def get_repo_base_url(cls):
"""
:return: Get the location of the base url for the repo. If it runs on an OS that is not supported or on windows this will return None.
"""
if not cls._repo_base_url:
repo_file = cls.get_repo_file()
if repo_file:
# QE-16037: Make sure file is readable.
Machine.chmod("755", repo_file, False, Machine.getAdminUser(),
None, Machine.getAdminPasswd())
file = open(repo_file, 'r')
content = file.read()
file.close()
if Machine.isCentOs() or Machine.isSuse():
m = re.search('^baseurl=(.*HDF.*)', content, re.MULTILINE)
if m and m.group(1):
cls._repo_base_url = m.group(1)
elif Machine.isDebian() or Machine.isUbuntu():
m = re.search('^deb (.*) HDF main', content, re.MULTILINE)
if m and m.group(1):
cls._repo_base_url = m.group(1)
return cls._repo_base_url
def collect_application_logs_for_failed_aborted_tests(cls):
"""
Collect application logs for failed and aborted tests
1) Read test_case_status.log to collected failed and aborted tests
2) Gather Jobsummary log files for all RMs
3) List down the applicationIds for failed and aborted tests
4) Gather logs
:return:
"""
curr_component = util.get_TESTSUITE_COMPONENT()
logger.info(curr_component)
m = re.search(FALCON_STR, curr_component)
isFalcon = bool(m)
LocalJobSummaryLogs = cls.get_local_job_summary_logs(curr_component)
logger.info(LocalJobSummaryLogs)
testcase_map = cls.get_failed_aborted_test_with_timestamp()
logger.info(testcase_map)
if not Machine.pathExists(None, None, cls.LOCAL_TMP_APP_STORAGE, None):
Machine.makedirs(Machine.getAdminUser(),
None, cls.LOCAL_TMP_APP_STORAGE,
Machine.getAdminPasswd())
Machine.chmod("777", cls.LOCAL_TMP_APP_STORAGE, False,
Machine.getAdminUser(), None,
Machine.getAdminPasswd(), True)
for testcase in testcase_map:
testcase_data = testcase_map[testcase]
cls.gather_application_log_from_JobSummary(
LocalJobSummaryLogs,
testcase_data["start_time"],
testcase_data["end_time"],
isFalcon=isFalcon)
def background_job_setup(cls, runSmokeTestSetup=True, config=None):
'''
Upload Data to HDFS before Upgrade starts
Creates /user/hrt_qa/test_rollingupgrade dir on HDFS
Upload 20 files to /user/hrt_qa/test_rollingupgrade
'''
if not cls._base_hdfs_dir:
cls._base_hdfs_dir = '/user/%s/test_rollingupgrade' % Config.get(
'hadoop', 'HADOOPQA_USER')
exit_code, stdout = HDFS.createDirectory(cls._base_hdfs_dir,
force=True)
ruAssert("HDFS", exit_code == 0,
'[BGJobSetup] could not create dir on hdfs.')
LOCAL_WORK_DIR = os.path.join(Config.getEnv('ARTIFACTS_DIR'),
'HDFS_RU_TEST')
localTestWorkDir1 = os.path.join(LOCAL_WORK_DIR, "Temp_data")
HadoopJobHelper.runCustomWordWriter(LOCAL_WORK_DIR, localTestWorkDir1,
20, 40, 1000)
HDFS.copyFromLocal(os.path.join(localTestWorkDir1, "*"),
cls._base_hdfs_dir)
# set up for loadGenerator
cls._lgTestDataDir = cls._base_hdfs_dir + '/testData'
cls._lgTestOutputDir = cls._base_hdfs_dir + '/lg_job'
cls._lgStructureDir = Machine.getTempDir() + "/structure"
# test dir setup
HDFS.deleteDirectory(cls._lgTestDataDir)
HDFS.deleteDirectory(cls._lgTestOutputDir)
command = "rm -rf " + cls._lgStructureDir
exit_code, stdout = Machine.runas(Machine.getAdminUser(), command,
None, None, None, "True",
Machine.getAdminPasswd())
command = "mkdir " + cls._lgStructureDir
stdout = Machine.runas(None, command, None, None, None, "True", None)
Machine.chmod("777", cls._lgStructureDir, "True",
Machine.getAdminUser(), None, Machine.getAdminPasswd())
HADOOP_TEST_JAR = cls.get_hadoop_test_jar()
TEST_USER = Config.get('hadoop', 'HADOOPQA_USER')
# structure generator
jobCmd = 'jar %s NNstructureGenerator -maxDepth 5 -minWidth 2 -maxWidth 5 -numOfFiles 100 -avgFileSize 3 -outDir %s' % (
HADOOP_TEST_JAR, cls._lgStructureDir)
exit_code, stdout = Hadoop.run(jobCmd)
ruAssert("HDFS", exit_code == 0,
"[BGJobSetup] StructureGenerator failed")
# data generator
jobCmd = 'jar %s NNdataGenerator -inDir %s -root %s' % (
HADOOP_TEST_JAR, cls._lgStructureDir, cls._lgTestDataDir)
exit_code, stdout = Hadoop.run(jobCmd)
ruAssert("HDFS", exit_code == 0, "[BGJobSetup] DataGenerator failed")
if runSmokeTestSetup:
logger.info("**** Running HDFS Smoke Test Setup ****")
cls.smoke_test_setup()
def get_local_job_summary_logs(cls, component):
"""
Copy Job_summary Logs to local dirs [artifacts/job_summary_local.log]
parameter: component : Component name for which log collection is taking place
return: List of Local copies of Job summary log
Note: Some components need special handling where there are multiple Job Summary Log files
such as HA and Falcon
"""
LocalJobSummaryLogs = []
try:
if component == FALCON_STR:
from beaver.component.falcon import Falcon # pylint: disable=redefined-outer-name
host1 = Falcon.get_cluster_1_masters()['rm']
host2 = Falcon.get_cluster_2_masters()['rm']
host3 = Falcon.get_cluster_3_masters()['rm']
for host in [host1, host2, host3]:
JobSummaryLog = ComponentLogUtil.MAPRED_getJobSummaryLogFile(
host)
LocalJobSummaryLog = os.path.join(
Config.getEnv('ARTIFACTS_DIR'),
"jobsummary_" + host + ".log")
Machine.copyToLocal(None, host, JobSummaryLog,
LocalJobSummaryLog, None)
if Machine.pathExists(None,
None,
LocalJobSummaryLog,
passwd=None):
LocalJobSummaryLogs.append(LocalJobSummaryLog)
else:
for host in ComponentLogUtil.YARN_getRMHANodes():
JobSummaryLog = ComponentLogUtil.MAPRED_getJobSummaryLogFile(
host)
LocalJobSummaryLog = os.path.join(
Config.getEnv('ARTIFACTS_DIR'),
"jobsummary_" + host + ".log")
Machine.copyToLocal(Machine.getAdminUser(), host,
JobSummaryLog, LocalJobSummaryLog,
Machine.getAdminPasswd())
Machine.chmod("777",
LocalJobSummaryLog,
user=Machine.getAdminUser(),
passwd=Machine.getAdminPasswd())
if Machine.pathExists(Machine.getAdminUser(),
None,
LocalJobSummaryLog,
passwd=Machine.getAdminPasswd()):
LocalJobSummaryLogs.append(LocalJobSummaryLog)
return LocalJobSummaryLogs
except Exception as e:
logger.info("Exception occurs at job_summary_log collection %s", e)
tb = traceback.format_exc()
logger.info(tb)
return LocalJobSummaryLogs
def create_Logs_directories( # pylint: disable=redefined-builtin,unused-argument
cls,
logHost,
logHostUser,
logHostBasePath,
passwd=None,
logoutput=False,
type="all"):
'''
Create log directories for log collection.
type : Choose the type to create directory
if type = "service-logs" , <logHostBasepath>/service-logs dir will be created on logHost
if type = "app-logs", <logHostBasepath>/app-logs dir will be created on logHost
if type == "test-logs", <logHostBasepath>/test-logs dir will be created on logHost
if type == "artifacts", <logHostBasepath>/artifacts dir will be created on logHost
if type == "test-logs", <logHostBasepath>/test-logs dir will be created on logHost
if type == "jenkins-logs", <logHostBasepath>/jenkins-logs dir will be created on logHost
if type == all, all of above directories will be created.
Returns None.
'''
cls.get_log_aggregation_Dir_Locations()
# create temp dir in gateway for app logs
if not Machine.pathExists(None, logHost, cls.LOCAL_TMP_CLUSTER_DIR,
None):
Machine.makedirs(Machine.getAdminUser(),
None, cls.LOCAL_TMP_CLUSTER_DIR,
Machine.getAdminPasswd())
Machine.chmod("777", cls.LOCAL_TMP_CLUSTER_DIR, False,
Machine.getAdminUser(), None,
Machine.getAdminPasswd(), True)
if not Machine.pathExists(None, None, cls.LOCAL_TMP_APP_STORAGE, None):
Machine.makedirs(Machine.getAdminUser(),
None, cls.LOCAL_TMP_APP_STORAGE,
Machine.getAdminPasswd())
Machine.chmod("777", cls.LOCAL_TMP_APP_STORAGE, False,
Machine.getAdminUser(), None,
Machine.getAdminPasswd(), True)
cls.createDirInternal(logHost,
logHostUser,
logHostBasePath,
passwd,
logoutput=logoutput)
# create base dirs in log server
cls.createDirInternal(logHost,
logHostUser,
logHostBasePath + "/" + "artifacts",
passwd,
logoutput=logoutput)
def getConfigValue(cls, propertyValue, defaultValue=None):
_localHiveConf = os.path.join(Config.getEnv('ARTIFACTS_DIR'), cls._localConfDir)
if not os.path.exists(_localHiveConf):
admin_user = Machine.getAdminUser()
Machine.copyToLocal(admin_user, cls.getHiveHost(), cls._configDir, _localHiveConf)
Machine.chmod('777', _localHiveConf, recursive=True, user=admin_user)
if cls._modifiedConfigs.has_key(propertyValue):
return cls._modifiedConfigs[propertyValue]
value = util.getPropertyValueFromConfigXMLFile(
os.path.join(_localHiveConf, 'hiveserver2-site.xml'), propertyValue
)
if value is None or value == '':
value = util.getPropertyValueFromConfigXMLFile(
os.path.join(_localHiveConf, 'hive-site.xml'), propertyValue, defaultValue=defaultValue
)
return value
def createLocalCSVDataFile(cls, numMB, filePath, fileName):
"""
File will be created in test_IndexTable folder
Max size = 4GB
"""
cities = ['San Jose', 'San Francisco', 'Santa Clara']
if numMB > 1024 * 4:
logger.info(
"--- Creating CSV data file of 4096 MB (max limit for this function), "
"instead of requested %s MB file ---", numMB)
numMB = 1024 * 4
# if the file or the path does not exist, we create it
if not Machine.pathExists(Machine.getAdminUser(), Machine.getfqdn(),
filePath, Machine.getAdminPasswd()):
Machine.makedirs(Machine.getAdminUser(), Machine.getfqdn(),
filePath, Machine.getAdminPasswd())
if not Machine.pathExists(
Machine.getAdminUser(), Machine.getfqdn(),
os.path.join(os.getcwd(), filePath, fileName),
Machine.getAdminPasswd()):
Machine.create_file(numMB, os.path.join(os.getcwd(), filePath),
fileName)
Machine.chmod('777',
os.path.join(filePath, fileName),
user=Machine.getAdminUser(),
host=Machine.getfqdn(),
passwd=Machine.getAdminPasswd())
# We insert the data.
insertFile = open(os.path.join(os.getcwd(), filePath, fileName), 'w')
for j in range(numMB):
query = ''
for i in range(1, 11000):
# 1MB more or less equals 10999 rows of these records.
randInt = str(i + (j * 10999))
firstName = 'first_name%s' % randInt
secondName = 'last_name%s' % randInt
city = random.choice(cities)
query += "%s,%s,%s,%s\n" % (str(randInt), firstName,
secondName, city)
insertFile.write(query)
insertFile.close()
def modifyConfig(cls,
changes,
nodeSelection,
isFirstUpdate=True,
host=None,
tmpConf=None):
nodes = HBase.getSelectedNodes(nodeSelection, host)
phoenix_conf = os.path.join(Config.get('phoenix', 'PHOENIX_HOME'),
'bin')
if tmpConf is None:
tmp_conf = os.path.join(Machine.getTempDir(), 'phoenixConf')
else:
tmp_conf = tmpConf
# We change the permissions for phoenix-home/bin
Machine.chmod('777',
tmp_conf,
user=Machine.getAdminUser(),
host=host,
passwd=Machine.getAdminPasswd())
configUtils.modifyConfig(changes, phoenix_conf, tmp_conf, nodes,
isFirstUpdate)
def run(self, randomwriter_bytes="10", local_dir_name="small_rw_jobs"): # pylint: disable=unused-argument
local_dir = os.path.join(Config.getEnv('ARTIFACTS_DIR'),
self.local_dir_name)
if not Machine.pathExists(None, None, local_dir, passwd=None):
Machine.makedirs(None, None, local_dir, None)
Machine.chmod("777", local_dir)
while self.signal:
input_dir = "rw_%d" % int(999999 * random.random())
HDFS.deleteDirectory(input_dir)
HadoopJobHelper.runRandomTextWriterJob(input_dir,
self.randomwriter_bytes,
bytesPerMap=1,
mapsPerHost=1,
jobArg="",
user=None,
config=None,
runInBackground=False,
redirect_file=os.path.join(
local_dir, input_dir))
def collect_application_log_for_Falcon_locally(cls, JobSummaryLogfile,
appId, user):
'''
Collects application logs for Falcon component and save it in Local Dir with <appId>.log filename
'''
host = re.search("jobsummary_(.*).log", JobSummaryLogfile).group(1)
if not Machine.pathExists(
None, None, os.path.join(cls.LOCAL_TMP_APP_STORAGE, host),
None):
Machine.makedirs(None, None,
os.path.join(cls.LOCAL_TMP_APP_STORAGE, host),
None)
Machine.chmod("777",
os.path.join(cls.LOCAL_TMP_APP_STORAGE, host),
recursive=True,
user=None,
host=None,
passwd=None,
logoutput=True)
filename = os.path.join(cls.LOCAL_TMP_APP_STORAGE, host,
appId + ".log")
try:
from beaver.component.falcon import Falcon # pylint: disable=redefined-outer-name
Falcon.get_application_log(host,
appId,
appOwner=user,
nodeAddress=None,
containerId=None,
logoutput=False,
grepFilter=None,
pipeToFileOutput=filename)
except Exception:
logger.error(
"Exception occured during collect_application_log_for_Falcon_locally() call"
)
logger.error(traceback.format_exc())
logger.info("Get application log for Falcon is broken")
def modifyConfigRemote(changes, OriginalConfDir, ConfDir, nodes, id=None): # pylint: disable=redefined-builtin
'''
Modifies hadoop config or config with similar structure.
Returns None.
Linux:
1. Create tmpModifyConfDir_<time> in artifacts dir based on source config directory in gateway
2. Modify contents in created directory.
3. Copy the directory to /tmp/hadoopConf in target machines
'''
_backuploc = getBackupConfigLocation(id=id)
tmpConfDir = os.path.join(
ARTIFACTS_DIR,
'tmpModifyConfDir_' + str(int(round(time.time() * 1000))))
Config.set(PYTHON_CONFIG_NAME,
TMP_CONF_DIR_VAR,
tmpConfDir,
overwrite=True)
tmpConfDir = Config.get(PYTHON_CONFIG_NAME, TMP_CONF_DIR_VAR)
for node in nodes:
Machine.rm(Machine.getAdminUser(), node, ConfDir, isdir=True)
Machine.rm(Machine.getAdminUser(),
Machine.getfqdn(),
tmpConfDir,
isdir=True)
logger.info("*** COPY ORIGINAL CONFIGS FROM REMOTE TO LOCAL ***")
Machine.copyToLocal(None, node, OriginalConfDir, tmpConfDir)
#if node == Machine.getfqdn():
# Machine.copy(OriginalConfDir,tmpConfDir)
for filename, values in changes.items():
filepath = os.path.join(tmpConfDir, filename)
if os.path.isfile(filepath):
logger.info("Modifying file locally: %s", filepath)
_fname, fext = os.path.splitext(filepath)
if fext == ".xml":
util.writePropertiesToConfigXMLFile(
filepath, filepath, values)
elif fext == ".json":
util.writePropertiesToConfigJSONFile(
filepath, filepath, values, ["global"],
"site.hbase-site.")
elif fext == ".properties":
util.writePropertiesToFile(filepath, filepath, values)
elif fext == ".cfg":
util.writePropertiesToFile(filepath, filepath, values)
elif fext == ".conf":
util.writePropertiesToConfFile(filepath, filepath, values)
elif fext == ".sh":
text = ""
for value in values:
text += "\n" + value
util.writeToFile(text, filepath, isAppend=True)
elif fext == ".yaml":
text = ""
for k, v in values.iteritems():
text += k + " : " + v
util.writeToFile(text, filepath, isAppend=True)
elif fext == ".cmd":
text = ""
for value in values:
text += "\n" + value
util.writeToFile(text, filepath, isAppend=True)
elif fext is None or fext == "" or fext == ".include":
text = ""
isFirst = True
for value in values:
if isFirst:
text += value
else:
text += "\n" + value
isFirst = False
util.writeToFile(text, filepath, isAppend=True)
logger.info("****** Copy back the configs to remote ******")
#if node!=Machine.getfqdn():
Machine.copyFromLocal(None, node, tmpConfDir, ConfDir)
Machine.chmod('777', ConfDir, recursive=True, host=node)
def modifyTicketLifetime(cls, lifetime, isFirstUpdate=True):
'''
Restart KDC with new ticket_lifetime
'''
#Extending lifetime beyond 24 hours is currently supported for Redhat linux and Debian only
logger.info("Modifying Keberos ticket lifetime")
if Machine.isDebian() or Machine.isHumboldt():
kdc_conf = '/etc/krb5kdc/kdc.conf'
Machine.chmod(perm="777",
filepath="/etc/krb5kdc",
recursive=True,
user=Machine.getAdminUser(),
passwd=Machine.getAdminPasswd())
with open(kdc_conf, "r+") as f:
text = f.read()
text = (re.sub(r"max_life = \S*", "max_life = %s" % lifetime,
text))
f.seek(0)
f.write(text)
f.truncate()
f.close()
if Machine.isRedHat() or Machine.isCentOs():
kdc_conf = '/var/kerberos/krb5kdc/kdc.conf'
Machine.chmod(perm="777",
filepath=kdc_conf,
user=Machine.getAdminUser(),
passwd=Machine.getAdminPasswd())
with open(kdc_conf, "r+") as f:
text = f.read()
if isFirstUpdate:
text = (re.sub(
"EXAMPLE.COM = {",
"EXAMPLE.COM = {\n max_life = %s" % lifetime, text))
else:
text = (re.sub(
r"EXAMPLE.COM = {\n max_life = \S*",
"EXAMPLE.COM = {\n max_life = %s" % lifetime, text))
f.seek(0)
f.write(text)
f.truncate()
f.close()
kerberos_conf = '/etc/krb5.conf'
Machine.chmod(perm="777",
filepath=kerberos_conf,
user=Machine.getAdminUser(),
passwd=Machine.getAdminPasswd())
with open(kerberos_conf, "r+") as f:
text = f.read()
text = (re.sub(r"ticket_lifetime = \S*",
"ticket_lifetime = %s" % lifetime, text))
f.seek(0)
f.write(text)
f.truncate()
f.close()
kadmin_local_loc = Machine.which("kadmin.local", "root")
logger.info("kadmin_local_loc: %s", kadmin_local_loc)
cmd = "echo 'modify_principal -maxlife %s %s@%s' | %s" % (
lifetime, Config.getEnv('USER'),
Config.get('machine', 'USER_REALM', ''), kadmin_local_loc)
exit_code, stdout = Machine.runas(Machine.getAdminUser(),
cmd,
passwd=Machine.getAdminPasswd())
if exit_code != 0:
logger.info(
"modify_principal failed with the following output:\n%s",
stdout)
cmd = "echo 'modify_principal -maxlife %s krbtgt/[email protected]' | %s" % (
lifetime, kadmin_local_loc)
exit_code, stdout = Machine.runas(Machine.getAdminUser(),
cmd,
passwd=Machine.getAdminPasswd())
if exit_code != 0:
logger.info(
"modify_principal failed with the following output:\n%s",
stdout)
cmd = "service krb5kdc restart"
exit_code, stdout = Machine.runas(Machine.getAdminUser(),
cmd,
passwd=Machine.getAdminPasswd())
if exit_code != 0:
logger.info("KDC failed to restart with the following output:\n%s",
stdout)
