def _collect(self):
"""
Low level function to collect alert data. The result is a tuple as:
res[0] = the result code
res[1] = the list of arguments supplied to the reporting text for the result code
"""
if self.metric_info is None:
raise Exception("Could not determine result. Specific metric collector is not defined.")
if self.uri_property_keys is None:
raise Exception("Could not determine result. URL(s) were not defined.")
# use the URI lookup keys to get a final URI value to query
alert_uri = self._get_uri_from_structure(self.uri_property_keys)
logger.debug("[Alert][{0}] Calculated metric URI to be {1} (ssl={2})".format(
self.get_name(), alert_uri.uri, str(alert_uri.is_ssl_enabled)))
host = BaseAlert.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
try:
port = int(get_port_from_url(alert_uri.uri))
except:
port = 6188
collect_result = None
value_list = []
if isinstance(self.metric_info, AmsMetric):
raw_data_points, http_code = self._load_metric(alert_uri.is_ssl_enabled, host, port, self.metric_info)
if not raw_data_points and http_code not in [200, 307]:
collect_result = self.RESULT_UNKNOWN
value_list.append('HTTP {0} response (metrics unavailable)'.format(str(http_code)))
elif not raw_data_points and http_code in [200, 307]:
raise Exception("[Alert][{0}] Unable to extract JSON from HTTP response".format(self.get_name()))
else:
data_points = self.metric_info.calculate_value(raw_data_points)
compute_result = self.metric_info.calculate_compute(data_points)
value_list.append(compute_result)
collect_result = self._get_result(value_list[0] if compute_result is None else compute_result)
logger.debug("[Alert][{0}] Computed result = {1}".format(self.get_name(), str(value_list)))
return (collect_result, value_list)
def _collect(self):
if self.metric_info is None:
raise Exception("Could not determine result. Specific metric collector is not defined.")
if self.uri_property_keys is None:
raise Exception("Could not determine result. URL(s) were not defined.")
# use the URI lookup keys to get a final URI value to query
alert_uri = self._get_uri_from_structure(self.uri_property_keys)
logger.debug("[Alert][{0}] Calculated metric URI to be {1} (ssl={2})".format(
self.get_name(), alert_uri.uri, str(alert_uri.is_ssl_enabled)))
host = BaseAlert.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
port = 80 # probably not very realistic
try:
port = int(get_port_from_url(alert_uri.uri))
except:
pass
collect_result = None
value_list = []
if isinstance(self.metric_info, JmxMetric):
jmx_property_values, http_code = self._load_jmx(alert_uri.is_ssl_enabled, host, port, self.metric_info)
if not jmx_property_values and http_code in [200, 307]:
collect_result = self.RESULT_UNKNOWN
value_list.append('HTTP {0} response (metrics unavailable)'.format(str(http_code)))
elif not jmx_property_values and http_code not in [200, 307]:
raise Exception("[Alert][{0}] Unable to extract JSON from JMX response".format(self.get_name()))
else:
value_list.extend(jmx_property_values)
check_value = self.metric_info.calculate(value_list)
value_list.append(check_value)
collect_result = self._get_result(value_list[0] if check_value is None else check_value)
logger.debug("[Alert][{0}] Resolved values = {1}".format(self.get_name(), str(value_list)))
return (collect_result, value_list)
def _build_web_query(self, alert_uri):
"""
Builds a URL out of the URI structure. If the URI is already a URL of
the form http[s]:// then this will return the URI as the URL; otherwise,
it will build the URL from the URI structure's elements
"""
# shortcut if the supplied URI starts with the information needed
string_uri = str(alert_uri.uri)
if string_uri.startswith('http://') or string_uri.startswith('https://'):
return alert_uri.uri
uri_path = None
if string_uri and string_uri != str(None):
uri_path = get_path_from_url(string_uri)
# start building the URL manually
host = BaseAlert.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
# maybe slightly realistic
port = 80
if alert_uri.is_ssl_enabled is True:
port = 443
# extract the port
try:
port = int(get_port_from_url(alert_uri.uri))
except:
pass
scheme = 'http'
if alert_uri.is_ssl_enabled is True:
scheme = 'https'
if OSCheck.is_windows_family():
# on windows 0.0.0.0 is invalid address to connect but on linux it resolved to 127.0.0.1
host = resolve_address(host)
if uri_path:
return "{0}://{1}:{2}/{3}".format(scheme, host, str(port), uri_path)
else:
return "{0}://{1}:{2}".format(scheme, host, str(port))
def _collect(self):
if self.metric_info is None:
raise Exception("Could not determine result. Specific metric collector is not defined.")
if self.uri_property_keys is None:
raise Exception("Could not determine result. URL(s) were not defined.")
# use the URI lookup keys to get a final URI value to query
alert_uri = self._get_uri_from_structure(self.uri_property_keys)
logger.debug("[Alert][{0}] Calculated metric URI to be {1} (ssl={2})".format(
self.get_name(), alert_uri.uri, str(alert_uri.is_ssl_enabled)))
host = BaseAlert.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
port = 80 # probably not very realistic
try:
port = int(get_port_from_url(alert_uri.uri))
except:
pass
collect_result = None
value_list = []
if isinstance(self.metric_info, JmxMetric):
value_list.extend(self._load_jmx(alert_uri.is_ssl_enabled, host, port, self.metric_info))
check_value = self.metric_info.calculate(value_list)
value_list.append(check_value)
collect_result = self.__get_result(value_list[0] if check_value is None else check_value)
logger.debug("[Alert][{0}] Resolved values = {1}".format(
self.get_name(), str(value_list)))
return (collect_result, value_list)
def _build_web_query(self, alert_uri):
"""
Builds a URL out of the URI structure. If the URI is already a URL of
the form http[s]:// then this will return the URI as the URL; otherwise,
it will build the URL from the URI structure's elements
"""
# shortcut if the supplied URI starts with the information needed
string_uri = str(alert_uri.uri)
if string_uri.startswith('http://') or string_uri.startswith(
'https://'):
return alert_uri.uri
# start building the URL manually
host = BaseAlert.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
# maybe slightly realistic
port = 80
if alert_uri.is_ssl_enabled is True:
port = 443
# extract the port
try:
port = int(get_port_from_url(alert_uri.uri))
except:
pass
scheme = 'http'
if alert_uri.is_ssl_enabled is True:
scheme = 'https'
if OSCheck.is_windows_family():
# on windows 0.0.0.0 is invalid address to connect but on linux it resolved to 127.0.0.1
host = resolve_address(host)
return "{0}://{1}:{2}".format(scheme, host, str(port))
namenode_address = format('hdfs://{dfs_ha_nameservices}')
namenode_hosts = default("/clusterHostInfo/namenode_host", None)
if type(namenode_hosts) is list:
namenode_host = namenode_hosts[0]
else:
namenode_host = namenode_hosts
has_namenode = not namenode_host == None
namenode_http_port = "50070"
namenode_rpc_port = "8020"
if has_namenode:
if 'dfs.namenode.http-address' in config['configurations']['hdfs-site']:
namenode_http_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.namenode.http-address'])
if dfs_ha_enabled and namenode_rpc:
namenode_rpc_port = get_port_from_url(namenode_rpc)
else:
if 'dfs.namenode.rpc-address' in config['configurations']['hdfs-site']:
namenode_rpc_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.namenode.rpc-address'])
rm_hosts = default("/clusterHostInfo/rm_host", None)
if type(rm_hosts) is list:
rm_host = rm_hosts[0]
else:
rm_host = rm_hosts
has_rm = not rm_host == None
jt_rpc_port = "8050"
rm_port = "8080"
jdbc_libs_dir = format("{hive_lib}/native/lib64")
lib_dir_available = os.path.exists(jdbc_libs_dir)
if sqla_db_used:
jars_path_in_archive = format("{tmp_dir}/sqla-client-jdbc/java/*")
libs_path_in_archive = format("{tmp_dir}/sqla-client-jdbc/native/lib64/*")
downloaded_custom_connector = format("{tmp_dir}/{jdbc_jar_name}")
libs_in_hive_lib = format("{jdbc_libs_dir}/*")
# Start, Common Hosts and Ports
ambari_server_hostname = config['ambariLevelParams']['ambari_server_host']
hive_metastore_hosts = default('/clusterHostInfo/hive_metastore_hosts', [])
hive_metastore_host = hive_metastore_hosts[0] if len(hive_metastore_hosts) > 0 else None
hive_metastore_port = get_port_from_url(config['configurations']['hive-site']['hive.metastore.uris'])
hive_server_hosts = default("/clusterHostInfo/hive_server_hosts", [])
hive_server_host = hive_server_hosts[0] if len(hive_server_hosts) > 0 else None
hive_server_interactive_hosts = default('/clusterHostInfo/hive_server_interactive_hosts', [])
hive_server_interactive_host = hive_server_interactive_hosts[0] if len(hive_server_interactive_hosts) > 0 else None
hive_server_interactive_ha = True if len(hive_server_interactive_hosts) > 1 else False
# End, Common Hosts and Ports
hive_transport_mode = config['configurations']['hive-site']['hive.server2.transport.mode']
if hive_transport_mode.lower() == "http":
hive_server_port = config['configurations']['hive-site']['hive.server2.thrift.http.port']
else:
hive_server_port = default('/configurations/hive-site/hive.server2.thrift.port',"10000")
def _collect(self):
"""
Low level function to collect alert data. The result is a tuple as:
res[0] = the result code
res[1] = the list of arguments supplied to the reporting text for the result code
"""
if self.metric_info is None:
raise Exception(
"Could not determine result. Specific metric collector is not defined."
)
if self.uri_property_keys is None:
raise Exception(
"Could not determine result. URL(s) were not defined.")
# use the URI lookup keys to get a final URI value to query
alert_uri = self._get_uri_from_structure(self.uri_property_keys)
if logger.isEnabledFor(logging.DEBUG):
logger.debug(
"[Alert][{0}] Calculated metric URI to be {1} (ssl={2})".
format(self.get_name(), alert_uri.uri,
str(alert_uri.is_ssl_enabled)))
host = inet_utils.get_host_from_url(alert_uri.uri)
if host is None:
host = self.host_name
try:
port = int(get_port_from_url(alert_uri.uri))
except:
port = 6188
collect_result = None
value_list = []
if isinstance(self.metric_info, AmsMetric):
raw_data_points, http_code = self._load_metric(
alert_uri.is_ssl_enabled, host, port, self.metric_info)
if not raw_data_points and http_code not in [200, 307]:
collect_result = self.RESULT_UNKNOWN
value_list.append(
'HTTP {0} response (metrics unavailable)'.format(
str(http_code)))
elif not raw_data_points and http_code in [200, 307]:
raise Exception(
"[Alert][{0}] Unable to extract JSON from HTTP response".
format(self.get_name()))
else:
data_points = self.metric_info.calculate_value(raw_data_points)
compute_result = self.metric_info.calculate_compute(
data_points)
value_list.append(compute_result)
collect_result = self._get_result(
value_list[0] if compute_result is None else compute_result
)
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] Computed result = {1}".format(
self.get_name(), str(value_list)))
return (collect_result, value_list)
def _collect(self):
# can be parameterized or static
# if not parameterized, this will return the static value
uri_value = self._get_configuration_value(self.uri)
if uri_value is None:
uri_value = self.host_name
logger.debug("[Alert][{0}] Setting the URI to this host since it wasn't specified".format(
self.get_name()))
# in some cases, a single property is a comma-separated list like
# host1:8080,host2:8081,host3:8083
uri_value_array = uri_value.split(',')
if len(uri_value_array) > 1:
for item in uri_value_array:
if self.host_name in item:
uri_value = item
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] Extracted {1} as the host name while parsing the CSV URI {2}".format(
self.get_name(), uri_value, str(uri_value_array)))
break
host = BaseAlert.get_host_from_url(uri_value)
if host is None:
host = self.host_name
try:
port = int(get_port_from_url(uri_value))
except:
if self.default_port is None:
label = 'Unable to determine port from URI {0}'.format(uri_value)
return (self.RESULT_UNKNOWN, [label])
port = self.default_port
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] Checking {1} on port {2}".format(
self.get_name(), host, str(port)))
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.settimeout(self.critical_timeout)
if OSCheck.is_windows_family():
# on windows 0.0.0.0 is invalid address to connect but on linux it resolved to 127.0.0.1
host = resolve_address(host)
start_time = time.time()
s.connect((host, port))
end_time = time.time()
milliseconds = end_time - start_time
seconds = milliseconds / 1000.0
# not sure why this happens sometimes, but we don't always get a
# socket exception if the connect() is > than the critical threshold
if seconds >= self.critical_timeout:
return (self.RESULT_CRITICAL, ['Socket Timeout', host, port])
result = self.RESULT_OK
if seconds >= self.warning_timeout:
result = self.RESULT_WARNING
return (result, [seconds, port])
except Exception as e:
return (self.RESULT_CRITICAL, [str(e), host, port])
finally:
if s is not None:
try:
s.close()
except:
# no need to log a close failure
pass
def _collect(self):
# can be parameterized or static
# if not parameterized, this will return the static value
uri_value = self._get_configuration_value(self.uri)
host_not_specified = False
if uri_value is None:
host_not_specified = True
uri_value = self.host_name
logger.debug("[Alert][{0}] Setting the URI to this host since it wasn't specified".format(
self.get_name()))
# in some cases, a single property is a comma-separated list like
# host1:8080,host2:8081,host3:8083
uri_value_array = uri_value.split(',')
if len(uri_value_array) > 1:
for item in uri_value_array:
if self.host_name in item:
uri_value = item
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] Extracted {1} as the host name while parsing the CSV URI {2}".format(
self.get_name(), uri_value, str(uri_value_array)))
break
host = BaseAlert.get_host_from_url(uri_value)
if host is None or host == "localhost" or host == "0.0.0.0":
host = self.host_name
host_not_specified = True
hosts = [host]
# If host is not specified in the uri, hence we are using current host name
# then also add public host name as a fallback.
if host_not_specified and host.lower() == self.host_name.lower() \
and self.host_name.lower() != self.public_host_name.lower():
hosts.append(self.public_host_name)
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] List of hosts = {1}".format(self.get_name(), hosts))
try:
port = int(get_port_from_url(uri_value))
except:
if self.default_port is None:
label = 'Unable to determine port from URI {0}'.format(uri_value)
return (self.RESULT_UNKNOWN, [label])
port = self.default_port
exceptions = []
for host in hosts:
if logger.isEnabledFor(logging.DEBUG):
logger.debug("[Alert][{0}] Checking {1} on port {2}".format(
self.get_name(), host, str(port)))
s = None
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.settimeout(self.critical_timeout)
if OSCheck.is_windows_family():
# on windows 0.0.0.0 is invalid address to connect but on linux it resolved to 127.0.0.1
host = resolve_address(host)
start_time = time.time()
s.connect((host, port))
if self.socket_command is not None:
s.sendall(self.socket_command)
data = s.recv(1024)
if self.socket_command_response is not None and data != self.socket_command_response:
raise Exception("Expected response {0}, Actual response {1}".format(
self.socket_command_response, data))
end_time = time.time()
milliseconds = end_time - start_time
seconds = milliseconds / 1000.0
# not sure why this happens sometimes, but we don't always get a
# socket exception if the connect() is > than the critical threshold
if seconds >= self.critical_timeout:
return (self.RESULT_CRITICAL, ['Socket Timeout', host, port])
result = self.RESULT_OK
if seconds >= self.warning_timeout:
result = self.RESULT_WARNING
return (result, [seconds, port])
except Exception as e:
exceptions.append(e)
finally:
if s is not None:
try:
s.close()
except:
# no need to log a close failure
pass
if exceptions:
return (self.RESULT_CRITICAL, [str(exceptions[0]), hosts[0], port])
namenode_address = format('hdfs://{dfs_ha_nameservices}')
namenode_hosts = default("/clusterHostInfo/namenode_host", None)
if type(namenode_hosts) is list:
namenode_host = namenode_hosts[0]
else:
namenode_host = namenode_hosts
has_namenode = not namenode_host == None
namenode_http_port = "50070"
namenode_rpc_port = "8020"
if has_namenode:
if 'dfs.namenode.http-address' in config['configurations']['hdfs-site']:
namenode_http_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.namenode.http-address'])
if dfs_ha_enabled and namenode_rpc:
namenode_rpc_port = get_port_from_url(namenode_rpc)
else:
if 'dfs.namenode.rpc-address' in config['configurations']['hdfs-site']:
namenode_rpc_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.namenode.rpc-address'])
rm_hosts = default("/clusterHostInfo/rm_host", None)
if type(rm_hosts) is list:
rm_host = rm_hosts[0]
else:
rm_host = rm_hosts
has_rm = not rm_host == None
jt_rpc_port = "8050"
rm_port = "8080"
check_db_connection_jar = format("/usr/lib/ambari-agent/{check_db_connection_jar_name}")
hive_jdbc_drivers_list = [
"com.microsoft.sqlserver.jdbc.SQLServerDriver",
"com.mysql.jdbc.Driver",
"org.postgresql.Driver",
"oracle.jdbc.driver.OracleDriver",
]
downloaded_custom_connector = format("{tmp_dir}/{jdbc_jar_name}")
prepackaged_ojdbc_symlink = format("{hive_lib}/ojdbc6.jar")
templeton_port = config["configurations"]["webhcat-site"]["templeton.port"]
# common
hive_metastore_hosts = config["clusterHostInfo"]["hive_metastore_host"]
hive_metastore_host = hive_metastore_hosts[0]
hive_metastore_port = get_port_from_url(config["configurations"]["hive-site"]["hive.metastore.uris"]) # "9083"
ambari_server_hostname = config["clusterHostInfo"]["ambari_server_host"][0]
hive_server_host = config["clusterHostInfo"]["hive_server_host"][0]
hive_server_hosts = config["clusterHostInfo"]["hive_server_host"]
hive_transport_mode = config["configurations"]["hive-site"]["hive.server2.transport.mode"]
if hive_transport_mode.lower() == "http":
hive_server_port = config["configurations"]["hive-site"]["hive.server2.thrift.http.port"]
else:
hive_server_port = default("/configurations/hive-site/hive.server2.thrift.port", "10000")
hive_url = format("jdbc:hive2://{hive_server_host}:{hive_server_port}")
hive_http_endpoint = default("/confiurations/hive-site/hive.server2.thrift.http.path", "cliservice")
hive_server_principal = config["configurations"]["hive-site"]["hive.server2.authentication.kerberos.principal"]
hive_server2_authentication = config["configurations"]["hive-site"]["hive.server2.authentication"]
# params from nifi-toolkit-env
nifi_toolkit_java_options = config['configurations']['nifi-toolkit-env'][
'nifi_toolkit_java_options'] if 'nifi-toolkit-env' in config[
'configurations'] else '-Xms128m -Xmx256m'
nifi_toolkit_conf_dir = '/etc/nifi-toolkit'
# autodetect jdk home
jdk64_home = config['ambariLevelParams']['java_home']
# autodetect ambari server for metrics
if 'metrics_collector_hosts' in config['clusterHostInfo']:
metrics_collector_host = str(
config['clusterHostInfo']['metrics_collector_hosts'][0])
metrics_collector_port = str(
get_port_from_url(config['configurations']['ams-site']
['timeline.metrics.service.webapp.address']))
else:
metrics_collector_host = ''
metrics_collector_port = ''
# detect zookeeper_quorum
zookeeper_port = default('/configurations/zoo.cfg/clientPort', None)
# get comma separated list of zookeeper hosts from clusterHostInfo
index = 0
zookeeper_quorum = ""
zk_hosts_property = 'zookeeper_hosts' if 'zookeeper_hosts' in config[
'clusterHostInfo'] else "zookeeper_server_hosts"
for host in config['clusterHostInfo'][zk_hosts_property]:
zookeeper_quorum += host + ":" + str(zookeeper_port)
index += 1
if index < len(config['clusterHostInfo'][zk_hosts_property]):
templeton_port = config['configurations']['webhcat-site']['templeton.port']
#constants for type2 jdbc
jdbc_libs_dir = format("{hive_lib}/native/lib64")
lib_dir_available = os.path.exists(jdbc_libs_dir)
if sqla_db_used:
jars_path_in_archive = format("{tmp_dir}/sqla-client-jdbc/java/*")
libs_path_in_archive = format("{tmp_dir}/sqla-client-jdbc/native/lib64/*")
downloaded_custom_connector = format("{tmp_dir}/sqla-client-jdbc.tar.gz")
libs_in_hive_lib = format("{jdbc_libs_dir}/*")
#common
hive_metastore_hosts = config['clusterHostInfo']['hive_metastore_host']
hive_metastore_host = hive_metastore_hosts[0]
hive_metastore_port = get_port_from_url(config['configurations']['hive-site']['hive.metastore.uris']) #"9083"
ambari_server_hostname = config['clusterHostInfo']['ambari_server_host'][0]
hive_server_host = config['clusterHostInfo']['hive_server_host'][0]
hive_server_hosts = config['clusterHostInfo']['hive_server_host']
hive_transport_mode = config['configurations']['hive-site']['hive.server2.transport.mode']
if hive_transport_mode.lower() == "http":
hive_server_port = config['configurations']['hive-site']['hive.server2.thrift.http.port']
else:
hive_server_port = default('/configurations/hive-site/hive.server2.thrift.port',"10000")
hive_url = format("jdbc:hive2://{hive_server_host}:{hive_server_port}")
hive_http_endpoint = default('/confiurations/hive-site/hive.server2.thrift.http.path', "cliservice")
hive_server_principal = config['configurations']['hive-site']['hive.server2.authentication.kerberos.principal']
hive_server2_authentication = config['configurations']['hive-site']['hive.server2.authentication']
