def check_status(self, notebook, snippet):
api = get_spark_api(self.user)
state = api.get_batch_status(snippet['result']['handle']['id'])
return {
'status': state,
}
def create_session(self, lang='scala', properties=None):
if not properties and USE_DEFAULT_CONFIGURATION.get():
user_config = DefaultConfiguration.objects.get_configuration_for_user(
app='spark', user=self.user)
if user_config is not None:
properties = user_config.properties_list
props = self.get_livy_props(lang, properties)
api = get_spark_api(self.user)
response = api.create_session(**props)
status = api.get_session(response['id'])
count = 0
while status['state'] == 'starting' and count < 120:
status = api.get_session(response['id'])
count += 1
time.sleep(1)
if status['state'] != 'idle':
info = '\n'.join(status['log']) if status['log'] else 'timeout'
raise QueryError(
_('The Spark session is %s and could not be created in the cluster: %s'
) % (status['state'], info))
return {
'type': lang,
'id': response['id'],
'properties': self.to_properties(props)
}
def execute(self, notebook, snippet):
api = get_spark_api(self.user)
if snippet['type'] == 'jar':
properties = {
'file': snippet['properties'].get('app_jar'),
'className': snippet['properties'].get('class'),
'args': snippet['properties'].get('arguments'),
}
elif snippet['type'] == 'py':
properties = {
'file': snippet['properties'].get('py_file'),
'args': snippet['properties'].get('argument', []),
}
else:
properties = {
'file': snippet['properties'].get('app_jar'),
'className': snippet['properties'].get('class'),
'args': snippet['properties'].get('arguments'),
'pyFiles': snippet['properties'].get('py_file'),
'files': snippet['properties'].get('files'),
# driverMemory
# driverCores
# executorMemory
# executorCores
# archives
}
response = api.submit_batch(properties)
return {'id': response['id'], 'has_result_set': True, 'properties': []}
def close_statement(self, notebook, snippet):
api = get_spark_api(self.user)
session_id = snippet['result']['handle']['id']
if session_id is not None:
api.close_batch(session_id)
return {'session': session_id, 'status': 0}
else:
return {'status': -1} # skipped
def close_session(self, session):
api = get_spark_api(self.user)
if session['id'] is not None:
try:
api.close(session['id'])
return {'session': session['id'], 'status': 0}
except RestException as e:
if e.code == 404 or e.code == 500: # TODO remove the 500
raise SessionExpired(e)
else:
return {'status': -1}
def check_status(self, notebook, snippet):
api = get_spark_api(self.user)
session = _get_snippet_session(notebook, snippet)
cell = snippet['result']['handle']['id']
try:
response = api.fetch_data(session['id'], cell)
return {
'status': response['state'],
}
except Exception as e:
message = force_unicode(str(e)).lower()
if re.search("session ('\d+' )?not found", message):
raise SessionExpired(e)
else:
raise e
def get_sample_data(self,
snippet,
database=None,
table=None,
column=None,
is_async=False,
operation=None):
if operation != 'hello':
raise NotImplementedError()
response = {}
api = get_spark_api(self.user)
api.get_status()
response['status'] = 0
response['rows'] = []
return response
def execute(self, notebook, snippet):
api = get_spark_api(self.user)
session = _get_snippet_session(notebook, snippet)
try:
response = api.submit_statement(session['id'],
snippet['statement'])
return {
'id': response['id'],
'has_result_set': True,
'sync': False
}
except Exception as e:
message = force_unicode(str(e)).lower()
if re.search(
"session ('\d+' )?not found", message
) or 'connection refused' in message or 'session is in state busy' in message:
raise SessionExpired(e)
else:
raise e
def get_api(self): return get_spark_api(self.user, self.interpreter)
def get_log(self, notebook, snippet, startFrom=0, size=None):
api = get_spark_api(self.user)
session = _get_snippet_session(notebook, snippet)
return api.get_log(session['id'], startFrom=startFrom, size=size)
def cancel(self, notebook, snippet):
api = get_spark_api(self.user)
session = _get_snippet_session(notebook, snippet)
response = api.cancel(session['id'])
return {'status': 0}
def fetch_result(self, notebook, snippet, rows, start_over):
api = get_spark_api(self.user)
session = _get_snippet_session(notebook, snippet)
cell = snippet['result']['handle']['id']
try:
response = api.fetch_data(session['id'], cell)
except Exception as e:
message = force_unicode(str(e)).lower()
if re.search("session ('\d+' )?not found", message):
raise SessionExpired(e)
else:
raise e
content = response['output']
if content['status'] == 'ok':
data = content['data']
images = []
try:
table = data['application/vnd.livy.table.v1+json']
except KeyError:
try:
images = [data['image/png']]
except KeyError:
images = []
if 'application/json' in data:
result = data['application/json']
data = result['data']
meta = [{
'name': field['name'],
'type': field['type'],
'comment': ''
} for field in result['schema']['fields']]
type = 'table'
else:
data = [[data['text/plain']]]
meta = [{
'name': 'Header',
'type': 'STRING_TYPE',
'comment': ''
}]
type = 'text'
else:
data = table['data']
headers = table['headers']
meta = [{
'name': h['name'],
'type': h['type'],
'comment': ''
} for h in headers]
type = 'table'
# Non start_over not supported
if not start_over:
data = []
return {'data': data, 'images': images, 'meta': meta, 'type': type}
elif content['status'] == 'error':
tb = content.get('traceback', None)
if tb is None or not tb:
msg = content.get('ename', 'unknown error')
evalue = content.get('evalue')
if evalue is not None:
msg = '%s: %s' % (msg, evalue)
else:
msg = ''.join(tb)
raise QueryError(msg)
def create_session(self, lang='scala', properties=None):
if not properties:
config = None
if USE_DEFAULT_CONFIGURATION.get():
config = DefaultConfiguration.objects.get_configuration_for_user(
app='spark', user=self.user)
if config is not None:
properties = config.properties_list
else:
properties = self.get_properties()
props = dict([(p['name'], p['value'])
for p in properties]) if properties is not None else {}
# HUE-4761: Hue's session request is causing Livy to fail with "JsonMappingException: Can not deserialize
# instance of scala.collection.immutable.List out of VALUE_STRING token" due to List type values
# not being formed properly, they are quoted csv strings (without brackets) instead of proper List
# types, this is for keys; archives, jars, files and pyFiles. The Mako frontend probably should be
# modified to pass the values as Livy expects but for now we coerce these types to be Lists.
# Issue only occurs when non-default values are used because the default path properly sets the
# empty list '[]' for these four values.
# Note also that Livy has a 90 second timeout for the session request to complete, this needs to
# be increased for requests that take longer, for example when loading large archives.
tmp_archives = props['archives']
if type(tmp_archives) is not list:
props['archives'] = tmp_archives.split(",")
LOG.debug("Check List type: archives was not a list")
tmp_jars = props['jars']
if type(tmp_jars) is not list:
props['jars'] = tmp_jars.split(",")
LOG.debug("Check List type: jars was not a list")
tmp_files = props['files']
if type(tmp_files) is not list:
props['files'] = tmp_files.split(",")
LOG.debug("Check List type: files was not a list")
tmp_py_files = props['pyFiles']
if type(tmp_py_files) is not list:
props['pyFiles'] = tmp_py_files.split(",")
LOG.debug("Check List type: pyFiles was not a list")
# Convert the conf list to a dict for Livy
LOG.debug("Property Spark Conf kvp list from UI is: " +
str(props['conf']))
props['conf'] = {
conf.get('key'): conf.get('value')
for i, conf in enumerate(props['conf'])
}
LOG.debug("Property Spark Conf dictionary is: " + str(props['conf']))
props['kind'] = lang
api = get_spark_api(self.user)
response = api.create_session(**props)
status = api.get_session(response['id'])
count = 0
while status['state'] == 'starting' and count < 120:
status = api.get_session(response['id'])
count += 1
time.sleep(1)
if status['state'] != 'idle':
info = '\n'.join(status['log']) if status['log'] else 'timeout'
raise QueryError(
_('The Spark session could not be created in the cluster: %s')
% info)
return {'type': lang, 'id': response['id'], 'properties': properties}
def get_log(self, notebook, snippet, startFrom=0, size=None):
api = get_spark_api(self.user)
return api.get_batch_log(snippet['result']['handle']['id'],
startFrom=startFrom,
size=size)
