AMBARI-1728. Cleanup INFO Logging at the ambari agent to make it more useful and less verbose. (swagle)

git-svn-id: https://svn.apache.org/repos/asf/incubator/ambari/trunk@1461886 13f79535-47bb-0310-9956-ffa450edef68

ambari-agent/conf/unix/ambari-agent.ini

@@ -19,6 +19,8 @@ secured_url_port=8441
 
 [agent]
 prefix=/var/lib/ambari-agent/data
+;loglevel=(DEBUG/INFO)
+loglevel=INFO
 
 [stack]
 installprefix=/var/ambari-agent/

ambari-agent/src/main/python/ambari_agent/ActionQueue.py

@@ -81,7 +81,10 @@ class ActionQueue(threading.Thread):
     return self._stop.isSet()
 
   def put(self, command):
-    logger.info("The " + command['commandType'] + " from the server is \n" + pprint.pformat(command))
+    logger.info("Adding " + command['commandType'] + " for service " +\
+                command['serviceName'] + " of cluster " +\
+                command['clusterName'] + " to the queue.")
+    logger.debug(pprint.pformat(command))
     self.commandQueue.put(command)
     pass
 
@@ -90,7 +93,7 @@ class ActionQueue(threading.Thread):
     while not self.stopped():
       while not self.commandQueue.empty():
         command = self.commandQueue.get()
-        logger.info("Took an element of Queue: " + pprint.pformat(command))
+        logger.debug("Took an element of Queue: " + pprint.pformat(command))
         if command['commandType'] == self.EXECUTION_COMMAND:
           try:
             #pass a copy of action since we don't want anything to change in the
@@ -118,8 +121,10 @@ class ActionQueue(threading.Thread):
             livestatus = LiveStatus(cluster, service, component,
               globalConfig, self.config)
             result = livestatus.build()
-            logger.info("Got live status for component " + component + " of service " + str(service) +\
-                        " of cluster " + str(cluster) + "\n" + pprint.pformat(result))
+            logger.debug("Got live status for component " + component +\
+                         " of service " + str(service) +\
+                        " of cluster " + str(cluster))
+            logger.debug(pprint.pformat(result))
             if result is not None:
               self.resultQueue.put((ActionQueue.STATUS_COMMAND, result))
           except Exception, err:
@@ -172,7 +177,6 @@ class ActionQueue(threading.Thread):
     return result
 
   def executeCommand(self, command):
-    logger.info("Executing command \n" + pprint.pformat(command))
     clusterName = command['clusterName']
     commandId = command['commandId']
     hostname = command['hostname']
@@ -183,6 +187,11 @@ class ActionQueue(threading.Thread):
     configurations = command['configurations']
     result = []
 
+    logger.info("Executing command with id = " + str(commandId) +\
+                " for role = " + command['role'] + " of " +\
+                "cluster " + clusterName)
+    logger.debug(pprint.pformat(command))
+
     taskId = command['taskId']
     # Preparing 'IN_PROGRESS' report
     self.commandInProgress = {

ambari-agent/src/main/python/ambari_agent/Controller.py

@@ -113,11 +113,11 @@ class Controller(threading.Thread):
     """ Put the required actions into the Queue """
     """ Verify if the action is to reboot or not """
     if not commands:
-      logger.info("No commands from the server : " + pprint.pformat(commands))
+      logger.debug("No commands from the server : " + pprint.pformat(commands))
     else:
       """Only add to the queue if not empty list """
       for command in commands:
-        logger.info("Adding command to the action queue: \n" +
+        logger.debug("Adding command to the action queue: \n" +\
                      pprint.pformat(command))
         self.actionQueue.put(command)
         pass
@@ -150,7 +150,7 @@ class Controller(threading.Thread):
         response = self.sendRequest(self.heartbeatUrl, data)
         response = json.loads(response)
 
-        logger.info('Got server response: ' + pprint.pformat(response))
+        logger.debug('Got server response: ' + pprint.pformat(response))
         
         serverId=int(response['responseId'])
 

ambari-agent/src/main/python/ambari_agent/Heartbeat.py

@@ -54,18 +54,21 @@ class Heartbeat:
                   'hostname'          : hostname.hostname(),
                   'nodeStatus'        : nodeStatus
                 }
+
+    if len(queueResult) != 0:
+      heartbeat['reports'] = queueResult['reports']
+      heartbeat['componentStatus'] = queueResult['componentStatus']
+      pass
+    logger.info("Heartbeat : " + pformat(heartbeat))
+
     if (int(id) >= 0) and state_interval > 0 and (int(id) % state_interval) == 0:
       hostInfo = HostInfo()
       nodeInfo = { }
       # for now, just do the same work as registration
       hostInfo.register(nodeInfo)
       heartbeat['agentEnv'] = nodeInfo
+      logger.debug("agentEnv : " + str(nodeInfo))
 
-    if len(queueResult) != 0:
-      heartbeat['reports'] = queueResult['reports']
-      heartbeat['componentStatus'] = queueResult['componentStatus']
-      pass
-    logger.info("Heartbeat dump: " + pformat(heartbeat))
     return heartbeat
 
 def main(argv=None):

ambari-agent/src/main/python/ambari_agent/LiveStatus.py

@@ -125,8 +125,8 @@ class LiveStatus:
                                     read_stack_version(component["componentName"])
         }
         break
-    logger.info("The live status for component " + str(self.component) + " of service " + \
-                str(self.service) + " is " + str(livestatus))
+    logger.debug("The live status for component " + str(self.component) +\
+                " of service " + str(self.service) + " is " + str(livestatus))
     return livestatus
 
 def main(argv=None):

ambari-agent/src/main/python/ambari_agent/NetUtil.py

@@ -41,14 +41,14 @@ class NetUtil:
     """Try to connect to a given url. Result is True if url returns HTTP code 200, in any other case
     (like unreachable server or wrong HTTP code) result will be False
     """
-    logger.info("DEBUG:: Connecting to the following url " + url);
+    logger.info("Connecting to the following url " + url);
     try:
       parsedurl = urlparse(url)
       ca_connection = httplib.HTTPSConnection(parsedurl[1])
       ca_connection.request("GET", parsedurl[2])
       response = ca_connection.getresponse()  
       status = response.status    
-      logger.info("DEBUG: Calling url received " + str(status))
+      logger.info("Calling url received " + str(status))
       
       if status == 200: 
         return True

ambari-agent/src/main/python/ambari_agent/PuppetExecutor.py

@@ -157,7 +157,7 @@ class PuppetExecutor:
     rubyLib = ""
     if os.environ.has_key("RUBYLIB"):
       rubyLib = os.environ["RUBYLIB"]
-      logger.info("RUBYLIB from Env " + rubyLib)
+      logger.debug("RUBYLIB from Env " + rubyLib)
     if not (self.facterLib() in rubyLib):
       rubyLib = rubyLib + ":" + self.facterLib()
     if not (self.puppetLib() in rubyLib):
@@ -166,10 +166,10 @@ class PuppetExecutor:
     tmperr =  open(tmperrfile, 'w')
     puppetEnv["RUBYLIB"] = rubyLib
     puppetEnv = self.configureEnviron(puppetEnv)
-    logger.info("Setting RUBYLIB as: " + rubyLib)
+    logger.debug("Setting RUBYLIB as: " + rubyLib)
     logger.info("Running command " + pprint.pformat(puppetcommand))
     puppet = self.lauch_puppet_subprocess(puppetcommand,tmpout, tmperr, puppetEnv)
-    logger.info("Launching watchdog thread")
+    logger.debug("Launching watchdog thread")
     self.event.clear()
     self.last_puppet_has_been_killed = False
     thread = Thread(target =  self.puppet_watchdog_func, args = (puppet, ))

ambari-agent/src/main/python/ambari_agent/PythonExecutor.py

@@ -53,7 +53,7 @@ class PythonExecutor:
     pythonCommand = self.pythonCommand(file)
     logger.info("Running command " + pprint.pformat(pythonCommand))
     process = self.lauch_python_subprocess(pythonCommand, tmpout, tmperr)
-    logger.info("Launching watchdog thread")
+    logger.debug("Launching watchdog thread")
     self.event.clear()
     self.python_process_has_been_killed = False
     thread = Thread(target =  self.python_watchdog_func, args = (process, ))

ambari-agent/src/main/python/ambari_agent/StatusCheck.py

@@ -102,7 +102,7 @@ class StatusCheck:
     try:
       pidPath = None
       pidPattern = self.serToPidDict[serviceCode]
-      logger.info('pidPattern: ' + pidPattern)
+      logger.debug('pidPattern: ' + pidPattern)
     except KeyError as e:
       logger.warn('There is no mapping for ' + serviceCode)
       return None
@@ -110,7 +110,7 @@ class StatusCheck:
       for pidFile in self.pidFilesDict.keys():
         if re.match(pidPattern, pidFile):
           pidPath = self.pidFilesDict[pidFile]          
-      logger.info('pidPath: ' + str(pidPath))
+      logger.debug('pidPath: ' + str(pidPath))
       result = self.getIsLive(pidPath)
       return result
     except KeyError:

ambari-agent/src/main/python/ambari_agent/main.py

@@ -80,17 +80,18 @@ def main():
   if options.verbose:
     logging.basicConfig(format=formatstr, level=logging.DEBUG, filename=logfile)
     logger.setLevel(logging.DEBUG)
+    logger.info("loglevel=logging.DEBUG")
   else:
     logging.basicConfig(format=formatstr, level=logging.INFO, filename=logfile)
     logger.setLevel(logging.INFO)
-
-  logger.debug("loglevel=logging.DEBUG")
+    logger.info("loglevel=logging.INFO")
 
   default_cfg = { 'agent' : { 'prefix' : '/home/ambari' } }
   config = ConfigParser.RawConfigParser(default_cfg)
   signal.signal(signal.SIGINT, signal_handler)
   signal.signal(signal.SIGTERM, signal_handler)
   signal.signal(signal.SIGUSR1, debug)
+
   if (len(sys.argv) >1) and sys.argv[1]=='stop':
     # stop existing Ambari agent
     pid = -1
@@ -143,6 +144,21 @@ def main():
 
   killstaleprocesses()
 
+  # Setting loglevel based on config file
+  try:
+    loglevel = config.get('agent', 'loglevel')
+    if loglevel is not None:
+      if loglevel == 'DEBUG':
+        logging.basicConfig(format=formatstr, level=logging.DEBUG, filename=logfile)
+        logger.setLevel(logging.DEBUG)
+        logger.info("Newloglevel=logging.DEBUG")
+      else:
+        logging.basicConfig(format=formatstr, level=logging.INFO, filename=logfile)
+        logger.setLevel(logging.INFO)
+        logger.debug("Newloglevel=logging.INFO")
+  except Exception, err:
+    logger.info("Default loglevel=DEBUG")
+
   server_url = 'https://' + config.get('server', 'hostname') + ':' + config.get('server', 'url_port')
   print("Connecting to the server at " + server_url + "...")
   logger.info('Connecting to the server at: ' + server_url)

ambari-agent/src/main/python/ambari_agent/manifestGenerator.py

@@ -25,7 +25,7 @@ from uuid import getnode as get_mac
 from shell import shellRunner
 from datetime import datetime
 import AmbariConfig
-
+import pprint
 
 logger = logging.getLogger()
 
@@ -47,8 +47,8 @@ def writeImports(outputFile, modulesdir, importsList):
 
 
 def generateManifest(parsedJson, fileName, modulesdir, ambariconfig):
-  logger.info("JSON Received:")
-  logger.info(json.dumps(parsedJson, sort_keys=True, indent=4))
+  logger.debug("JSON Received:")
+  logger.debug(json.dumps(parsedJson, sort_keys=True, indent=4))
 #reading json
   hostname = parsedJson['hostname']
   clusterHostInfo = {} 
@@ -164,6 +164,7 @@ def writeHostAttributes(outputFile, hostAttributes):
 #write flat configurations
 def writeFlatConfigurations(outputFile, flatConfigs):
   flatDict = {}
+  logger.info("Generating global configurations =>\n" + pprint.pformat(flatConfigs))
   for flatConfigName in flatConfigs.iterkeys():
     for flatConfig in flatConfigs[flatConfigName].iterkeys():
       flatDict[flatConfig] = flatConfigs[flatConfigName][flatConfig]
@@ -175,9 +176,8 @@ def writeNonGlobalConfigurations(outputFile, xmlConfigs):
   outputFile.write('$configuration =  {\n')
 
   for configName in xmlConfigs.iterkeys():
-
     config = xmlConfigs[configName]
-    
+    logger.info("Generating " + configName + ",configurations =>\n" + pprint.pformat(config))
     outputFile.write(configName + '=> {\n')
     coma = ''
     for configParam in config.iterkeys():
@@ -235,6 +235,7 @@ def writeTasks(outputFile, roles, ambariconfig, clusterHostInfo=None,
 
     stageNum = stageNum + 1
   outputFile.write('}\n')
+
 def normalizeTaskParams(taskParams):
   result = ''
   coma = ''