Merge pull request #52 from DataDog/51-pid-file

Fixing pid file location

.travis.yml

@@ -12,3 +12,4 @@ before_script:
   - sudo apt-get install python-mysqldb
   - sudo apt-get install nginx
   - sudo apt-get install apache2
+  - sudo apt-get install sysstat

agent.py

@@ -19,6 +19,10 @@
 import time
 import urllib
 
+# Constants
+PID_DIR="/var/run/dd-agent"
+PID_FILE="dd-agent.pid"
+
 # Watchdog implementation
 from threading import Timer
 WATCHDOG_MULTIPLIER = 10 # will fire if no checks have been collected in N * checkFreq, 150s by default
@@ -31,29 +35,14 @@
 
 # Custom modules
 from checks.common import checks
+from checks.ec2 import EC2
 from config import get_config, get_system_stats, get_parsed_args
 from daemon import Daemon
 from emitter import http_emitter
 
 # Override the generic daemon class to run our checks
 class agent(Daemon):    
 
-    EC2_URL = "http://169.254.169.254/latest/meta-data/instance-id"
-
-    @staticmethod
-    def get_ec2_instance_id():
-        """Fetch EC2 instance ID if possible. If not on EC2 returns None"""
-        try:
-            url = urllib.urlopen(agent.EC2_URL)
-            instanceId = url.read()
-            assert instanceId.startswith("i-"), "Malformed instance-id: %s" % instanceId
-            return instanceId
-
-        except Exception, e:
-            logging.getLogger('agent').exception('Cannot determine instance-id. Is this machine on EC2?')
-
-        return None
-
     def late(self, cks, threshold, crash=True):
         """Determine whether the agent run is late and optionally kill it if so.
         """
@@ -75,27 +64,25 @@ def run(self, agentConfig=None, run_forever=True):
         agentLogger.debug('Creating checks instance')
 
         if agentConfig is None:
-            agentConfig, rawConfig = get_config()
-        else:
-            rawConfig = {}
+            agentConfig = get_config()
 
         # Try to fetch instance Id from EC2 if not hostname has been set
         # in the config file
         if agentConfig.get('hostname') is None and agentConfig.get('useEC2InstanceId'):
-            instanceId = self.get_ec2_instance_id()
+            instanceId = EC2.get_instance_id()
             if instanceId is not None:
                 agentLogger.info("Running on EC2, instanceId: %s" % instanceId)
                 agentConfig['hostname'] = instanceId
             else:
-                agentLogger.info('Not running on EC2')
+                agentLogger.info('Not running on EC2, using hostname to identify this server')
 
         emitter = http_emitter
 
         checkFreq = int(agentConfig['checkFreq'])
         lateThresh = checkFreq * WATCHDOG_MULTIPLIER
 
         # Checks instance
-        c = checks(agentConfig, rawConfig, emitter)
+        c = checks(agentConfig, emitter)
 
         # Run once
         c.doChecks(True, systemStats)
@@ -120,9 +107,10 @@ def run(self, agentConfig=None, run_forever=True):
             agentLogger.debug("Getting ready to sleep for %s seconds." % lateThresh)
 
 def setupLogging(agentConfig):
-    """Used by ddagent.py as well"""
+    """Configure logging to use syslog whenever possible.
+    Also controls debugMode."""
     if agentConfig['debugMode']:
-        logFile = os.path.join(agentConfig['tmpDirectory'], 'dd-agent.log')
+        logFile = "/tmp/dd-agent.log"
         logging.basicConfig(filename=logFile, filemode='w', level=logging.DEBUG, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
         logging.info("Logging to %s" % logFile)
     else:
@@ -146,49 +134,68 @@ def setupLogging(agentConfig):
             sys.stderr.write("Error while setting up syslog logging (%s). No logging available" % str(e))
             logging.disable(logging.ERROR)
 
-def getPidFile(command, agentConfig, clean):
-    """Used by ddagent.py as well"""
-
-    if command == 'init': 
-        # This path added for newer Linux packages which run under
-        # a separate dd-agent user account.
-        if os.path.exists('/var/run/dd-agent/'):
-            pidFile = '/var/run/dd-agent/dd-agent.pid'
-        else:
-            pidFile = '/var/run/dd-agent.pid'
-
+def getPidFile(pid_dir=PID_DIR):
+    """Find a good spot for the pid file.
+    By default PID_DIR/PID_FILE
+    """
+    try:
+        # Can we write to the directory
+        if os.access(pid_dir, os.W_OK):
+            pidfile = os.path.join(pid_dir, PID_FILE)
+            logging.info("Pid file is: %s" % pidfile)
+            return pidfile
+    except:
+        logging.exception("Cannot locate pid file, defaulting to /tmp/%s" % PID_FILE)
+        # continue
+
+    # if all else fails
+    if os.access("/tmp", os.W_OK):
+        logging.warn("Pid file: /tmp/%s" % PID_FILE)
+        return os.path.join("/tmp", PID_FILE)
     else:
-        pidFile = os.path.join(agentConfig['pidfileDirectory'], 'dd-agent.pid')
-
-    if clean:
-        logging.debug('Agent called with --clean option, removing .pid')
-        try:
-            os.remove(pidFile)
-        except OSError:
-            # Did not find pid file
-            pass
-
-    return pidFile
+        # Can't save pid file, bail out
+        logging.error("Cannot save pid file anywhere")
+        sys.exit(-2)
+
+def cleanPidFile(pid_dir=PID_DIR):
+    try:
+        logging.debug("Cleaning up pid file %s" % getPidFile(pid_dir))
+        os.remove(getPidFile(pid_dir))
+        return True
+    except:
+        logging.exception("Could not clean up pid file")
+        return False
+
+def getPid(pid_dir=PID_DIR):
+    "Retrieve the actual pid"
+    try:
+        pf = open(getPidFile(pid_dir))
+        pid_s = pf.read()
+        pf.close()
+
+        return int(pid_s.strip())
+    except:
+        logging.exception("Cannot read pid")
+        return None
 
 # Control of daemon     
 if __name__ == '__main__':  
 
     options, args = get_parsed_args()
-    agentConfig, rawConfig = get_config()
+    agentConfig = get_config()
 
     # Logging
     setupLogging(agentConfig)
 
-    # FIXME
-    # Ever heard of optparse?
-
     argLen = len(sys.argv)
 
     if len(args) > 0:
         command = args[0]
+
+        if options.clean:
+            cleanPidFile()
 
-        # Daemon instance from agent class
-        pidFile = getPidFile(command, agentConfig, options.clean)
+        pidFile = getPidFile()
         daemon = agent(pidFile)
 
         if 'start' == command:
@@ -208,16 +215,8 @@ def getPidFile(command, agentConfig, clean):
             daemon.run()
 
         elif 'status' == command:
-            try:
-                pf = file(pidFile,'r')
-                pid = int(pf.read().strip())
-                pf.close()
-            except IOError:
-                pid = None
-            except SystemExit:
-                pid = None
-
-            if pid:
+            pid = getPid()
+            if pid is not None:
                 sys.stdout.write('dd-agent is running as pid %s.\n' % pid)
                 logging.info("dd-agent is running as pid %s." % pid)
             else:

checks/common.py

@@ -65,9 +65,8 @@ def wrapper(*args, **kwargs):
     return wrapper
 
 class checks:
-    def __init__(self, agentConfig, rawConfig, emitter):
+    def __init__(self, agentConfig, emitter):
         self.agentConfig = agentConfig
-        self.rawConfig = rawConfig
         self.plugins = None
         self.emitter = emitter
         self.last_post_ts = None

checks/ec2.py

@@ -8,13 +8,14 @@
 class EC2(Check):
     """Retrieve EC2 metadata
     """
-    URL = "http://169.254.169.254/1.0/meta-data/"
+    URL = "http://169.254.169.254/latest/meta-data/"
     TIMEOUT = 0.1 # second
 
     def __init__(self, logger):
         Check.__init__(self, logger)
 
-    def get_metadata(self):
+    @staticmethod
+    def get_metadata():
         """Use the ec2 http service to introspect the instance. This adds latency if not running on EC2
         """
         # >>> import urllib2
@@ -42,7 +43,7 @@ def get_metadata(self):
                 assert type(v) in (types.StringType, types.UnicodeType) and len(v) > 0, "%s is not a string" % v
                 metadata[k] = v
             except:
-                self.logger.exception("(Ignore if !ec2) Cannot extract EC2 metadata %s" % k)
+                pass
 
         try:
             if socket_to is None:
@@ -52,3 +53,10 @@ def get_metadata(self):
             pass
 
         return metadata
+
+    @staticmethod
+    def get_instance_id():
+        try:
+            return EC2.get_metadata().get("instance-id", None)
+        except:
+            return None

config.py

@@ -9,6 +9,7 @@
 
 # CONSTANTS
 DATADOG_CONF = "datadog.conf"
+DEFAULT_CHECK_FREQUENCY = 15 # seconds
 
 def get_parsed_args():
     parser = OptionParser()
@@ -25,7 +26,7 @@ def get_parsed_args():
     return options, args
 
 def get_version():
-    return "2.2.16"
+    return "2.2.17"
 
 def skip_leading_wsp(f):
     "Works on a file, returns a file-like object"
@@ -42,11 +43,9 @@ def get_config(parse_args = True, cfg_path=None):
     agentConfig = {}
     agentConfig['debugMode'] = False
     # not really a frequency, but the time to sleep between checks
-    agentConfig['checkFreq'] = 15
+    agentConfig['checkFreq'] = DEFAULT_CHECK_FREQUENCY
     agentConfig['version'] = get_version()
 
-    rawConfig = {}
-
     # Config handling
     try:
         # Find the right config file
@@ -67,7 +66,11 @@ def get_config(parse_args = True, cfg_path=None):
         config = ConfigParser.ConfigParser()
         config.readfp(skip_leading_wsp(open(config_path)))
 
+        #
         # Core config
+        #
+
+        # Where to send the data
         if options is not None and options.use_forwarder:
             listen_port = 17123
             if config.has_option('Main','listen_port'):
@@ -80,14 +83,10 @@ def get_config(parse_args = True, cfg_path=None):
         if agentConfig['ddUrl'].endswith('/'):
             agentConfig['ddUrl'] = agentConfig['ddUrl'][:-1]
 
+        # Which API key to use
         agentConfig['apiKey'] = config.get('Main', 'api_key')
-
-        if os.path.exists('/var/log/dd-agent/'):
-            agentConfig['tmpDirectory'] = '/var/log/dd-agent/'
-        else:
-            agentConfig['tmpDirectory'] = '/tmp/' # default which may be overriden in the config later
-        agentConfig['pidfileDirectory'] = agentConfig['tmpDirectory']
-
+
+        # Debug mode
         agentConfig['debugMode'] = config.get('Main', 'debug_mode').lower() in ("yes", "true")
 
         if config.has_option('Main', 'use_ec2_instance_id'):    
@@ -98,7 +97,10 @@ def get_config(parse_args = True, cfg_path=None):
             agentConfig['useEC2InstanceId'] = False
 
         if config.has_option('Main', 'check_freq'):
-            agentConfig['checkFreq'] = int(config.get('Main', 'check_freq'))
+            try:
+                agentConfig['checkFreq'] = int(config.get('Main', 'check_freq'))
+            except:
+                agentConfig['checkFreq'] = DEFAULT_CHECK_FREQUENCY
 
         if config.has_option('Main','hostname'):
             agentConfig['hostname'] = config.get('Main','hostname')
@@ -161,12 +163,6 @@ def get_config(parse_args = True, cfg_path=None):
         if config.has_option('Main', 'nginx_status_url'):
             agentConfig['nginxStatusUrl'] = config.get('Main', 'nginx_status_url')
 
-        if config.has_option('Main', 'tmp_directory'):
-            agentConfig['tmpDirectory'] = config.get('Main', 'tmp_directory')
-
-        if config.has_option('Main', 'pidfile_directory'):
-            agentConfig['pidfileDirectory'] = config.get('Main', 'pidfile_directory')
-
         if config.has_option('Main', 'plugin_directory'):
             agentConfig['pluginDirectory'] = config.get('Main', 'plugin_directory')
 
@@ -308,13 +304,7 @@ def get_config(parse_args = True, cfg_path=None):
             sys.stderr.write('You have configured MongoDB for monitoring, but the pymongo module is not installed.\n')
             sys.exit(2)
 
-    for section in config.sections():
-        rawConfig[section] = {}
-
-        for option in config.options(section):
-            rawConfig[section][option] = config.get(section, option)
-
-    return agentConfig, rawConfig
+    return agentConfig
 
 def get_system_stats():
     systemStats = {

datadog.conf.example

@@ -17,7 +17,7 @@ debug_mode: no
 
 # Use the amazon EC2 instance-id instead of hostname (unless hostname is
 # explicitly set)
-use_ec2_instance_id: no
+use_ec2_instance_id: yes
 
 # Use mount points instead of volumes to track disk and fs metrics
 use_mount: no

ddagent.py

@@ -227,7 +227,7 @@ def main():
 
     import tornado.httpclient
 
-    agentConfig, rawConfig = get_config(parse_args = False)
+    agentConfig = get_config(parse_args = False)
 
     port = agentConfig.get('listen_port', None)
     if port is None: