+
+@dataclass(frozen=True)
+class SbdDeviceStatusDto(DataTransferObject):
+    path: str
+    exists: bool
+    is_block_device: bool
+
+
+@dataclass(frozen=True)
+class SbdCheckResultDto(DataTransferObject):
+    sbd_service: ServiceStatusDto
+    watchdog: Optional[SbdWatchdogStatusDto] = None
+    device_list: Optional[list[SbdDeviceStatusDto]] = None

pcs/daemon/app/api_v0.py

@@ -16,6 +16,7 @@
     PermissionMetadataDependenciesDto,
     PermissionMetadataDto,
 )
+from pcs.common.sbd_dto import SbdCheckResultDto
 from pcs.common.str_tools import format_list
 from pcs.daemon import log
 from pcs.daemon.app.api_v0_tools import (
@@ -699,6 +700,56 @@ async def _handle_request(self) -> None:
             raise self._error(reports_to_str(result.reports))
 
 
+class CheckSbdHandler(_BaseApiV0Handler):
+    async def _handle_request(self) -> None:
+        watchdog = self.get_argument("watchdog", "")
+        device_list_json = self.get_argument("device_list", "[]")
+        try:
+            device_list = json.loads(device_list_json)
+        except json.JSONDecodeError as e:
+            raise self._error("Invalid input data format") from e
+        if not isinstance(device_list, list) or not all(
+            isinstance(device, str) for device in device_list
+        ):
+            raise self._error("Invalid input data format")
+
+        result = await self._run_library_command(
+            "sbd.check_sbd",
+            dict(watchdog=watchdog, device_list=device_list),
+        )
+        if not result.success:
+            raise self._error(reports_to_str(result.reports))
+        # Not using to_dict to keep backward compatibility with older
+        # pcs versions which don't expect watchdog and device_list keys
+        # to be present in the response when they have no values. Also the
+        # response was different than our DTOs so we make them into legacy
+        # format by hand.
+        sbd_status = cast(SbdCheckResultDto, result.result)
+        response: dict[str, Any] = dict(
+            sbd=dict(
+                installed=sbd_status.sbd_service.installed,
+                enabled=sbd_status.sbd_service.enabled,
+                running=sbd_status.sbd_service.running,
+            ),
+        )
+        if sbd_status.watchdog is not None:
+            response["watchdog"] = dict(
+                path=sbd_status.watchdog.path,
+                exist=sbd_status.watchdog.exists,
+                is_supported=sbd_status.watchdog.is_supported,
+            )
+        if sbd_status.device_list is not None:
+            response["device_list"] = [
+                dict(
+                    path=device.path,
+                    exist=device.exists,
+                    block_device=device.is_block_device,
+                )
+                for device in sbd_status.device_list
+            ]
+        self.write(json.dumps(response))
+
+
 class SetCorosyncConf(_BaseApiV0Handler):
     async def _handle_request(self) -> None:
         self._check_required_params({"corosync_conf"})
@@ -806,6 +857,7 @@ def r(url: str) -> str:
         # cluster config
         (r("set_corosync_conf"), SetCorosyncConf, params),
         # sbd
+        (r("check_sbd"), CheckSbdHandler, params),
         (r("get_sbd_config"), GetSbdConfigHandler, params),
         (r("set_sbd_config"), SetSbdConfigHandler, params),
     ]

pcs/daemon/async_tasks/worker/command_mapping.py

@@ -491,6 +491,10 @@ class _Cmd:
         cmd=services.pacemaker_remote_off_local,
         required_permission=p.WRITE,
     ),
+    "sbd.check_sbd": _Cmd(
+        cmd=sbd.check_sbd,
+        required_permission=p.READ,
+    ),
     "sbd.disable_sbd": _Cmd(
         cmd=sbd.disable_sbd,
         required_permission=p.WRITE,
@@ -570,6 +574,7 @@ class _Cmd:
     "resource_agent.list_standards",
     # The sbd URLs are ready to be exposed in APIv2, just waiting for all the
     # other URLs to get moved to APIv2
+    "sbd.check_sbd",
     "sbd.get_node_sbd_config_text",
     "sbd.set_node_sbd_config_text",
     # There is a lot of url handlers managing cluster services and they should

pcs/lib/commands/sbd.py

@@ -3,12 +3,15 @@
 from pcs import settings
 from pcs.common import reports
 from pcs.common.node_communicator import RequestTarget
+from pcs.common.sbd_dto import (
+    SbdCheckResultDto,
+    SbdDeviceStatusDto,
+    SbdWatchdogStatusDto,
+)
+from pcs.common.services_dto import ServiceStatusDto
 from pcs.common.types import StringSequence
 from pcs.common.validate import is_integer
-from pcs.lib import (
-    sbd,
-    validate,
-)
+from pcs.lib import sbd, validate
 from pcs.lib.cib.tools import get_resources
 from pcs.lib.communication.nodes import GetOnlineTargets
 from pcs.lib.communication.sbd import (
@@ -153,6 +156,57 @@ def _get_full_target_dict[T](
     }
 
 
+def check_sbd(
+    lib_env: LibraryEnvironment,
+    watchdog: Optional[str] = None,
+    device_list: Optional[StringSequence] = None,
+) -> SbdCheckResultDto:
+    """
+    Check whether sbd is installed, enabled and running.
+    - If watchdog is specified, also check whether the watchdog exists on the
+      local node.
+    - If device_list is specified, also check whether paths specified in the
+      list exist on the local node and if they are block devices.
+
+    watchdog -- watchdog path to check
+    device_list -- list of paths to check
+    """
+    sbd_status = ServiceStatusDto(
+        service=settings.sbd_service_name,
+        installed=lib_env.service_manager.is_installed(
+            settings.sbd_service_name
+        ),
+        enabled=lib_env.service_manager.is_enabled(settings.sbd_service_name),
+        running=lib_env.service_manager.is_running(settings.sbd_service_name),
+    )
+
+    watchdog_dto: Optional[SbdWatchdogStatusDto] = None
+    if watchdog:
+        available_watchdogs = sbd.get_available_watchdogs(lib_env.cmd_runner())
+        exists = watchdog in available_watchdogs
+        # The support status provided by sbd is unreliable, so we are reporting
+        # every device as supported for now
+        # https://bugzilla.redhat.com/show_bug.cgi?id=1578891
+        # https://github.com/ClusterLabs/pcs/commit/1ca412e151bf531533a525640899fbd109aa
+        watchdog_dto = SbdWatchdogStatusDto(
+            path=watchdog,
+            exists=exists,
+            is_supported=exists,
+        )
+
+    device_list_dto: Optional[list[SbdDeviceStatusDto]] = None
+    if device_list:
+        device_list_dto = [
+            sbd.check_sbd_device_exists(device) for device in device_list
+        ]
+
+    return SbdCheckResultDto(
+        sbd_service=sbd_status,
+        watchdog=watchdog_dto,
+        device_list=device_list_dto,
+    )
+
+
 def enable_sbd(  # noqa: PLR0913
     lib_env: LibraryEnvironment,
     default_watchdog: Optional[str],

pcs/lib/sbd.py

@@ -1,25 +1,20 @@
 import fcntl
+import os
 import re
-from os import path
-from typing import (
-    Mapping,
-    Optional,
-    Union,
-)
+import stat
+from typing import Mapping, Optional, Union
 
 from pcs import settings
 from pcs.common import reports
+from pcs.common.sbd_dto import SbdDeviceStatusDto
 from pcs.common.services.interfaces import ServiceManagerInterface
 from pcs.common.types import StringSequence
 from pcs.common.validate import is_integer
 from pcs.lib import validate
 from pcs.lib.corosync.config_facade import ConfigFacade as CorosyncConfFacade
 from pcs.lib.errors import LibraryError
 from pcs.lib.external import CommandRunner
-from pcs.lib.tools import (
-    dict_to_environment_file,
-    environment_file_to_dict,
-)
+from pcs.lib.tools import dict_to_environment_file, environment_file_to_dict
 
 DEVICE_INITIALIZATION_OPTIONS_MAPPING = {
     "watchdog-timeout": "-1",
@@ -212,7 +207,7 @@ def validate_nodes_devices(
                 reports.messages.SbdDevicePathNotAbsolute(device, node_label)
             )
             for device in device_list
-            if not device or not path.isabs(device)
+            if not device or not os.path.isabs(device)
         )
     return report_item_list
 
@@ -341,11 +336,34 @@ def initialize_block_devices(
     )
 
 
+def check_sbd_device_exists(
+    device: str,
+) -> SbdDeviceStatusDto:
+    """
+    Check whether a path exists on the local node and if it is a block device.
+
+    device -- device path to be checked
+    """
+    try:
+        mode = os.stat(device).st_mode
+        return SbdDeviceStatusDto(
+            path=device,
+            exists=True,
+            is_block_device=stat.S_ISBLK(mode),
+        )
+    except OSError:
+        return SbdDeviceStatusDto(
+            path=device,
+            exists=False,
+            is_block_device=False,
+        )
+
+
 def get_local_sbd_device_list() -> list[str]:
     """
     Returns list of devices specified in local SBD config
     """
-    if not path.exists(settings.sbd_config):
+    if not os.path.exists(settings.sbd_config):
         return []
 
     cfg = environment_file_to_dict(get_local_sbd_config())
@@ -403,7 +421,7 @@ def _get_local_sbd_watchdog_timeout() -> int:
     """
     Return the value of SBD_WATCHDOG_TIMEOUT used in local SBD config
     """
-    if not path.exists(settings.sbd_config):
+    if not os.path.exists(settings.sbd_config):
         return _DEFAULT_SBD_WATCHDOG_TIMEOUT
 
     cfg = environment_file_to_dict(get_local_sbd_config())

pcs/stonith.py

@@ -380,18 +380,6 @@ def sbd_watchdog_list(lib: Any, argv: Argv, modifiers: InputModifiers) -> None:
         print_to_stderr("No available watchdog")
 
 
-def sbd_watchdog_list_json(
-    lib: Any, argv: Argv, modifiers: InputModifiers
-) -> None:
-    """
-    Options: no options
-    """
-    modifiers.ensure_only_supported()
-    if argv:
-        raise CmdLineInputError()
-    print(json.dumps(lib.sbd.get_local_available_watchdogs()))
-
-
 def sbd_watchdog_test(lib: Any, argv: Argv, modifiers: InputModifiers) -> None:
     """
     Options:

pcs_test/Makefile.am

@@ -343,6 +343,7 @@ EXTRA_DIST		= \
 			  tier0/lib/commands/resource/test_restart.py \
 			  tier0/lib/commands/resource/test_stop.py \
 			  tier0/lib/commands/sbd/__init__.py \
+			  tier0/lib/commands/sbd/test_check_sbd.py \
 			  tier0/lib/commands/sbd/test_disable_sbd.py \
 			  tier0/lib/commands/sbd/test_enable_sbd.py \
 			  tier0/lib/commands/sbd/test_get_cluster_sbd_config.py \