def _install_psutils():
_install_gcc()
_install_pip()
if waagent.Run("pip install psutil", chk_err=False) == 0:
waagent.Log("psutil installed")
else:
waagent.Log("Error installing psutil")
def install_package(package_name):
if DistroName == "centos" or DistroName == "redhat":
cmd = "yum -y install " + package_name
elif DistroName == "ubuntu":
cmd = "apt-get -y install " + package_name
elif DistroName == "suse":
if not os.listdir('/etc/zypp/repos.d'):
waagent.Run("zypper ar http://download.opensuse.org/distribution/13.2/repo/oss/suse/ opensuse")
cmd = "zypper -n --gpg-auto-import-keys install --force-resolution -l " + package_name
else:
cmd = "zypper -n install --force-resolution -l " + package_name
else:
raise Exception("Unsupported Linux Distro.")
waagent.Log("The command to install {0}: {1}".format(package_name, cmd))
attempt = 1
while(True):
waagent.Log("Installing package {0} (Attempt {1})".format(package_name, attempt))
retcode, retoutput = waagent.RunGetOutput(cmd)
if retcode == 0:
waagent.Log("package {0} installation succeeded".format(package_name))
break
else:
waagent.Log("package {0} installation failed {1}:\n {2}".format(package_name, retcode, retoutput))
if attempt < 3:
attempt += 1
time.sleep(5)
if DistroName == 'suse' and retcode == 104:
waagent.Run("zypper ar http://download.opensuse.org/distribution/13.2/repo/oss/suse/ opensuse")
cmd = "zypper -n --gpg-auto-import-keys install --force-resolution -l " + package_name
continue
else:
raise Exception("failed to install package {0}:{1}".format(package_name, retcode))
def enable(hutil):
pidFile = os.path.join(aem.LibDir, "pid");
#Check whether monitor process is running.
#If it does, return. Otherwise clear pid file
if os.path.isfile(pidFile):
pid = waagent.GetFileContents(pidFile)
if os.path.isdir(os.path.join("/proc", pid)):
if hutil.is_seq_smaller():
hutil.do_exit(0, 'Enable', 'success', '0',
'Azure Enhanced Monitor is already running')
else:
waagent.Log("Stop old daemon: {0}".format(pid))
os.kill(int(pid), 9)
os.remove(pidFile)
args = [os.path.join(os.getcwd(), __file__), "daemon"]
devnull = open(os.devnull, 'w')
child = subprocess.Popen(args, stdout=devnull, stderr=devnull)
if child.pid == None or child.pid < 1:
hutil.do_exit(1, 'Enable', 'error', '1',
'Failed to launch Azure Enhanced Monitor')
else:
hutil.save_seq()
waagent.SetFileContents(pidFile, str(child.pid))
waagent.Log(("Daemon pid: {0}").format(child.pid))
hutil.do_exit(0, 'Enable', 'success', '0',
'Azure Enhanced Monitor is enabled')
def init_suse_hostsfile(host_name, ipaddrs):
hostsfile = '/etc/hosts'
if not os.path.isfile(hostsfile):
return
try:
newhpcd_entries = ''
for ipaddr in ipaddrs:
newhpcd_entries += '{0:24}{1:30}#HPCD\n'.format(ipaddr, host_name)
curhpcd_entries = ''
newcontent = ''
hpcentryexists = False
with open(hostsfile, 'r') as F:
for line in F.readlines():
if re.match(r"^[0-9\.]+\s+[^\s#]+\s+#HPCD\s*$", line):
curhpcd_entries += line
hpcentryexists = True
elif re.match(r"^[0-9\.]+\s+[^\s#]+\s+#HPC\s*$", line):
hpcentryexists = True
else:
newcontent += line
if newhpcd_entries != curhpcd_entries:
if hpcentryexists:
waagent.Log("Clean the HPC related host entries from hosts file")
waagent.Log("Add the following HPCD host entries:\n{0}".format(newhpcd_entries))
if newcontent and newcontent[-1] != '\n':
newcontent += '\n'
newcontent += newhpcd_entries
waagent.ReplaceFileContentsAtomic(hostsfile,newcontent)
os.chmod(hostsfile, 0o644)
except :
raise
def daemon(hutil):
publicConfig = hutil.get_public_settings()
privateConfig = hutil.get_protected_settings()
config = aem.EnhancedMonitorConfig(publicConfig, privateConfig)
monitor = aem.EnhancedMonitor(config)
hutil.set_verbose_log(config.isVerbose())
InitExtensionEventLog(hutil.get_name())
while True:
waagent.Log("Collecting performance counter.")
startTime = time.time()
try:
monitor.run()
message = ("deploymentId={0} roleInstance={1} OK"
"").format(config.getVmDeploymentId(),
config.getVmRoleInstance())
hutil.do_status_report("Enable", "success", 0, message)
except Exception as e:
waagent.Error("{0} {1}".format(printable(e),
traceback.format_exc()))
hutil.do_status_report("Enable", "error", 0, "{0}".format(e))
waagent.Log("Finished collection.")
timeElapsed = time.time() - startTime
timeToWait = (aem.MonitoringInterval - timeElapsed)
#Make sure timeToWait is in the range [0, aem.MonitoringInterval)
timeToWait = timeToWait % aem.MonitoringInterval
time.sleep(timeToWait)
def _check_and_install_package(pkg, cmd = None):
if not cmd:
cmd = pkg
if waagent.Run("command -v {0}".format(cmd), chk_err=False) == 0:
waagent.Log("{0} was already installed".format(pkg))
else:
waagent.Log("Start to install {0}".format(pkg))
install_package(pkg)
waagent.Log("{0} was successfully installed".format(pkg))
def _is_nodemanager_daemon(pid):
retcode, output = waagent.RunGetOutput("ps -p {0} -o cmd=".format(pid))
if retcode == 0:
waagent.Log("The cmd for process {0} is {1}".format(pid, output))
pattern = r'(.*[/\s])?{0}\s+[-/]*daemon$'.format(os.path.basename(__file__))
if re.match(pattern, output):
return True
waagent.Log("The process {0} is not HPC Linux node manager daemon".format(pid))
return False
def _install_pip():
ec = waagent.Run("curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py --tlsv1.2", chk_err=False)
if ec != 0:
waagent.Log("get-pip.py was not downloaded, {0}".format(ec))
raise Exception("failed to install package python-pip:{0}".format(ec))
waagent.Log("get-pip.py was downloaded")
ec = waagent.Run("python get-pip.py", chk_err=False)
if ec != 0:
waagent.Log("get-pip.py run failed, {0}".format(ec))
raise Exception("failed to install package python-pip:{0}".format(ec))
waagent.Log("python-pip was installed")
def write(self, counters, maxRetry = 3, eventFile=EventFile):
for i in range(0, maxRetry):
try:
self._write(counters, eventFile)
waagent.Log(("Write {0} counters to event file."
"").format(len(counters)))
return
except IOError, e:
waagent.Warn(("Write to perf counters file failed: {0}"
"").format(printable(e)))
waagent.Log("Retry: {0}".format(i))
time.sleep(1)
def _mount_cgroup():
if not os.path.isdir('/cgroup'):
os.mkdir('/cgroup')
if not os.listdir('/cgroup'):
retcode, mount_msg = waagent.RunGetOutput('mount -t cgroup cgroup /cgroup')
waagent.Log("mount /cgroup directory {0}:{1}".format(retcode, mount_msg))
if retcode == 0:
waagent.Log("/cgroup directory is successfully mounted.")
else:
raise Exception("failed to mount /cgroup directory")
else:
waagent.Log("/cgroup directory was already mounted.")
def _install_cgroup_tool():
if waagent.Run("command -v cgexec", chk_err=False) == 0:
waagent.Log("cgroup tools was already installed")
else:
waagent.Log("Start to install cgroup tools")
if DistroName == "ubuntu":
cg_pkgname = 'cgroup-bin'
elif (DistroName == "centos" or DistroName == "redhat") and re.match("^6", DistroVersion):
cg_pkgname = 'libcgroup'
else:
cg_pkgname = 'libcgroup-tools'
install_package(cg_pkgname)
waagent.Log("cgroup tool was successfully installed")
def _add_dns_search(domain_fqdn):
need_update = False
new_content = ''
for line in (open('/etc/resolv.conf','r')).readlines():
if re.match('^search.* {0}'.format(domain_fqdn), line):
waagent.Log('{0} was already added in /etc/resolv.conf'.format(domain_fqdn))
return
if re.match('^search', line):
need_update = True
new_content += line.replace('search', 'search {0}'.format(domain_fqdn))
else:
new_content += line
if need_update:
waagent.Log('Adding {0} to /etc/resolv.conf'.format(domain_fqdn))
waagent.SetFileContents('/etc/resolv.conf', new_content)
def setUp(self):
print(
'\n\n============================================================================================'
)
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error)
hutil.do_parse_context('TEST')
global MyPatching
MyPatching = FakePatching(hutil)
if MyPatching is None:
sys.exit(1)
distro = DistInfo()[0]
if 'centos' in distro or 'Oracle' in distro or 'redhat' in distro:
MyPatching.cron_restart_cmd = 'service crond restart'
try:
os.remove('mrseq')
except:
pass
waagent.SetFileContents(MyPatching.package_downloaded_path, '')
waagent.SetFileContents(MyPatching.package_patched_path, '')
def main():
#Global Variables definition
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
hutil = None
try:
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
dummy_command("Disable", "success", "Disable succeeded")
elif re.match("^([-/]*)(uninstall)", a):
dummy_command("Uninstall", "success", "Uninstall succeeded")
elif re.match("^([-/]*)(install)", a):
dummy_command("Install", "success", "Install succeeded")
elif re.match("^([-/]*)(enable)", a):
hutil = parse_context("Enable")
enable(hutil)
elif re.match("^([-/]*)(daemon)", a):
hutil = parse_context("Executing")
daemon(hutil)
elif re.match("^([-/]*)(update)", a):
dummy_command("Update", "success", "Update succeeded")
except Exception as e:
err_msg = "Failed with error: {0}, {1}".format(e,
traceback.format_exc())
waagent.Error(err_msg)
if hutil is not None:
hutil.error(err_msg)
hutil.do_exit(1, 'Enable', 'failed', '0',
'Enable failed: {0}'.format(err_msg))
def main():
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." %(ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error,
ExtensionShortName)
global MyPatching
MyPatching = GetMyPatching(hutil)
if MyPatching is None:
sys.exit(1)
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
elif re.match("^([-/]*)(download)", a):
download()
elif re.match("^([-/]*)(patch)", a):
patch()
elif re.match("^([-/]*)(oneoff)", a):
oneoff()
def main():
if len(sys.argv) == 1:
unittest.main()
return
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error)
hutil.do_parse_context('TEST')
global MyPatching
MyPatching = FakePatching(hutil)
if MyPatching == None:
sys.exit(1)
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
elif re.match("^([-/]*)(download)", a):
download()
elif re.match("^([-/]*)(patch)", a):
patch()
elif re.match("^([-/]*)(oneoff)", a):
oneoff()
def main():
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error, ExtensionShortName)
global myProvision
myProvision = GetMyProvision(hutil)
if myProvision == None:
sys.exit(1)
if os.path.isfile("mrseq"):
os.remove("mrseq")
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
elif re.match("^([-/]*)(provision)", a):
provision()
def main():
waagent.LoggerInit('/var/log/waagent.log','/dev/stdout')
waagent.Log("%s started to handle." %(ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error)
hutil.try_parse_context()
global public_settings
public_settings = hutil.get_public_settings()
if not public_settings:
waagent.AddExtensionEvent(name=ExtensionShortName, op='MainInProgress', isSuccess=True, message="Public settings are NOT provided.")
public_settings = {}
global protected_settings
protected_settings = hutil.get_protected_settings()
if not protected_settings:
waagent.AddExtensionEvent(name=ExtensionShortName, op='MainInProgress', isSuccess=True, message="protected settings are NOT provided.")
protected_settings = {}
global distro_category
distro_category = get_distro_category()
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
def init_globals():
"""Initialize all the globals in a function so that we can catch any exceptions that might be raised."""
global hutil, WorkDir, MDSDFileResourcesDir, MDSDRoleName, MDSDFileResourcesPrefix, MDSDPidFile, MDSDPidPortFile
global EnableSyslog, ExtensionOperationType, MdsdFolder, StartDaemonFilePath, MDSD_LISTEN_PORT, omfileconfig
# global rsyslog_ommodule_for_check, RunGetOutput, MdsdFolder, omi_universal_pkg_name
# global DebianConfig, RedhatConfig, UbuntuConfig1510OrHigher, SUSE11_MDSD_SSL_CERTS_FILE
# global SuseConfig11, SuseConfig12, CentosConfig, All_Dist
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("LinuxAzureDiagnostic started to handle.")
hutil = Util.HandlerUtility(waagent.Log, waagent.Error)
init_distro_specific_actions()
WorkDir = os.getcwd()
MDSDFileResourcesDir = "/var/run/mdsd"
MDSDRoleName = 'lad_mdsd'
MDSDFileResourcesPrefix = os.path.join(MDSDFileResourcesDir, MDSDRoleName)
MDSDPidFile = os.path.join(WorkDir, 'mdsd.pid')
MDSDPidPortFile = MDSDFileResourcesPrefix + '.pidport'
EnableSyslog = True
ExtensionOperationType = None
MdsdFolder = os.path.join(WorkDir, 'bin')
StartDaemonFilePath = os.path.join(os.getcwd(), __file__)
MDSD_LISTEN_PORT = 29131
omfileconfig = os.path.join(WorkDir, 'omfileconfig')
init_public_and_private_settings()
def main():
waagent.LoggerInit('/var/log/waagent.log','/dev/stdout')
waagent.Log("%s started to handle." %(ExtensionShortName))
global hutil
hutil = Util.HandlerUtility(waagent.Log, waagent.Error, ExtensionShortName)
hutil.try_parse_context()
global public_settings
public_settings = hutil.get_public_settings()
if not public_settings:
public_settings = {}
global protected_settings
protected_settings = hutil.get_protected_settings()
if not protected_settings:
protected_settings = {}
global distro_category
distro_category = get_distro_category()
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
def main():
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout', True)
waagent.Log("%s started to handle." % (ExtensionShortName))
try:
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
hutil = parse_context("Disable")
disable(hutil)
elif re.match("^([-/]*)(uninstall)", a):
hutil = parse_context("Uninstall")
uninstall(hutil)
elif re.match("^([-/]*)(install)", a):
hutil = parse_context("Install")
install(hutil)
elif re.match("^([-/]*)(enable)", a):
hutil = parse_context("Enable")
enable(hutil)
elif re.match("^([-/]*)(update)", a):
dummy_command("Update", "success", "Update succeeded")
except Exception as e:
err_msg = ("Failed to enable the extension with error: {0}, "
"{1}").format(e, traceback.format_exc())
waagent.Error(err_msg)
hutil.error(err_msg)
hutil.do_exit(1, 'Enable', 'failed', '0',
'Enable failed: {0}'.format(e))
def main():
waagent.LoggerInit('/var/log/waagent.log','/dev/stdout')
waagent.Log("{0} started to handle.".format(ExtensionShortName))
if not os.path.isdir(aem.LibDir):
os.makedirs(aem.LibDir)
for command in sys.argv[1:]:
if re.match("^([-/]*)(install)", command):
grace_exit("install", "success", "Install succeeded")
if re.match("^([-/]*)(uninstall)", command):
grace_exit("uninstall", "success", "Uninstall succeeded")
if re.match("^([-/]*)(update)", command):
grace_exit("update", "success", "Update succeeded")
try:
if re.match("^([-/]*)(enable)", command):
hutil = parse_context("enable")
enable(hutil)
elif re.match("^([-/]*)(disable)", command):
hutil = parse_context("disable")
disable(hutil)
elif re.match("^([-/]*)(daemon)", command):
hutil = parse_context("enable")
daemon(hutil)
except Exception as e:
hutil.error("{0}, {1}".format(e, traceback.format_exc()))
hutil.do_exit(1, command, 'failed','0',
'{0} failed:{1}'.format(command, e))
def waagent_log_info(message):
"""
Log informational message, being cautious of possibility that waagent may
not be imported
"""
if 'Utils.WAAgentUtil' in sys.modules:
waagent.Log(message)
else:
print('Info: {0}'.format(message))
def getStorageMetrics(account, key, hostBase, table, startKey, endKey):
try:
waagent.Log("Retrieve storage metrics data.")
tableService = TableService(account_name = account,
account_key = key,
host_base = hostBase)
ofilter = ("PartitionKey ge '{0}' and PartitionKey lt '{1}'"
"").format(startKey, endKey)
oselect = ("TotalRequests,TotalIngress,TotalEgress,AverageE2ELatency,"
"AverageServerLatency,RowKey")
metrics = tableService.query_entities(table, ofilter, oselect)
waagent.Log("{0} records returned.".format(len(metrics)))
return metrics
except Exception, e:
waagent.Error(("Failed to retrieve storage metrics data: {0} {1}"
"").format(printable(e), traceback.format_exc()))
AddExtensionEvent(message=FAILED_TO_RETRIEVE_STORAGE_DATA)
return None
def main():
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % ExtensionShortName)
operation = "disable"
status = "success"
msg = "Disabled successfully."
hutil = parse_context(operation)
hutil.log("Start to disable.")
hutil.log(msg)
hutil.do_exit(0, operation, status, '0', msg)
def main():
#Global Variables definition
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
operation = "update"
status = "success"
msg = "Updated successfully."
hutil = parse_context(operation)
hutil.log("Start to update.")
hutil.log(msg)
hutil.do_exit(0, operation, status, '0', msg)
def disable():
waagent.Log("disable() called.")
hutil = parse_context('Disable')
#TODO where to kill the node manager
#Check whether monitor process is running.
#If it does, kill it. Otherwise clear pid file
if os.path.isfile(DaemonPidFilePath):
pid = waagent.GetFileContents(DaemonPidFilePath)
waagent.Log("Discovered daemon pid: {0}".format(pid))
if os.path.isdir(os.path.join("/proc", pid)) and _is_nodemanager_daemon(pid):
waagent.Log(("Stop HPC node manager daemon: {0}").format(pid))
os.killpg(int(pid), 9)
cleanup_host_entries()
hutil.do_exit(0, 'Disable', 'success', '0',
'HPC node manager daemon is disabled')
waagent.Log("Remove the daemon pid file: {0}".format(DaemonPidFilePath))
os.remove(DaemonPidFilePath)
else:
waagent.Log("No daemon pid file discovered.")
hutil.do_exit(0, 'Disable', 'success', '0',
'HPC node manager daemon disabled')
def _update_dns_record(domain_fqdn):
s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
while True:
try:
s.connect((domain_fqdn, 53))
break
except Exception as e:
waagent.Log('Failed to connect to {0}:53: {1}'.format(domain_fqdn, e))
ipaddr = s.getsockname()[0]
host_fqdn = "{0}.{1}".format(socket.gethostname().split('.')[0], domain_fqdn)
dns_cmd = 'echo -e "server {0}\nzone {0}\nupdate delete {1}\nupdate add {1} 864000 A {2}\nsend\n" | nsupdate -v'.format(domain_fqdn, host_fqdn, ipaddr)
waagent.Log('The command to update ip to dns server is: {0}'.format(dns_cmd))
retry = 0
while retry < 60:
dns_ret, dns_msg = waagent.RunGetOutput(dns_cmd)
if dns_ret == 0:
waagent.Log("Succeeded to update ip to dns server.")
return
else:
retry = retry + 1
waagent.Log("Failed to update ip to dns server: {0}, {1}".format(dns_ret, dns_msg))
time.sleep(10)
def main():
waagent.LoggerInit('/var/log/waagent.log', '/dev/stdout')
waagent.Log("%s started to handle." % (ExtensionShortName))
for a in sys.argv[1:]:
if re.match("^([-/]*)(disable)", a):
disable()
elif re.match("^([-/]*)(uninstall)", a):
uninstall()
elif re.match("^([-/]*)(install)", a):
install()
elif re.match("^([-/]*)(enable)", a):
enable()
elif re.match("^([-/]*)(update)", a):
update()
def disable(hutil):
pidFile = os.path.join(aem.LibDir, "pid");
#Check whether monitor process is running.
#If it does, kill it. Otherwise clear pid file
if os.path.isfile(pidFile):
pid = waagent.GetFileContents(pidFile)
if os.path.isdir(os.path.join("/proc", pid)):
waagent.Log(("Stop daemon: {0}").format(pid))
os.kill(int(pid), 9)
os.remove(pidFile)
hutil.do_exit(0, 'Disable', 'success', '0',
'Azure Enhanced Monitor is disabled')
os.remove(pidFile)
hutil.do_exit(0, 'Disable', 'success', '0',
'Azure Enhanced Monitor is not running')
