healthd

#!/usr/bin/env python2

"""
    healthd
    System health monitor daemon for SONiC
"""

import signal
import threading

from sonic_py_common.daemon_base import DaemonBase
from swsssdk import SonicV2Connector

from health_checker.manager import HealthCheckerManager

SYSLOG_IDENTIFIER = 'healthd'


class HealthDaemon(DaemonBase):
    """
    A daemon that run as a service to perform system health checker with a configurable interval. Also set system LED 
    according to the check result and store the check result to redis.
    """
    SYSTEM_HEALTH_TABLE_NAME = 'SYSTEM_HEALTH_INFO'

    def __init__(self):
        """
        Constructor of HealthDaemon.
        """
        DaemonBase.__init__(self, SYSLOG_IDENTIFIER)
        self._db = SonicV2Connector(host="127.0.0.1")
        self._db.connect(self._db.STATE_DB)
        self.stop_event = threading.Event()

    def deinit(self):
        """
        Destructor. Remove all entries in $SYSTEM_HEALTH_TABLE_NAME table.
        :return: 
        """
        self._clear_system_health_table()

    def _clear_system_health_table(self):
        self._db.delete_all_by_pattern(self._db.STATE_DB, HealthDaemon.SYSTEM_HEALTH_TABLE_NAME)

    # Signal handler
    def signal_handler(self, sig, frame):
        """
        Signal handler
        :param sig: Signal number
        :param frame: not used
        :return:
        """
        if sig == signal.SIGHUP:
            self.log_notice("Caught SIGHUP - ignoring...")
        elif sig == signal.SIGINT:
            self.log_notice("Caught SIGINT - exiting...")
            self.stop_event.set()
        elif sig == signal.SIGTERM:
            self.log_notice("Caught SIGTERM - exiting...")
            self.stop_event.set()
        else:
            self.log_warning("Caught unhandled signal '" + sig + "'")

    def run(self):
        """
        Check system health in an infinite loop.
        :return: 
        """
        self.log_notice("Starting up...")

        import sonic_platform.platform
        chassis = sonic_platform.platform.Platform().get_chassis()
        manager = HealthCheckerManager()
        if not manager.config.config_file_exists():
            self.log_warning("System health configuration file not found, exit...")
            return
        while 1:
            state, stat = manager.check(chassis)
            if state == HealthCheckerManager.STATE_RUNNING:
                self._process_stat(chassis, manager.config, stat)

            if self.stop_event.wait(manager.config.interval):
                break

        self.deinit()

    def _process_stat(self, chassis, config, stat):
        from health_checker.health_checker import HealthChecker
        self._clear_system_health_table()
        for category, info in stat.items():
            for obj_name, obj_data in info.items():
                if obj_data[HealthChecker.INFO_FIELD_OBJECT_STATUS] == HealthChecker.STATUS_NOT_OK:
                    self._db.set(self._db.STATE_DB, HealthDaemon.SYSTEM_HEALTH_TABLE_NAME, obj_name,
                                 obj_data[HealthChecker.INFO_FIELD_OBJECT_MSG])

        self._db.set(self._db.STATE_DB, HealthDaemon.SYSTEM_HEALTH_TABLE_NAME, 'summary', HealthChecker.summary)


#
# Main =========================================================================
#
def main():
    health_monitor = HealthDaemon()
    health_monitor.run()


if __name__ == '__main__':
    main()