Fix: sbd: bootstrap parameters better cope with Azure for sbd and cor…

…osync(bsc#1175896)

crmsh/bootstrap.py

@@ -34,7 +34,7 @@
 from . import term
 from . import lock
 from . import userdir
-from .constants import SSH_OPTION, QDEVICE_HELP_INFO
+from .constants import SSH_OPTION, QDEVICE_HELP_INFO, PROFILES_DICT
 from . import ocfs2
 from . import qdevice
 
@@ -118,6 +118,7 @@ def __init__(self):
         self.interfaces_inst = None
         self.with_other_user = True
         self.cluster_is_running = None
+        self.profiles_dict = {}
         self.default_nic_list = []
         self.default_ip_list = []
         self.local_ip_list = []
@@ -186,6 +187,17 @@ def init_sbd_manager(self):
         from .sbd import SBDManager
         self.sbd_manager = SBDManager(self)
 
+    def load_profiles(self):
+        """
+        Load profiles data for different cloud provider
+        """
+        env_name = None
+        if utils.is_in_azure():
+            env_name = "Azure"
+        if not env_name:
+            return
+        self.profiles_dict = PROFILES_DICT[env_name]
+
 
 _context = None
 
@@ -1164,6 +1176,18 @@ def gen_mcastaddr():
     csync2_update(corosync.conf())
 
 
+def profiles_adjust_for_corosync():
+    """
+    Adjust corosync's parameters according profiles
+    """
+    if not _context.profiles_dict:
+        return
+    corosync.set_value("totem.token", _context.profiles_dict["corosync_token"])
+    corosync.set_value("totem.consensus", _context.profiles_dict["corosync_consensus"])
+    corosync.set_value("totem.max_messages", _context.profiles_dict["corosync_max_messages"])
+    corosync.set_value("totem.token_retransmits_before_loss_const", _context.profiles_dict["corosync_token_retransmits_before_loss_const"])
+
+
 def init_corosync():
     """
     Configure corosync (unicast or multicast, encrypted?)
@@ -1184,6 +1208,7 @@ def requires_unicast():
         init_corosync_unicast()
     else:
         init_corosync_multicast()
+    profiles_adjust_for_corosync()
 
 
 def init_sbd():
@@ -2016,6 +2041,7 @@ def bootstrap_init(context):
 
     _context.initialize_qdevice()
     _context.validate_option()
+    _context.load_profiles()
     _context.init_sbd_manager()
 
     # Need hostname resolution to work, want NTP (but don't block ssh_remote or csync2_remote)

crmsh/constants.py

@@ -490,4 +490,16 @@
 
 
 SSH_OPTION = "-o StrictHostKeyChecking=no"
+
+
+PROFILES_DICT = {
+        "Azure": {
+            "corosync_token": 30000,
+            "corosync_consensus": 36000,
+            "corosync_max_messages": 20,
+            "corosync_token_retransmits_before_loss_const": 10,
+            "sbd_watchdog_timeout": 60,
+            "sbd_msgwait": 120
+            }
+        }
 # vim:ts=4:sw=4:et:

crmsh/sbd.py

@@ -47,6 +47,7 @@ def __init__(self, context):
         self._sbd_watchdog_timeout = 0
         self._is_s390 = "390" in os.uname().machine
         self._context = context
+        self._sbd_msgwait = None
 
     @staticmethod
     def _get_device_uuid(dev, node=None):
@@ -150,8 +151,12 @@ def _initialize_sbd(self):
         """
         if self.diskless_sbd:
             return
+        opt = ""
+        if self._context.profiles_dict:
+            self._sbd_msgwait = self._context.profiles_dict["sbd_msgwait"]
+            opt = "-4 {}".format(self._sbd_msgwait)
         for dev in self._sbd_devices:
-            rc, _, err = bootstrap.invoke("sbd -d {} create".format(dev))
+            rc, _, err = bootstrap.invoke("sbd {} -d {} create".format(opt, dev))
             if not rc:
                 bootstrap.error("Failed to initialize SBD device {}: {}".format(dev, err))
 
@@ -179,6 +184,8 @@ def _determine_sbd_watchdog_timeout(self):
         When using diskless SBD, determine value of SBD_WATCHDOG_TIMEOUT
         """
         if not self.diskless_sbd:
+            if self._context.profiles_dict:
+                self._sbd_watchdog_timeout = self._context.profiles_dict["sbd_watchdog_timeout"]
             return
         # add sbd after qdevice started
         if utils.is_qdevice_configured() and utils.service_is_active("corosync-qdevice.service"):
@@ -241,6 +248,28 @@ def _enable_sbd_service(self):
             # in init process
             bootstrap.invoke("systemctl enable sbd.service")
 
+    def _adjust_systemd(self):
+        """
+        Adjust start timeout for sbd when has profiles data
+        """
+        if not self._context.profiles_dict:
+            return
+
+        # TimeoutStartUSec default is 1min 30s, need to parse as seconds
+        cmd = "systemctl show service -p TimeoutStartUSec|\
+                sed -e 's/min/ 60/; s/s//g; s/.*=//g'|\
+                awk '{if (NF == 3) a=$1*$2+$3; else if (NF == 1) a=$1;print a}'"
+        default_start_timeout = int(utils.get_stdout_or_raise_error(cmd))
+        if self._sbd_msgwait <= default_start_timeout:
+            return
+
+        systemd_sbd_dir = "/etc/systemd/system/sbd.service.d"
+        utils.mkdirp(systemd_sbd_dir)
+        systemd_timeout_sec = self._sbd_msgwait * 1.2
+        cmd = 'echo -e "[Service]\nTimeoutSec={}" | tee {}/sbd_delay_start.conf'.format(systemd_timeout_sec, systemd_sbd_dir)
+        utils.get_stdout_or_raise_error(cmd)
+        utils.get_stdout_or_raise_error("systemctl daemon-reload")
+
     def _warn_diskless_sbd(self, peer=None):
         """
         Give warning when configuring diskless sbd
@@ -278,6 +307,7 @@ def sbd_init(self):
             self._update_configuration()
         self._determine_stonith_watchdog_timeout()
         self._enable_sbd_service()
+        self._adjust_systemd()
 
     def configure_sbd_resource(self):
         """

crmsh/utils.py

@@ -2067,6 +2067,13 @@ def detect_cloud():
     return None
 
 
+def is_in_azure():
+    """
+    Detect if is in Azure
+    """
+    return detect_cloud() == "microsoft-azure"
+
+
 def debug_timestamp():
     return datetime.datetime.now().strftime('%Y/%m/%d %H:%M:%S')