Add log collection tool and thread

azurelinuxagent/agent.py

@@ -24,21 +24,23 @@
 from __future__ import print_function
 
 import os
-import sys
 import re
 import subprocess
+import sys
 import threading
 import traceback
 
-import azurelinuxagent.common.logger as logger
-import azurelinuxagent.common.event as event
 import azurelinuxagent.common.conf as conf
-from azurelinuxagent.common.version import AGENT_NAME, AGENT_LONG_VERSION, \
-                                     DISTRO_NAME, DISTRO_VERSION, \
-                                     PY_VERSION_MAJOR, PY_VERSION_MINOR, \
-                                     PY_VERSION_MICRO, GOAL_STATE_AGENT_VERSION
+import azurelinuxagent.common.event as event
+import azurelinuxagent.common.logger as logger
+from azurelinuxagent.common.future import ustr
+from azurelinuxagent.common.logcollector import LogCollector, OUTPUT_RESULTS_FILE_PATH
 from azurelinuxagent.common.osutil import get_osutil
 from azurelinuxagent.common.utils import fileutil
+from azurelinuxagent.common.version import AGENT_NAME, AGENT_LONG_VERSION, \
+    DISTRO_NAME, DISTRO_VERSION, \
+    PY_VERSION_MAJOR, PY_VERSION_MINOR, \
+    PY_VERSION_MICRO, GOAL_STATE_AGENT_VERSION
 
 
 class Agent(object):
@@ -49,17 +51,17 @@ def __init__(self, verbose, conf_file_path=None):
         self.conf_file_path = conf_file_path
         self.osutil = get_osutil()
 
-        #Init stdout log
+        # Init stdout log
         level = logger.LogLevel.VERBOSE if verbose else logger.LogLevel.INFO
         logger.add_logger_appender(logger.AppenderType.STDOUT, level)
 
-        #Init config
+        # Init config
         conf_file_path = self.conf_file_path \
                 if self.conf_file_path is not None \
                     else self.osutil.get_agent_conf_file_path()
         conf.load_conf_from_file(conf_file_path)
 
-        #Init log
+        # Init log
         verbose = verbose or conf.get_logs_verbose()
         level = logger.LogLevel.VERBOSE if verbose else logger.LogLevel.INFO
         logger.add_logger_appender(logger.AppenderType.FILE, level,
@@ -104,7 +106,6 @@ def daemon(self):
         child_args = None \
             if self.conf_file_path is None \
                 else "-configuration-path:{0}".format(self.conf_file_path)
-
         from azurelinuxagent.daemon import get_daemon_handler
         daemon_handler = get_daemon_handler()
         daemon_handler.run(child_args=child_args)
@@ -151,6 +152,22 @@ def show_configuration(self):
         for k in sorted(configuration.keys()):
             print("{0} = {1}".format(k, configuration[k]))
 
+    def collect_logs(self, is_full_mode):
+        if is_full_mode:
+            print("Running log collector mode full")
+        else:
+            print("Running log collector mode normal")
+
+        try:
+            log_collector = LogCollector(is_full_mode)
+            archive = log_collector.collect_logs_and_get_archive()
+            print("Log collection successfully completed. Archive can be found at {0} "
+                  "and detailed log output can be found at {1}".format(archive, OUTPUT_RESULTS_FILE_PATH))
+        except Exception as e: # pylint: disable=C0103
+            print("Log collection completed unsuccessfully. Error: {0}".format(ustr(e)))
+            print("Detailed log output can be found at {0}".format(OUTPUT_RESULTS_FILE_PATH))
+            sys.exit(1)
+
 
 def main(args=[]): # pylint: disable=R0912,W0102
     """
@@ -159,7 +176,7 @@ def main(args=[]): # pylint: disable=R0912,W0102
     """
     if len(args) <= 0: # pylint: disable=len-as-condition
         args = sys.argv[1:]
-    command, force, verbose, debug, conf_file_path = parse_args(args)
+    command, force, verbose, debug, conf_file_path, log_collector_full_mode = parse_args(args)
     if command == "version":
         version()
     elif command == "help":
@@ -183,11 +200,14 @@ def main(args=[]): # pylint: disable=R0912,W0102
                 agent.run_exthandlers(debug)
             elif command == "show-configuration":
                 agent.show_configuration()
+            elif command == "collect-logs":
+                agent.collect_logs(log_collector_full_mode)
         except Exception:
             logger.error(u"Failed to run '{0}': {1}",
                          command,
                          traceback.format_exc())
 
+
 def parse_args(sys_args): # pylint: disable=R0912
     """
     Parse command line arguments
@@ -197,45 +217,50 @@ def parse_args(sys_args): # pylint: disable=R0912
     verbose = False
     debug = False
     conf_file_path = None
-    for a in sys_args: # pylint: disable=C0103
-        m = re.match("^(?:[-/]*)configuration-path:([\w/\.\-_]+)", a) # pylint: disable=W1401,C0103
+    log_collector_full_mode = False
+
+    for arg in sys_args:
+        m = re.match("^(?:[-/]*)configuration-path:([\w/\.\-_]+)", arg) # pylint: disable=W1401,C0103
         if not m is None:
             conf_file_path = m.group(1)
             if not os.path.exists(conf_file_path):
                 print("Error: Configuration file {0} does not exist".format(
                         conf_file_path), file=sys.stderr) 
-                usage()
+                print(usage())
                 sys.exit(1)
-
-        elif re.match("^([-/]*)deprovision\\+user", a):
+        elif re.match("^([-/]*)deprovision\\+user", arg):
             cmd = "deprovision+user"
-        elif re.match("^([-/]*)deprovision", a):
+        elif re.match("^([-/]*)deprovision", arg):
             cmd = "deprovision"
-        elif re.match("^([-/]*)daemon", a):
+        elif re.match("^([-/]*)daemon", arg):
             cmd = "daemon"
-        elif re.match("^([-/]*)start", a):
+        elif re.match("^([-/]*)start", arg):
             cmd = "start"
-        elif re.match("^([-/]*)register-service", a):
+        elif re.match("^([-/]*)register-service", arg):
             cmd = "register-service"
-        elif re.match("^([-/]*)run-exthandlers", a):
+        elif re.match("^([-/]*)run-exthandlers", arg):
             cmd = "run-exthandlers"
-        elif re.match("^([-/]*)version", a):
+        elif re.match("^([-/]*)version", arg):
             cmd = "version"
-        elif re.match("^([-/]*)verbose", a):
+        elif re.match("^([-/]*)verbose", arg):
             verbose = True
-        elif re.match("^([-/]*)debug", a):
+        elif re.match("^([-/]*)debug", arg):
             debug = True
-        elif re.match("^([-/]*)force", a):
+        elif re.match("^([-/]*)force", arg):
             force = True
-        elif re.match("^([-/]*)show-configuration", a):
+        elif re.match("^([-/]*)show-configuration", arg):
             cmd = "show-configuration"
-        elif re.match("^([-/]*)(help|usage|\\?)", a):
+        elif re.match("^([-/]*)(help|usage|\\?)", arg):
             cmd = "help"
+        elif re.match("^([-/]*)collect-logs", arg):
+            cmd = "collect-logs"
+        elif re.match("^([-/]*)full", arg):
+            log_collector_full_mode = True
         else:
             cmd = "help"
             break
 
-    return cmd, force, verbose, debug, conf_file_path
+    return cmd, force, verbose, debug, conf_file_path, log_collector_full_mode
 
 
 def version():
@@ -250,6 +275,7 @@ def version():
                                        PY_VERSION_MICRO))
     print("Goal state agent: {0}".format(GOAL_STATE_AGENT_VERSION))
 
+
 def usage():
     """
     Return agent usage message
@@ -258,11 +284,12 @@ def usage():
     s += ("usage: {0} [-verbose] [-force] [-help] " # pylint: disable=C0103
            "-configuration-path:<path to configuration file>" 
            "-deprovision[+user]|-register-service|-version|-daemon|-start|"
-           "-run-exthandlers|-show-configuration]"
+           "-run-exthandlers|-show-configuration|-collect-logs [-full]"
            "").format(sys.argv[0])
     s += "\n" # pylint: disable=C0103
     return s
 
+
 def start(conf_file_path=None):
     """
     Start agent daemon in a background process and set stdout/stderr to
@@ -274,5 +301,6 @@ def start(conf_file_path=None):
         args.append('-configuration-path:{0}'.format(conf_file_path))
     subprocess.Popen(args, stdout=devnull, stderr=devnull)
 
+
 if __name__ == '__main__' :
     main()

azurelinuxagent/common/cgroupapi.py

@@ -41,6 +41,7 @@
 VM_AGENT_CGROUP_NAME = "walinuxagent.service"
 EXTENSIONS_ROOT_CGROUP_NAME = "walinuxagent.extensions"
 UNIT_FILES_FILE_SYSTEM_PATH = "/etc/systemd/system"
+SYSTEMD_RUN_PATH = "/run/systemd/system/"
 
 
 class CGroupsApi(object):
@@ -94,15 +95,15 @@ def create():
         """
         Factory method to create the correct API for the current platform
         """
-        return SystemdCgroupsApi() if CGroupsApi._is_systemd() else FileSystemCgroupsApi()
+        return SystemdCgroupsApi() if CGroupsApi.is_systemd() else FileSystemCgroupsApi()
 
     @staticmethod
-    def _is_systemd():
+    def is_systemd():
         """
         Determine if systemd is managing system services; the implementation follows the same strategy as, for example,
         sd_booted() in libsystemd, or /usr/sbin/service
         """
-        return os.path.exists('/run/systemd/system/')
+        return os.path.exists(SYSTEMD_RUN_PATH)
 
     @staticmethod
     def _foreach_controller(operation, message):

azurelinuxagent/common/conf.py

@@ -97,6 +97,7 @@ def load_conf_from_file(conf_file_path, conf=__conf__):
     "OS.CheckRdmaDriver": False,
     "Logs.Verbose": False,
     "Logs.Console": True,
+    "Logs.Collect": False,
     "Extensions.Enabled": True,
     "Provisioning.AllowResetSysUser": False,
     "Provisioning.RegenerateSshHostKeyPair": False,
@@ -149,7 +150,8 @@ def load_conf_from_file(conf_file_path, conf=__conf__):
     "Provisioning.PasswordCryptSaltLength": 10,
     "HttpProxy.Port": None,
     "ResourceDisk.SwapSizeMB": 0,
-    "Autoupdate.Frequency": 3600
+    "Autoupdate.Frequency": 3600,
+    "Logs.CollectPeriod": 3600
 }
 
 
@@ -223,6 +225,14 @@ def get_logs_console(conf=__conf__):
     return conf.get_switch("Logs.Console", True)
 
 
+def get_collect_logs(conf=__conf__):
+    return conf.get_switch("Logs.Collect", False)
+
+
+def get_collect_logs_period(conf=__conf__):
+    return conf.get_int("Logs.CollectPeriod", 3600)
+
+
 def get_lib_dir(conf=__conf__):
     return conf.get("Lib.Dir", "/var/lib/waagent")
 
@@ -374,9 +384,11 @@ def get_password_crypt_salt_len(conf=__conf__):
 def get_monitor_hostname(conf=__conf__):
     return conf.get_switch("Provisioning.MonitorHostName", False)
 
+
 def get_monitor_hostname_period(conf=__conf__):
     return conf.get_int("Provisioning.MonitorHostNamePeriod", 30)
 
+
 def get_httpproxy_host(conf=__conf__):
     return conf.get("HttpProxy.Host", None)
 
@@ -395,10 +407,12 @@ def get_resourcedisk_format(conf=__conf__):
 
 def get_resourcedisk_enable_swap(conf=__conf__):
     return conf.get_switch("ResourceDisk.EnableSwap", False)
-
+
+
 def get_resourcedisk_enable_swap_encryption(conf=__conf__):
     return conf.get_switch("ResourceDisk.EnableSwapEncryption", False)
 
+
 def get_resourcedisk_mountpoint(conf=__conf__):
     return conf.get("ResourceDisk.MountPoint", "/mnt/resource")
 

azurelinuxagent/common/event.py

@@ -101,6 +101,7 @@ class WALAEventOperation: # pylint: disable=R0903,no-init
     InitializeHostPlugin = "InitializeHostPlugin"
     InvokeCommandUsingSystemd = "InvokeCommandUsingSystemd"
     Log = "Log"
+    LogCollection = "LogCollection"
     OSInfo = "OSInfo"
     Partition = "Partition"
     PluginSettingsVersionMismatch = "PluginSettingsVersionMismatch"