Add 'crm sbd' sub-level (jsc#PED-8256)

crmsh/bootstrap.py

@@ -44,6 +44,8 @@
 from .sh import ShellUtils
 from .ui_node import NodeMgmt
 from .user_of_host import UserOfHost, UserNotFoundError
+from .sbd import SBDUtils, SBDManager, SBDTimeout
+from . import watchdog
 import crmsh.healthcheck
 
 
@@ -56,21 +58,18 @@
 COROSYNC_AUTH = "/etc/corosync/authkey"
 CRM_CFG = "/etc/crm/crm.conf"
 PROFILES_FILE = "/etc/crm/profiles.yml"
-SYSCONFIG_SBD = "/etc/sysconfig/sbd"
 SYSCONFIG_PCMK = "/etc/sysconfig/pacemaker"
 SYSCONFIG_NFS = "/etc/sysconfig/nfs"
 PCMK_REMOTE_AUTH = "/etc/pacemaker/authkey"
 COROSYNC_CONF_ORIG = tmpfiles.create()[1]
 SERVICES_STOP_LIST = ["corosync-qdevice.service", "corosync.service", "hawk.service", CSYNC2_SERVICE]
-WATCHDOG_CFG = "/etc/modules-load.d/watchdog.conf"
 BOOTH_DIR = "/etc/booth"
 BOOTH_CFG = "/etc/booth/booth.conf"
 BOOTH_AUTH = "/etc/booth/authkey"
-SBD_SYSTEMD_DELAY_START_DIR = "/etc/systemd/system/sbd.service.d"
 FILES_TO_SYNC = (BOOTH_DIR, corosync.conf(), COROSYNC_AUTH, CSYNC2_CFG, CSYNC2_KEY, "/etc/ctdb/nodes",
         "/etc/drbd.conf", "/etc/drbd.d", "/etc/ha.d/ldirectord.cf", "/etc/lvm/lvm.conf", "/etc/multipath.conf",
-        "/etc/samba/smb.conf", SYSCONFIG_NFS, SYSCONFIG_PCMK, SYSCONFIG_SBD, PCMK_REMOTE_AUTH, WATCHDOG_CFG,
-        PROFILES_FILE, CRM_CFG, SBD_SYSTEMD_DELAY_START_DIR)
+        "/etc/samba/smb.conf", SYSCONFIG_NFS, SYSCONFIG_PCMK, SBDManager.SYSCONFIG_SBD, PCMK_REMOTE_AUTH, watchdog.Watchdog.WATCHDOG_CFG,
+        PROFILES_FILE, CRM_CFG, SBDManager.SBD_SYSTEMD_DELAY_START_DIR)
 
 INIT_STAGES_EXTERNAL = ("ssh", "csync2", "corosync", "sbd", "cluster", "ocfs2", "admin", "qdevice")
 INIT_STAGES_INTERNAL = ("csync2_remote", "qnetd_remote", "remote_auth")
@@ -136,7 +135,7 @@
         self.profiles_dict = {}
         self.default_nic = None
         self.default_ip_list = []
-        self.rm_list = [SYSCONFIG_SBD, CSYNC2_CFG, corosync.conf(), CSYNC2_KEY,
+        self.rm_list = [SBDManager.SYSCONFIG_SBD, CSYNC2_CFG, corosync.conf(), CSYNC2_KEY,
                 COROSYNC_AUTH, "/var/lib/heartbeat/crm/*", "/var/lib/pacemaker/cib/*",
                 "/var/lib/corosync/*", "/var/lib/pacemaker/pengine/*", PCMK_REMOTE_AUTH,
                 "/var/lib/csync2/*", "~/.config/crm/*"]
@@ -215,12 +214,21 @@
         """
         Validate sbd options
         """
+        with_sbd_option = self.sbd_devices or self.diskless_sbd
+        sbd_installed = utils.package_is_installed("sbd")
+
+        if with_sbd_option and not sbd_installed:
+            utils.fatal(SBDManager.SBD_NOT_INSTALLED_MSG)
         if self.sbd_devices and self.diskless_sbd:
             utils.fatal("Can't use -s and -S options together")
+        if self.sbd_devices:
+            SBDUtils.verify_sbd_device(self.sbd_devices)
         if self.stage == "sbd":
-            if not self.sbd_devices and not self.diskless_sbd and self.yes_to_all:
+            if not sbd_installed:
+                utils.fatal(SBDManager.SBD_NOT_INSTALLED_MSG)
+            if not with_sbd_option and self.yes_to_all:
                 utils.fatal("Stage sbd should specify sbd device by -s or diskless sbd by -S option")
-            if ServiceManager().service_is_active("sbd.service") and not config.core.force:
+            if ServiceManager().service_is_active(constants.SBD_SERVICE) and not config.core.force:
                 utils.fatal("Can't configure stage sbd: sbd.service already running! Please use crm option '-F' if need to redeploy")
             if self.cluster_is_running:
                 utils.check_all_nodes_reachable()
@@ -297,8 +305,7 @@
         self._validate_sbd_option()
 
     def init_sbd_manager(self):
-        from .sbd import SBDManager
-        self.sbd_manager = SBDManager(self)
+        self.sbd_manager = SBDManager(bootstrap_context=self)
 
     def detect_platform(self):
         """
@@ -400,7 +407,7 @@
 
 
 def confirm(msg):
-    if _context.yes_to_all:
+    if config.core.force or (_context and _context.yes_to_all):
         return True
     disable_completion()
     rc = logger_utils.confirm(msg)
@@ -410,12 +417,12 @@
 
 
 def disable_completion():
-    if _context.ui_context:
+    if _context and _context.ui_context:
         _context.ui_context.disable_completion()
 
 
 def enable_completion():
-    if _context.ui_context:
+    if _context and _context.ui_context:
         _context.ui_context.setup_readline()
 
 
@@ -500,7 +507,7 @@
         return False
 
     # if peer_node is None, this is in the init process
-    if _context.cluster_node is None:
+    if not _context or _context.cluster_node is None:
         return True
     # In join process
     # If the joining node is already online but can't find the init node
@@ -769,11 +776,10 @@
 
     Return success node list
     """
-    from .sbd import SBDTimeout
     # not _context means not in init or join process
     if not _context and \
             utils.package_is_installed("sbd") and \
-            ServiceManager().service_is_enabled("sbd.service") and \
+            ServiceManager().service_is_enabled(constants.SBD_SERVICE) and \
             SBDTimeout.is_sbd_delay_start():
         target_dir = "/run/systemd/system/sbd.service.d/"
         cmd1 = "mkdir -p {}".format(target_dir)
@@ -1400,7 +1406,7 @@
     import crmsh.sbd
     if _context.stage == "sbd":
         crmsh.sbd.clean_up_existing_sbd_resource()
-    _context.sbd_manager.sbd_init()
+    _context.sbd_manager.init_and_deploy_sbd()
 
 
 def init_upgradeutil():
@@ -1437,7 +1443,9 @@
 rsc_defaults rsc-options: resource-stickiness=1 migration-threshold=3
 """)
 
-    _context.sbd_manager.configure_sbd_resource_and_properties()
+    if ServiceManager().service_is_enabled(constants.SBD_SERVICE):
+        _context.sbd_manager.configure_sbd()
+
 
 
 def init_admin():
@@ -2089,8 +2097,7 @@
     shell.get_stdout_or_raise_error("rm -f {}".format(' '.join(_context.rm_list)), remote)
     # restore original sbd configuration file from /usr/share/fillup-templates/sysconfig.sbd
     if utils.package_is_installed("sbd", remote_addr=remote):
-        from .sbd import SBDManager
-        cmd = "cp {} {}".format(SBDManager.SYSCONFIG_SBD_TEMPLATE, SYSCONFIG_SBD)
+        cmd = "cp {} {}".format(SBDManager.SYSCONFIG_SBD_TEMPLATE, SBDManager.SYSCONFIG_SBD)
         shell.get_stdout_or_raise_error(cmd, remote)
 
 
@@ -2737,8 +2744,7 @@
     """
     Adjust stonith-timeout for sbd and other scenarios
     """
-    if ServiceManager().service_is_active("sbd.service"):
-        from .sbd import SBDTimeout
+    if ServiceManager().service_is_active(constants.SBD_SERVICE):
         SBDTimeout.adjust_sbd_timeout_related_cluster_configuration()
     else:
         value = get_stonith_timeout_generally_expected()
@@ -2801,7 +2807,12 @@
     """
     Sync files between cluster nodes
     """
-    if _context.skip_csync2:
+    if _context:
+        skip_csync2 = _context.skip_csync2
+    else:
+        skip_csync2 = not ServiceManager().service_is_active(CSYNC2_SERVICE)
+
+    if skip_csync2:
         utils.cluster_copy_file(path, nodes=_context.node_list_in_cluster, output=False)
     else:
         csync2_update(path)

crmsh/constants.py

@@ -450,4 +450,9 @@
 HIDDEN_COMMANDS = {'ms'}
 
 NO_SSH_ERROR_MSG = "ssh-related operations are disabled. crmsh works in local mode."
+
+PCMK_SERVICE = "pacemaker.service"
+SBD_SERVICE = "sbd.service"
+
+SHOW_SBD_START_TIMEOUT_CMD = "systemctl show -p TimeoutStartUSec sbd.service --value"
 # vim:ts=4:sw=4:et:

crmsh/ocfs2.py

@@ -119,7 +119,7 @@ def _check_sbd_and_ocfs2_dev(self):
         """
         from . import sbd
         if ServiceManager().service_is_enabled("sbd.service"):
-            sbd_device_list = sbd.SBDManager.get_sbd_device_from_config()
+            sbd_device_list = sbd.SBDUtils.get_sbd_device_from_config()
             for dev in self.ocfs2_devices:
                 if dev in sbd_device_list:
                     self._dynamic_raise_error("{} cannot be the same with SBD device".format(dev))

crmsh/qdevice.py

@@ -15,6 +15,7 @@
 from . import lock
 from . import log
 from .service_manager import ServiceManager
+from .sbd import SBDManager, SBDTimeout, SBDUtils
 
 
 logger = log.setup_logger(__name__)
@@ -614,15 +615,14 @@ def adjust_sbd_watchdog_timeout_with_qdevice(self):
         """
         Adjust SBD_WATCHDOG_TIMEOUT when configuring qdevice and diskless SBD
         """
-        from .sbd import SBDManager, SBDTimeout
         utils.check_all_nodes_reachable()
-        self.using_diskless_sbd = SBDManager.is_using_diskless_sbd()
+        self.using_diskless_sbd = SBDUtils.is_using_diskless_sbd()
         # add qdevice after diskless sbd started
         if self.using_diskless_sbd:
-            res = SBDManager.get_sbd_value_from_config("SBD_WATCHDOG_TIMEOUT")
+            res = SBDUtils.get_sbd_value_from_config("SBD_WATCHDOG_TIMEOUT")
             if not res or int(res) < SBDTimeout.SBD_WATCHDOG_TIMEOUT_DEFAULT_WITH_QDEVICE:
                 sbd_watchdog_timeout_qdevice = SBDTimeout.SBD_WATCHDOG_TIMEOUT_DEFAULT_WITH_QDEVICE
-                SBDManager.update_configuration({"SBD_WATCHDOG_TIMEOUT": str(sbd_watchdog_timeout_qdevice)})
+                SBDManager.update_sbd_configuration({"SBD_WATCHDOG_TIMEOUT": str(sbd_watchdog_timeout_qdevice)})
                 utils.set_property("stonith-timeout", SBDTimeout.get_stonith_timeout())
 
     @qnetd_lock_for_same_cluster_name

crmsh/report/collect.py

@@ -447,11 +447,16 @@ def collect_sbd_info(context: core.Context) -> None:
         return
 
     sbd_f = os.path.join(context.work_dir, constants.SBD_F)
-    cmd = ". {};export SBD_DEVICE;{};{}".format(constants.SBDCONF, "sbd dump", "sbd list")
+    cmd_list = [
+        f". {constants.SBDCONF};export SBD_DEVICE;sbd dump;sbd list",
+        "crm sbd configure show",
+        "crm sbd status"
+    ]
     with open(sbd_f, "w") as f:
-        f.write("\n\n#=====[ Command ] ==========================#\n")
-        f.write(f"# {cmd}\n")
-        f.write(utils.get_cmd_output(cmd))
+        for cmd in cmd_list:
+            f.write("\n\n#=====[ Command ] ==========================#\n")
+            f.write(f"# {cmd}\n")
+            f.write(utils.get_cmd_output(cmd))
 
     logger.debug(f"Dump SBD config file into {utils.real_path(sbd_f)}")
 

crmsh/report/utils.py

@@ -739,7 +739,7 @@ def get_cmd_output(cmd: str, timeout: int = None) -> str:
         out_str += f"{out}\n"
     if err:
         out_str += f"{err}\n"
-    return out_str
+    return crmutils.strip_ansi_escape_sequences(out_str)
 
 
 def get_timespan_str(context: core.Context) -> str: