Introduce chassisd to monitor status of cards on chassis

sonic-chassisd/scripts/chassisd

@@ -0,0 +1,224 @@
+#!/usr/bin/env python2
+
+"""
+    chassisd
+    Module information update daemon for SONiC
+    This daemon will loop to collect all modules related information and then write the information to state DB.
+    The loop interval is CHASSIS_INFO_UPDATE_PERIOD_SECS in seconds.
+"""
+
+try:
+    import sys
+    import time
+    import signal
+    import threading
+    import os
+    from swsscommon import swsscommon
+    from sonic_py_common import daemon_base, logger
+    from sonic_platform_base.module_base import ModuleBase
+    from sonic_py_common.task_base import ProcessTaskBase
+except ImportError, e:
+    raise ImportError (str(e) + " - required module not found")
+
+#
+# Constants ====================================================================
+#
+
+SYSLOG_IDENTIFIER = "chassisd"
+
+CHASSIS_CFG_TABLE = 'CHASSIS_MODULE'
+
+CHASSIS_INFO_TABLE = 'CHASSIS_TABLE'
+CHASSIS_INFO_KEY_TEMPLATE = 'CHASSIS {}'
+CHASSIS_INFO_CARD_NUM_FIELD = 'module_num'
+
+CHASSIS_MODULE_INFO_TABLE = 'CHASSIS_MODULE_TABLE'
+CHASSIS_MODULE_INFO_KEY_TEMPLATE = 'CHASSIS_MODULE {}'
+CHASSIS_MODULE_INFO_NAME_FIELD = 'name'
+CHASSIS_MODULE_INFO_DESC_FIELD = 'desc'
+CHASSIS_MODULE_INFO_SLOT_FIELD = 'slot'
+CHASSIS_MODULE_INFO_OPERSTATUS_FIELD = 'oper_status'
+
+CHASSIS_INFO_UPDATE_PERIOD_SECS = 10
+
+CHASSIS_LOAD_ERROR = 1
+CHASSIS_NOT_AVAIL = 2
+
+platform_chassis = None
+
+SELECT_TIMEOUT = 1000
+#
+# Helper functions =============================================================
+#
+info_dict_keys = [CHASSIS_MODULE_INFO_NAME_FIELD,
+                  CHASSIS_MODULE_INFO_DESC_FIELD,
+                  CHASSIS_MODULE_INFO_SLOT_FIELD,
+                  CHASSIS_MODULE_INFO_OPERSTATUS_FIELD]
+
+def get_module_info(module_index):
+    """
+    Retrieves module info of this module
+    """
+    module_info_dict = {}
+    module_info_dict = dict.fromkeys(info_dict_keys, 'N/A')
+    module_info_dict[CHASSIS_MODULE_INFO_NAME_FIELD] = str(platform_chassis.get_module(module_index - 1).get_name())
+    module_info_dict[CHASSIS_MODULE_INFO_DESC_FIELD] = str(platform_chassis.get_module(module_index - 1).get_description())
+    module_info_dict[CHASSIS_MODULE_INFO_SLOT_FIELD] = str(platform_chassis.get_module(module_index - 1).get_slot())
+    module_info_dict[CHASSIS_MODULE_INFO_OPERSTATUS_FIELD] = str(platform_chassis.get_module(module_index - 1).get_status())
+    return module_info_dict
+
+def module_db_update(config_tbl, module_tbl, num_modules):
+    for module_index in range(1, num_modules + 1):
+        module_info_dict = get_module_info(module_index)
+        if module_info_dict is not None:
+            fvs = swsscommon.FieldValuePairs([(CHASSIS_MODULE_INFO_DESC_FIELD, module_info_dict[CHASSIS_MODULE_INFO_DESC_FIELD]),
+                                              (CHASSIS_MODULE_INFO_SLOT_FIELD, module_info_dict[CHASSIS_MODULE_INFO_SLOT_FIELD]),
+                                              (CHASSIS_MODULE_INFO_OPERSTATUS_FIELD, module_info_dict[CHASSIS_MODULE_INFO_OPERSTATUS_FIELD])])
+            module_tbl.set(module_info_dict[CHASSIS_MODULE_INFO_NAME_FIELD], fvs)
+
+#
+# Config Handling task ========================================================
+#
+class config_handling_task(ProcessTaskBase):
+    def __init__(self):
+        ProcessTaskBase.__init__(self)
+
+        # TODO: Refactor to eliminate the need for this Logger instance
+        self.logger = logger.Logger(SYSLOG_IDENTIFIER)
+
+    def task_worker(self):
+        config_db = daemon_base.db_connect("CONFIG_DB")
+
+        #Subscribe to CHASSIS_MODULE table notifications in the Config DB
+        sel = swsscommon.Select()
+        sst = swsscommon.SubscriberStateTable(config_db, CHASSIS_CFG_TABLE)
+        sel.addSelectable(sst)
+
+        # Listen indefinitely for changes to the CFG_CHASSIS_MODULE_TABLE table in the Config DB
+        while True:
+            # Use timeout to prevent ignoring the signals we want to handle
+            # in signal_handler() (e.g. SIGTERM for graceful shutdown)
+            (state, c) = sel.select(SELECT_TIMEOUT)
+
+            if state == swsscommon.Select.TIMEOUT:
+                # Do not flood log when select times out
+                continue
+            if state != swsscommon.Select.OBJECT:
+                self.logger.log_warning("sel.select() did not  return swsscommon.Select.OBJECT")
+                continue
+
+            (key, op, fvp) = sst.pop()
+
+            if not key.startswith(ModuleBase.MODULE_TYPE_CONTROL) and not key.startswith(ModuleBase.MODULE_TYPE_LINE) and not key.startswith(ModuleBase.MODULE_TYPE_FABRIC):
+                self.logger.log_warning("Incorrect module-name {}. Should start with {} or {} or {}".format(key, ModuleBase.MODULE_TYPE_CONTROL, ModuleBase.MODULE_TYPE_LINE, ModuleBase.MODULE_TYPE_FABRIC))
+                continue
+
+            fvp_dict = dict(fvp)
+
+            module_index = platform_chassis.get_module_index(key)
+
+            #Continue if the index is invalid
+            if module_index < 0:
+                self.logger.log_error("Unable to get module-index for key {} during op {}". format(key, op))
+                continue
+
+            if op == "SET":
+                #Setting the module to administratively down state
+                self.logger.log_info("Changing module {} to admin DOWN state".format(key))
+                platform_chassis.get_module(module_index).set_admin_state(0)
+            elif op == "DEL":
+                #Setting the module to administratively up state
+                self.logger.log_info("Changing module {} to admin UP state".format(key))
+                platform_chassis.get_module(module_index).set_admin_state(1)
+
+        return 1
+
+#
+# Daemon =======================================================================
+#
+
+class ChassisdDaemon(daemon_base.DaemonBase):
+    def __init__(self, log_identifier):
+        super(ChassisdDaemon, self).__init__(log_identifier)
+
+        self.logger = logger.Logger(SYSLOG_IDENTIFIER)
+        self.stop = threading.Event()
+
+    # Signal handler
+    def signal_handler(self, sig, frame):
+        if sig == signal.SIGHUP:
+            self.log_info("Caught SIGHUP - ignoring...")
+        elif sig == signal.SIGINT:
+            self.log_info("Caught SIGINT - exiting...")
+            self.stop.set()
+        elif sig == signal.SIGTERM:
+            self.log_info("Caught SIGTERM - exiting...")
+            self.stop.set()
+        else:
+            self.log_warning("Caught unhandled signal '" + sig + "'")
+
+    # Run daemon
+    def run(self):
+        global platform_chassis
+
+        self.log_info("Starting up...")
+
+        # Load new platform api class
+        try:
+            import sonic_platform.platform
+            platform_chassis = sonic_platform.platform.Platform().get_chassis()
+        except Exception as e:
+            self.log_error("Failed to load chassis due to {}".format(repr(e)))
+            sys.exit(CHASSIS_LOAD_ERROR)
+
+        # Connect to STATE_DB and create chassis module/chassis info tables
+        state_db = daemon_base.db_connect("STATE_DB")
+        chassis_tbl = swsscommon.Table(state_db, CHASSIS_INFO_TABLE)
+        module_tbl = swsscommon.Table(state_db, CHASSIS_MODULE_INFO_TABLE)
+
+        # Connect to CONFIG_DB and check with chassis cfg table
+        config_db = daemon_base.db_connect("CONFIG_DB")
+        config_tbl = swsscommon.Table(config_db, CHASSIS_CFG_TABLE)
+
+        # Post number-of-modules info to STATE_DB
+        num_modules = platform_chassis.get_num_modules()
+        fvs = swsscommon.FieldValuePairs([(CHASSIS_INFO_CARD_NUM_FIELD, str(num_modules))])
+        chassis_tbl.set(CHASSIS_INFO_KEY_TEMPLATE.format(1), fvs)
+
+        if num_modules == 0:
+            self.log_error("Chassisd has no modules available")
+
+        #Start configuration handling task on control module
+        if platform_chassis.get_controlcard_slot() == platform_chassis.get_my_slot():
+            config_update = config_handling_task()
+            config_update.task_run()
+
+        # Start main loop
+        self.log_info("Start daemon main loop")
+
+        while not self.stop.wait(CHASSIS_INFO_UPDATE_PERIOD_SECS):
+            module_db_update(config_tbl, module_tbl, num_modules)
+
+        self.log_info("Stop daemon main loop")
+
+        if config_update is not None:
+            config_update.task_stop()
+
+        # Delete all the information from DB and then exit
+        for module_index in range(1, num_modules + 1):
+            module_tbl._del(CHASSIS_MODULE_INFO_KEY_TEMPLATE.format(module_index))
+
+        chassis_tbl._del(CHASSIS_INFO_KEY_TEMPLATE.format(1))
+
+        self.log_info("Shutting down...")
+
+#
+# Main =========================================================================
+#
+
+def main():
+    chassisd = ChassisdDaemon(SYSLOG_IDENTIFIER)
+    chassisd.run()
+
+if __name__ == '__main__':
+    main()

sonic-chassisd/setup.py

@@ -0,0 +1,29 @@
+from setuptools import setup
+
+setup(
+    name='sonic-chassisd',
+    version='1.0',
+    description='Chassis daemon for SONiC',
+    license='Apache 2.0',
+    author='SONiC Team',
+    author_email='linuxnetdev@microsoft.com',
+    url='https://github.com/Azure/sonic-platform-daemons',
+    maintainer='Manju Prabhu',
+    maintainer_email='manjunath.prabhu@nokia.com',
+    scripts=[
+        'scripts/chassisd',
+    ],
+    classifiers=[
+        'Development Status :: 4 - Beta',
+        'Environment :: No Input/Output (Daemon)',
+        'Intended Audience :: Developers',
+        'Intended Audience :: Information Technology',
+        'Intended Audience :: System Administrators',
+        'License :: OSI Approved :: Apache Software License',
+        'Natural Language :: English',
+        'Operating System :: POSIX :: Linux',
+        'Programming Language :: Python :: 2.7',
+        'Topic :: System :: Hardware',
+    ],
+    keywords='sonic SONiC chassis Chassis daemon chassisd',
+)