Tracebacks on agent.py info and changed how get_config() receives options and args, fixes #329

agent.py

@@ -63,7 +63,7 @@ def _handle_sigterm(self, signum, frame):
         if self.collector:
             self.collector.stop()
 
-    def run(self):
+    def run(self, config=None):
         """Main loop of the collector"""
 
         # Gracefully exit on sigterm.
@@ -73,7 +73,10 @@ def run(self):
         CollectorStatus().persist()
 
         # Intialize the collector.
-        agentConfig = self._set_agent_config_hostname(get_config())
+        if not config:
+            config = get_config(parse_args=True)
+
+        agentConfig = self._set_agent_config_hostname(config)
         systemStats = get_system_stats()
         emitters = self._get_emitters(agentConfig)
         self.collector = Collector(agentConfig, emitters, systemStats)
@@ -168,7 +171,7 @@ def setup_logging(agentConfig):
 
 def main():
     options, args = get_parsed_args()
-    agentConfig = get_config()
+    agentConfig = get_config(options=options)
 
     # Logging
     setup_logging(agentConfig)
@@ -188,7 +191,6 @@ def main():
         return 2
 
     command = args[0]
-
     if command not in COMMANDS:
         sys.stderr.write("Unknown command: %s\n" % command)
         return 3
@@ -217,7 +219,7 @@ def main():
 
         elif 'foreground' == command:
             logging.info('Running in foreground')
-            agent.run()
+            agent.run(config=agentConfig)
 
     # Commands that don't need the agent to be initialized.
     else:
@@ -229,7 +231,7 @@ def main():
                 sys.stdout.write('dd-agent is not running.\n')
 
         elif 'info' == command:
-            return CollectorStatus.print_latest_status()
+            return CollectorStatus.print_latest_status(verbose=options.verbose)
 
     return 0
 

checks/__init__.py

@@ -10,6 +10,7 @@
 import time
 import types
 import os
+import sys
 
 from util import LaconicFilter
 from checks import check_status
@@ -417,7 +418,8 @@ def run(self):
                 instance_status = check_status.InstanceStatus(i, check_status.STATUS_OK)
             except Exception, e:
                 self.log.exception("Check '%s' instance #%s failed" % (self.name, i))
-                instance_status = check_status.InstanceStatus(i, check_status.STATUS_ERROR, e)
+                # Send the traceback (located at sys.exc_info()[2]) into the InstanceStatus otherwise a traceback won't be able to be printed
+                instance_status = check_status.InstanceStatus(i, check_status.STATUS_ERROR, e, sys.exc_info()[2])
             instance_statuses.append(instance_status)
         return instance_statuses
 

checks/check_status.py

@@ -11,6 +11,7 @@
 import platform
 import sys
 import tempfile
+import traceback
 
 # project
 import config
@@ -162,7 +163,8 @@ def load_latest_status(cls):
             return None
 
     @classmethod
-    def print_latest_status(cls):
+    def print_latest_status(cls, verbose=False):
+        cls.verbose = verbose
         Stylizer.ENABLED = False
         try:
             if sys.stdout.isatty():
@@ -190,11 +192,16 @@ def _get_pickle_path(cls):
 
 class InstanceStatus(object):
 
-    def __init__(self, instance_id, status, error=None):
+    def __init__(self, instance_id, status, error=None, tb=None):
         self.instance_id = instance_id
         self.status = status
         self.error = repr(error)
 
+        if (type(tb).__name__ == 'traceback'):
+            self.traceback = traceback.format_tb(tb)
+        else:
+            self.traceback = None
+
     def has_error(self):
         return self.status != STATUS_OK
 
@@ -289,6 +296,19 @@ def body_lines(self):
                     "    - Collected %s metrics & %s events" % (cs.metric_count, cs.event_count),
                     ""
                 ]
+
+                if self.verbose and s.traceback is not None:
+                    # Formatting the traceback to look like a python traceback
+                    check_lines.append("    Traceback (most recent call last):")
+
+                    # Format the traceback lines to look good in the output
+                    for tb_line in s.traceback:
+                        lines = tb_line.split('\n')
+                        for line in lines:
+                            if line.strip() == '':
+                                continue
+                            check_lines.append('    ' + line)
+
                 lines += check_lines
 
         # Emitter status

config.py

@@ -30,6 +30,8 @@ def get_parsed_args():
                         default=False,dest='use_forwarder')
     parser.add_option('-n', '--disable-dd', action='store_true', default=False,
                         dest="disable_dd")
+    parser.add_option('-v', '--verbose', action='store_true', default=False,
+                        dest='verbose', help='Print out available tracebacks for errors in checks')
     try:
         options, args = parser.parse_args()
     except SystemExit:
@@ -141,11 +143,9 @@ def get_config_path(cfg_path=None, os_name=None):
     sys.stderr.write("Please supply a configuration file at %s or in the directory where the agent is currently deployed.\n" % str(exc))
     sys.exit(3)
 
-def get_config(parse_args = True, cfg_path=None, init_logging=False, options=None):
+def get_config(parse_args=False, cfg_path=None, init_logging=False, options=None):
     if parse_args:
         options, args = get_parsed_args()
-    elif not options:
-        args = None
 
     # General config
     agentConfig = {