[crmsh-5.0] Dev: sbd: Remove sbd configuration directories while removing cluster node (#1976)

Backport from #1962

Author: liangxin1300

crmsh/bootstrap.py

@@ -138,7 +138,7 @@ def __init__(self):
         self.default_nic = None
         self.default_ip_list = []
         self.corosync_conf_orig = None
-        self.rm_list = [sbd.SBDManager.SYSCONFIG_SBD, CSYNC2_CFG, corosync.conf(), CSYNC2_KEY,
+        self.rm_list = [CSYNC2_CFG, corosync.conf(), CSYNC2_KEY,
                 COROSYNC_AUTH, "/var/lib/pacemaker/cib/*",
                 "/var/lib/corosync/*", "/var/lib/pacemaker/pengine/*", PCMK_REMOTE_AUTH,
                 "/var/lib/csync2/*", "~/.config/crm/*"]
@@ -2220,10 +2220,8 @@ def rm_configuration_files(remote=None):
     """
     shell = sh.cluster_shell()
     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):
-        cmd = "cp {} {}".format(sbd.SBDManager.SYSCONFIG_SBD_TEMPLATE, sbd.SBDManager.SYSCONFIG_SBD)
-        shell.get_stdout_or_raise_error(cmd, remote)
+    if os.path.exists(sbd.SBDManager.SYSCONFIG_SBD):
+        sbd.cleanup_sbd_configurations(remote)
 
 
 def remove_node_from_cluster(node, dead_node=False):

crmsh/sbd.py

@@ -9,7 +9,6 @@
 from . import corosync
 from . import xmlutil
 from . import watchdog
-from . import parallax
 from .service_manager import ServiceManager
 from .sh import ShellUtils
 
@@ -787,6 +786,19 @@ def cleanup_existing_sbd_resource():
         utils.ext_cmd("crm configure delete {}".format(' '.join(sbd_id_list)))
 
 
+def cleanup_sbd_configurations(remote=None):
+    shell = sh.cluster_shell()
+    sysconfig_sbd_bak = f"{SBDManager.SYSCONFIG_SBD}.bak"
+    logger.info("Rename %s to %s on node %s",
+                SBDManager.SYSCONFIG_SBD, sysconfig_sbd_bak, remote or utils.this_node())
+    cmd = f"test -f {SBDManager.SYSCONFIG_SBD} && mv {SBDManager.SYSCONFIG_SBD} {sysconfig_sbd_bak} || exit 0"
+    shell.get_stdout_or_raise_error(cmd, host=remote)
+
+    for _dir in [SBDManager.SBD_SYSTEMD_DELAY_START_DIR, SBDManager.SBD_SYSTEMD_DELAY_START_DISABLE_DIR]:
+        cmd = f"test -d {_dir} && rm -rf {_dir} && systemctl daemon-reload || exit 0"
+        shell.get_stdout_or_raise_error(cmd, host=remote)
+
+
 def purge_sbd_from_cluster():
     '''
     Purge SBD from cluster, the process includes:
@@ -804,17 +816,10 @@ def purge_sbd_from_cluster():
         if service_manager.service_is_enabled(constants.SBD_SERVICE, node):
             logger.info("Disable %s on node %s", constants.SBD_SERVICE, node)
             service_manager.disable_service(constants.SBD_SERVICE, node)
-
-    config_bak = f"{SBDManager.SYSCONFIG_SBD}.bak"
-    logger.info("Move %s to %s on all nodes", SBDManager.SYSCONFIG_SBD, config_bak)
-    utils.cluster_run_cmd(f"mv {SBDManager.SYSCONFIG_SBD} {config_bak}")
+        cleanup_sbd_configurations(node)
 
     out = sh.cluster_shell().get_stdout_or_raise_error("stonith_admin -L")
     res = re.search("([0-9]+) fence device[s]* found", out)
     # after disable sbd.service, check if sbd is the last stonith device
     if res and int(res.group(1)) <= 1:
         utils.cleanup_stonith_related_properties()
-
-    for _dir in [SBDManager.SBD_SYSTEMD_DELAY_START_DIR, SBDManager.SBD_SYSTEMD_DELAY_START_DISABLE_DIR]:
-        cmd = f"test -d {_dir} && rm -rf {_dir} || exit 0"
-        parallax.parallax_call(cluster_nodes, cmd)

test/unittests/test_bootstrap.py

@@ -1977,16 +1977,17 @@ def test_remove_self_other_nodes(self, mock_this_node, mock_list, mock_run, mock
         mock_run.assert_called_once_with("node2", "crm cluster remove -y -c node1")
         mock_error.assert_called_once_with("Failed to remove this node from node2: err")
 
-    @mock.patch('crmsh.utils.package_is_installed')
+    @mock.patch('crmsh.sbd.cleanup_sbd_configurations')
+    @mock.patch('os.path.exists')
     @mock.patch('crmsh.sh.ClusterShell.get_stdout_or_raise_error')
-    def test_rm_configuration_files(self, mock_run, mock_installed):
+    def test_rm_configuration_files(self, mock_run, mock_exists, mock_rm_sbd):
         bootstrap._context = mock.Mock(rm_list=["file1", "file2"])
-        mock_installed.return_value = True
+        mock_exists.return_value = True
         bootstrap.rm_configuration_files()
         mock_run.assert_has_calls([
             mock.call('rm -f file1 file2', None),
-            mock.call('cp /usr/share/fillup-templates/sysconfig.sbd /etc/sysconfig/sbd', None)
             ])
+        mock_rm_sbd.assert_called_once_with(None)
 
     @mock.patch('crmsh.utils.get_iplist_from_name')
     @mock.patch('crmsh.corosync.get_values')

test/unittests/test_sbd.py

@@ -804,15 +804,27 @@ def test_cleanup_existing_sbd_resource(self, mock_CrmMonXmlParser, mock_logger_i
             call("Remove sbd resource '%s'", 'sbd_resource')
         ])
 
-    @patch('crmsh.parallax.parallax_call')
+    @patch('logging.Logger.info')
+    @patch('crmsh.sh.cluster_shell')
+    def test_cleanup_sbd_configurations(self, mock_cluster_shell, mock_logger_info):
+        mock_cluster_shell_inst = Mock()
+        mock_cluster_shell.return_value = mock_cluster_shell_inst
+        mock_cluster_shell_inst.get_stdout_or_raise_error = Mock()
+        sbd.cleanup_sbd_configurations()
+        mock_cluster_shell_inst.get_stdout_or_raise_error.assert_has_calls([
+            call(f"test -f {sbd.SBDManager.SYSCONFIG_SBD} && mv {sbd.SBDManager.SYSCONFIG_SBD} {sbd.SBDManager.SYSCONFIG_SBD}.bak || exit 0", host=None),
+            call(f"test -d {sbd.SBDManager.SBD_SYSTEMD_DELAY_START_DIR} && rm -rf {sbd.SBDManager.SBD_SYSTEMD_DELAY_START_DIR} && systemctl daemon-reload || exit 0", host=None),
+            call(f"test -d {sbd.SBDManager.SBD_SYSTEMD_DELAY_START_DISABLE_DIR} && rm -rf {sbd.SBDManager.SBD_SYSTEMD_DELAY_START_DISABLE_DIR} && systemctl daemon-reload || exit 0", host=None),
+        ])
+
+    @patch('crmsh.sbd.cleanup_sbd_configurations')
     @patch('crmsh.utils.cleanup_stonith_related_properties')
     @patch('crmsh.sbd.sh.cluster_shell')
-    @patch('crmsh.utils.cluster_run_cmd')
     @patch('logging.Logger.info')
     @patch('crmsh.sbd.ServiceManager')
     @patch('crmsh.utils.list_cluster_nodes')
     @patch('crmsh.sbd.cleanup_existing_sbd_resource')
-    def test_purge_sbd_from_cluster(self, mock_cleanup_existing_sbd_resource, mock_list_cluster_nodes, mock_ServiceManager, mock_logger_info, mock_cluster_run_cmd, mock_cluster_shell, mock_cleanup_stonith_related_properties, mock_parallax_call):
+    def test_purge_sbd_from_cluster(self, mock_cleanup_existing_sbd_resource, mock_list_cluster_nodes, mock_ServiceManager, mock_logger_info, mock_cluster_shell, mock_cleanup_stonith_related_properties, mock_rm_sbd_configuration_files):
         mock_list_cluster_nodes.return_value = ['node1', 'node2']
         mock_ServiceManager.return_value.service_is_enabled.side_effect = [True, True]
         stonith_data = """stonith-sbd
@@ -823,6 +835,6 @@ def test_purge_sbd_from_cluster(self, mock_cleanup_existing_sbd_resource, mock_l
         mock_logger_info.assert_has_calls([
             call("Disable %s on node %s", constants.SBD_SERVICE, 'node1'),
             call("Disable %s on node %s", constants.SBD_SERVICE, 'node2'),
-            call("Move %s to %s on all nodes", sbd.SBDManager.SYSCONFIG_SBD, sbd.SBDManager.SYSCONFIG_SBD+'.bak')
         ])
         mock_cleanup_stonith_related_properties.assert_called_once()
+        mock_rm_sbd_configuration_files.assert_has_calls([call("node1"), call("node2")])