[crmsh-4.4] Dev: utils: Change the way to get pacemaker's version

crmsh/cibconfig.py

@@ -28,7 +28,7 @@
 from .ra import get_ra, get_properties_list, get_pe_meta, get_properties_meta, RAInfo
 from .utils import ext_cmd, safe_open_w, pipe_string, safe_close_w, crm_msec
 from .utils import ask, lines2cli, olist
-from .utils import page_string, cibadmin_can_patch, str2tmp, ensure_sudo_readable
+from .utils import page_string, str2tmp, ensure_sudo_readable
 from .utils import run_ptest, is_id_valid, edit_file, get_boolean, filter_string
 from .xmlutil import is_child_rsc, rsc_constraint, sanitize_cib, rename_id, get_interesting_nodes
 from .xmlutil import is_pref_location, get_topnode, new_cib, get_rscop_defaults_meta_node
@@ -2619,7 +2619,7 @@ def commit(self, force=False, replace=False):
         'Commit the configuration to the CIB.'
         if not self.is_cib_sane():
             return False
-        if not replace and cibadmin_can_patch():
+        if not replace:
             rc = self._patch_cib(force)
         else:
             rc = self._replace_cib(force)
@@ -2783,9 +2783,8 @@ def initialize(self, cib=None):
             cib = text2elem(cib)
         if not self._import_cib(cib):
             return False
-        if cibadmin_can_patch():
-            self.cib_orig = copy.deepcopy(self.cib_elem)
-            sanitize_cib_for_patching(self.cib_orig)
+        self.cib_orig = copy.deepcopy(self.cib_elem)
+        sanitize_cib_for_patching(self.cib_orig)
         sanitize_cib(self.cib_elem)
         show_unrecognized_elems(self.cib_elem)
         self._populate()

crmsh/constants.py

@@ -521,7 +521,6 @@
 REJOIN_COUNT = 60
 REJOIN_INTERVAL = 10
 DC_DEADTIME_DEFAULT = 20
-
 ADVISED_ACTION_LIST = ['monitor', 'start', 'stop', 'promote', 'demote']
 ADVISED_KEY_LIST = ['timeout', 'interval', 'role']
 DEFAULT_INTERVAL_IN_ACTION = "20s"

crmsh/history.py

@@ -553,7 +553,6 @@ def set_node_colors(self):
             i = (i+1) % len(self.nodecolors)
 
     def _report_setup_source(self):
-        constants.pcmk_version = None
         # is this an crm report or a crm_report?
         for descname in ("description.txt", "report.summary"):
             self.desc = os.path.join(self.loc, descname)

crmsh/log_patterns.py

@@ -142,7 +142,7 @@
 
 
 def patterns(cib_f=None):
-    is118 = utils.is_pcmk_118(cib_f=cib_f)
+    is118 = utils.is_larger_than_pcmk_118(cib_f=cib_f)
     if is118:
         return _patterns_118
     else:

crmsh/main.py

@@ -158,12 +158,6 @@ def set_interactive():
         options.interactive = True
 
 
-def compatibility_setup():
-    if not utils.is_pcmk_118():
-        del constants.attr_defaults["node"]
-        constants.cib_no_section_rc = 22
-
-
 def add_quotes(args):
     '''
     Add quotes if there's whitespace in one of the
@@ -250,7 +244,6 @@ def main_input_loop(context, user_args):
     Main input loop for crmsh. Parses input
     line by line.
     """
-    compatibility_setup()
     rc = handle_noninteractive_use(context, user_args)
     if rc is not None:
         return rc

crmsh/ui_cluster.py

@@ -186,10 +186,18 @@ def do_stop(self, context, *args):
                 else:
                     logger.info("Cluster services already stopped on {}".format(node))
                 node_list.remove(node)
+            elif not utils.service_is_active("pacemaker.service", remote_addr=node):
+                utils.stop_service("corosync", remote_addr=node)
+                logger.info("Cluster services stopped on {}".format(node))
+                node_list.remove(node)
         if not node_list:
             return
 
-        if not utils.get_dc(timeout=5):
+        dc_deadtime = utils.get_property("dc-deadtime") or constants.DC_DEADTIME_DEFAULT
+        dc_timeout = int(dc_deadtime.strip('s')) + 5
+        try:
+            utils.check_function_with_timeout(utils.get_dc, wait_timeout=dc_timeout)
+        except TimeoutError:
             logger.error("No DC found currently, please wait if the cluster is still starting")
             return False
 

crmsh/ui_configure.py

@@ -898,7 +898,6 @@ def _commit(self, force=False, replace=False):
             logger.info("apparently there is nothing to commit")
             logger.info("try changing something first")
             return True
-        replace = replace or not utils.cibadmin_can_patch()
         rc1 = True
         if replace and not force:
             rc1 = cib_factory.is_current_cib_equal()

crmsh/ui_node.py

@@ -417,7 +417,7 @@ def do_clearstate(self, context, node=None):
         if not config.core.force and \
                 not utils.ask("Do you really want to drop state for node %s?" % node):
             return False
-        if utils.is_pcmk_118():
+        if utils.is_larger_than_pcmk_118():
             cib_elem = xmlutil.cibdump2elem()
             if cib_elem is None:
                 return False

crmsh/utils.py

@@ -853,10 +853,8 @@ def append_file(dest, src):
         return False
 
 
-def get_dc(timeout=None):
-    cmd = "crmadmin -D"
-    if timeout:
-        cmd += " -t {}".format(timeout)
+def get_dc():
+    cmd = "crmadmin -D -t 1"
     rc, s, _ = get_stdout_stderr(add_sudo(cmd))
     if rc != 0:
         return None
@@ -1468,32 +1466,15 @@ def load_graphviz_file(ini_f):
     return True, _graph_d
 
 
-def get_pcmk_version(dflt):
-    version = dflt
-
-    crmd = pacemaker_controld()
-    if crmd:
-        cmd = crmd
-    else:
-        return version
-
-    try:
-        rc, s, err = get_stdout_stderr("%s version" % (cmd))
-        if rc != 0:
-            logger.error("%s exited with %d [err: %s][out: %s]", cmd, rc, err, s)
-        else:
-            if err.startswith("CRM Version:"):
-                version = s.split()[0]
-            else:
-                version = s.split()[2]
-            logger.debug("found pacemaker version: %s", version)
-    except Exception as msg:
-        logger.warning("could not get the pacemaker version, bad installation?")
-        logger.warning(msg)
+def get_pcmk_version():
+    cmd = "/usr/sbin/pacemakerd --version"
+    out = get_stdout_or_raise_error(cmd)
+    version = out.split()[1]
+    logger.debug("Found pacemaker version: %s", version)
     return version
 
 
-def get_cib_property(cib_f, attr, dflt):
+def get_cib_property(cib_f, attr, dflt=None):
     """A poor man's get attribute procedure.
     We don't want heavy parsing, this needs to be relatively
     fast.
@@ -1561,14 +1542,17 @@ def is_larger_than_min_version(version, min_version):
 def is_min_pcmk_ver(min_ver, cib_f=None):
     if not constants.pcmk_version:
         if cib_f:
-            constants.pcmk_version = get_cib_property(cib_f, "dc-version", "1.1.11")
-            logger.debug("found pacemaker version: %s in cib: %s", constants.pcmk_version, cib_f)
+            constants.pcmk_version = get_cib_property(cib_f, "dc-version")
+            if constants.pcmk_version:
+                logger.debug("Found pacemaker version: %s in cib: %s", constants.pcmk_version, cib_f)
+            else:
+                fatal(f"Failed to get 'dc-version' from {cib_f}")
         else:
-            constants.pcmk_version = get_pcmk_version("1.1.11")
+            constants.pcmk_version = get_pcmk_version()
     return is_larger_than_min_version(constants.pcmk_version, min_ver)
 
 
-def is_pcmk_118(cib_f=None):
+def is_larger_than_pcmk_118(cib_f=None):
     return is_min_pcmk_ver("1.1.8", cib_f=cib_f)
 
 
@@ -1586,12 +1570,6 @@ def cibadmin_features():
     return []
 
 
-@memoize
-def cibadmin_can_patch():
-    # cibadmin -P doesn't handle comments in <1.1.11 (unless patched)
-    return is_min_pcmk_ver("1.1.11")
-
-
 # quote function from python module shlex.py in python 3.3
 
 _find_unsafe = re.compile(r'[^\w@%+=:,./-]').search
@@ -3005,7 +2983,8 @@ def get_property(name):
     """
     Get cluster properties
     """
-    cmd = "crm configure get_property " + name
+    cib_path = os.getenv('CIB_file', constants.CIB_RAW_FILE)
+    cmd = "CIB_file={} crm configure get_property {}".format(cib_path, name)
     rc, stdout, _ = get_stdout_stderr(cmd)
     return stdout if rc == 0 else None
 
@@ -3131,4 +3110,20 @@ def read_from_file(infile):
     with open(infile, 'rt', encoding='utf-8', errors='replace') as f:
         data = f.read()
     return to_ascii(data)
+
+
+def check_function_with_timeout(check_function, wait_timeout=30, interval=1):
+    """
+    Run check_function in a loop
+    Return when check_function is true
+    Raise TimeoutError when timeout
+    """
+    current_time = int(time.time())
+    timeout = current_time + wait_timeout
+    while current_time <= timeout:
+        if check_function():
+            return
+        time.sleep(interval)
+        current_time = int(time.time())
+    raise TimeoutError
 # vim:ts=4:sw=4:et:

crmsh/xmlutil.py

@@ -121,7 +121,7 @@ def cibdump2elem(section=None):
     rc, outp, errp = sudocall(cmd)
     if rc == 0:
         return text2elem(outp)
-    elif rc != constants.cib_no_section_rc:
+    else:
         logger.error("running %s: %s", cmd, errp)
     return None
 

test/features/bootstrap_bugs.feature

@@ -112,3 +112,22 @@ Feature: Regression test for bootstrap bugs
     When    Run "crm cluster remove HANODE2 -y" on "hanode1"
     Then    Cluster service is "stopped" on "hanode2"
     And     Online nodes are "hanode1"
+
+  @clean
+  Scenario: Stop service quickly(bsc#1203601)
+    Given   Cluster service is "stopped" on "hanode1"
+    And     Cluster service is "stopped" on "hanode2"
+    When    Run "crm cluster init -y" on "hanode1"
+    Then    Cluster service is "started" on "hanode1"
+    When    Run "crm cluster join -c hanode1 -y" on "hanode2"
+    Then    Cluster service is "started" on "hanode2"
+    When    Run "crm cluster stop --all" on "hanode1"
+    Then    Cluster service is "stopped" on "hanode1"
+    And     Cluster service is "stopped" on "hanode2"
+    When    Run "crm cluster start --all;crm cluster stop --all" on "hanode1"
+    Then    Cluster service is "stopped" on "hanode1"
+    And     Cluster service is "stopped" on "hanode2"
+    When    Run "systemctl start corosync" on "hanode1"
+    Then    Service "corosync" is "started" on "hanode1"
+    When    Run "crm cluster stop" on "hanode1"
+    Then    Service "corosync" is "stopped" on "hanode1"

test/unittests/test_ui_cluster.py

@@ -100,20 +100,23 @@ def test_do_stop_already_stopped(self, mock_parse_nodes, mock_active, mock_info)
     @mock.patch('crmsh.utils.is_quorate')
     @mock.patch('crmsh.utils.is_dlm_running')
     @mock.patch('crmsh.utils.get_dc')
+    @mock.patch('crmsh.utils.check_function_with_timeout')
+    @mock.patch('crmsh.utils.get_property')
     @mock.patch('crmsh.utils.service_is_active')
     @mock.patch('crmsh.ui_cluster.parse_option_for_nodes')
-    def test_do_stop(self, mock_parse_nodes, mock_active, mock_get_dc, mock_dlm_running, mock_is_quorate, mock_set_dlm, mock_stop, mock_info, mock_debug):
+    def test_do_stop(self, mock_parse_nodes, mock_active, mock_get_property, mock_check, mock_get_dc, mock_dlm_running, mock_is_quorate, mock_set_dlm, mock_stop, mock_info, mock_debug):
         context_inst = mock.Mock()
         mock_parse_nodes.return_value = ["node1"]
-        mock_active.side_effect = [True, True]
+        mock_active.side_effect = [True, True, True]
         mock_dlm_running.return_value = True
         mock_is_quorate.return_value = False
-        mock_get_dc.return_value = "node1"
+        mock_get_property.return_value = "20s"
 
         self.ui_cluster_inst.do_stop(context_inst, "node1")
 
         mock_active.assert_has_calls([
             mock.call("corosync.service", remote_addr="node1"),
+            mock.call("pacemaker.service", remote_addr="node1"),
             mock.call("corosync-qdevice.service")
             ])
         mock_stop.assert_has_calls([
@@ -123,3 +126,4 @@ def test_do_stop(self, mock_parse_nodes, mock_active, mock_get_dc, mock_dlm_runn
             ])
         mock_info.assert_called_once_with("Cluster services stopped on node1")
         mock_debug.assert_called_once_with("Quorum is lost; Set enable_quorum_fencing=0 and enable_quorum_lockspace=0 for dlm")
+        mock_check.assert_called_once_with(mock_get_dc, wait_timeout=25)

test/unittests/test_utils.py

@@ -1616,11 +1616,13 @@ def test_list_cluster_nodes(mock_run, mock_env, mock_isfile, mock_file2elem):
         ])
 
 
+@mock.patch('os.getenv')
 @mock.patch('crmsh.utils.get_stdout_stderr')
-def test_get_property(mock_run):
+def test_get_property(mock_run, mock_env):
     mock_run.return_value = (0, "data", None)
+    mock_env.return_value = "cib.xml"
     assert utils.get_property("no-quorum-policy") == "data"
-    mock_run.assert_called_once_with("crm configure get_property no-quorum-policy")
+    mock_run.assert_called_once_with("CIB_file=cib.xml crm configure get_property no-quorum-policy")
 
 
 @mock.patch('crmsh.utils.get_stdout_or_raise_error')