add check feature 'on_exit'

README.rst

@@ -256,6 +256,7 @@ Below are the default settings for all service checks, see `Configuring checks f
 :check_fail: 2
 :check_disabled: true
 :on_disabled: withdraw
+:on_exit: withdraw
 :ip_check_disabled: false
 :custom_bird_reconfigure_cmd_timeout: 2
 
@@ -447,6 +448,7 @@ Here are few examples::
     check_rise        = 2
     check_disabled    = false
     on_disabled       = withdraw
+    on_exit           = none
     ip_prefix         = 10.52.12.1/32
 
     [foo6.bar.com]
@@ -456,6 +458,7 @@ Here are few examples::
     check_fail        = 2
     check_disabled    = false
     on_disabled       = withdraw
+    on_exit           = none
     ip_prefix         = fd12:aba6:57db:ffff::1/128
     ip_check_disabled = false
 
@@ -489,6 +492,14 @@ A service is considered HEALTHY after these many consecutive successful health c
 
 What to do when check is disabled, either ``withdraw`` or ``advertise``
 
+* **on_exit** Defaults to **none**
+
+What to do when anycast-healthchecker exits, either ``none`` or ``withdraw`` or ``advertise``.
+
+  * ``none`` keep state how it was from checks
+  * ``advertise`` ensure the ip_prefix to be advertised
+  * ``withdraw`` ensure the ip_prefix to be withdrawn
+
 * **ip_prefix** Unset by default
 
 IP prefix associated with the service. It **must be** assigned to the interface set in ``interface`` parameter unless ``ip_check_disabled`` is set to ``true``. Prefix length is optional and defaults to 32 for IPv4 addresses and to 128 for IPv6 addresses.

anycast_healthchecker/__init__.py

@@ -18,6 +18,7 @@
         'check_fail': 2,
         'check_disabled': 'true',
         'on_disabled': 'withdraw',
+        'on_exit': 'none',
         'ip_check_disabled': 'false',
         'custom_bird_reconfigure_cmd_timeout': 2,
     },

anycast_healthchecker/healthchecker.py

@@ -12,9 +12,12 @@
 from anycast_healthchecker.utils import (SERVICE_OPTIONS_TYPE,
                                          get_ip_prefixes_from_bird,
                                          get_ip_prefixes_from_config,
+                                         get_ip_prefixes_from_config_on_exit,
                                          reconfigure_bird,
                                          write_temp_bird_conf,
                                          archive_bird_conf,
+                                         AddOperation,
+                                         DeleteOperation,
                                          ServiceCheckDiedError,
                                          run_custom_bird_reconfigure)
 
@@ -226,3 +229,47 @@ def run(self):
                         self.bird_configuration[ip_version]['reconfigure_cmd'])
                 else:
                     run_custom_bird_reconfigure(operation)
+
+    def on_exit(self):
+        for ip_version in self.bird_configuration:
+            config_file = self.bird_configuration[ip_version]['config_file']
+            ip_prefixes_in_bird = get_ip_prefixes_from_bird(config_file)
+            _ip_prefix_withdraw = get_ip_prefixes_from_config_on_exit(
+                self.config,
+                self.services,
+                ip_version,
+                'withdraw')
+            _ip_prefix_advertise = get_ip_prefixes_from_config_on_exit(
+                self.config,
+                self.services,
+                ip_version,
+                'advertise')
+
+            for withdraw_ip_prefix in set(ip_prefixes_in_bird).intersection(_ip_prefix_withdraw):
+                _delete_operation = DeleteOperation(
+                    name='on_exit',
+                    ip_prefix=withdraw_ip_prefix,
+                    ip_version=ip_version,
+                    bird_reconfigure_timeout=(
+                        self.config['daemon'].get('custom_bird_reconfigure_cmd_timeout')
+                    ),
+                    bird_reconfigure_cmd=(
+                        self.config['daemon'].get('custom_bird_reconfigure_cmd')
+                    )
+                )
+                self._update_bird_conf_file(_delete_operation)
+
+            for add_ip_prefix in set(_ip_prefix_advertise).difference(ip_prefixes_in_bird):
+                _add_operation = AddOperation(
+                    name='on_exit',
+                    ip_prefix=add_ip_prefix,
+                    ip_version=ip_version,
+                    bird_reconfigure_timeout=(
+                        self.config['daemon'].get('custom_bird_reconfigure_cmd_timeout')
+                    ),
+                    bird_reconfigure_cmd=(
+                        self.config['daemon'].get('custom_bird_reconfigure_cmd')
+                    )
+                )
+                self._update_bird_conf_file(_add_operation)
+

anycast_healthchecker/main.py

@@ -79,8 +79,10 @@ def main():
     pidfile = config.get('daemon', 'pidfile')
     update_pidfile(pidfile)
 
+    checker = healthchecker.HealthChecker(config, bird_configuration)
+
     # Register our shutdown handler to various termination signals.
-    shutdown_handler = partial(shutdown, pidfile)
+    shutdown_handler = partial(shutdown, pidfile, checker)
     signal.signal(signal.SIGHUP, shutdown_handler)
     signal.signal(signal.SIGTERM, shutdown_handler)
     signal.signal(signal.SIGABRT, shutdown_handler)
@@ -93,7 +95,6 @@ def main():
     ip_prefixes_sanity_check(config, bird_configuration)
 
     # Create our master process.
-    checker = healthchecker.HealthChecker(config, bird_configuration)
     logger.info("starting %s version %s", PROGRAM_NAME, __version__)
     checker.run()
 

anycast_healthchecker/utils.py

@@ -34,6 +34,7 @@
     'check_fail': 'getint',
     'check_disabled': 'getboolean',
     'on_disabled': 'get',
+    'on_exit': 'get',
     'ip_prefix': 'get',
     'interface': 'get',
     'ip_check_disabled': 'getboolean',
@@ -144,6 +145,27 @@ def get_ip_prefixes_from_config(config, services, ip_version):
 
     return ip_prefixes
 
+def get_ip_prefixes_from_config_on_exit(config, services, ip_version, on_exit_status):
+    """Build a set of IP prefixes found in service configuration files.
+
+    Arguments:
+        config (obg): A configparser object which holds our configuration.
+        services (list): A list of section names which are the name of the
+        service checks.
+        ip_version (int): IP protocol version
+
+    Returns:
+        A set of IP prefixes.
+
+    """
+    ip_prefixes = set()
+
+    for service in services:
+        if config.get(service, 'on_exit') == on_exit_status:
+            ip_prefix = ipaddress.ip_network(config.get(service, 'ip_prefix'))
+            if ip_prefix.version == ip_version:
+                ip_prefixes.add(ip_prefix.with_prefixlen)
+    return ip_prefixes
 
 def ip_prefixes_sanity_check(config, bird_configuration):
     """Sanity check on IP prefixes.
@@ -414,6 +436,17 @@ def service_configuration_check(config):
                            val=config.get(service, 'on_disabled')))
             raise ValueError(msg)
 
+
+        if (config.get(service, 'on_exit') != 'withdraw' and
+                config.get(service, 'on_exit') != 'advertise' and
+                config.get(service, 'on_exit') != 'none'):
+            msg = ("'on_exit' option has invalid value ({val}) for "
+                   "service check {name}, 'on_exit option should be set "
+                   "either to 'withdraw' or to 'advertise' or to 'none'"
+                   .format(name=service,
+                           val=config.get(service, 'on_exit')))
+            raise ValueError(msg)
+
         ip_prefixes.append(config.get(service, 'ip_prefix'))
 
         if not valid_ip_prefix(config.get(service, 'ip_prefix')):
@@ -923,7 +956,7 @@ def write_pid(pidfile):
         sys.exit("failed to write pidfile:{e}".format(e=exc))
 
 
-def shutdown(pidfile, signalnb=None, frame=None):
+def shutdown(pidfile, healthchecker, signalnb=None, frame=None):
     """Clean up pidfile upon shutdown.
 
     Notice:
@@ -942,6 +975,7 @@ def shutdown(pidfile, signalnb=None, frame=None):
     """
     log = logging.getLogger(PROGRAM_NAME)
     log.info("received %s at %s", signalnb, frame)
+    healthchecker.on_exit()
     log.info("going to remove pidfile %s", pidfile)
     # no point to catch possible errors when we delete the pid file
     os.unlink(pidfile)