def test_add_sessions_command_parses():
# Do not skip and python
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
connection_string = "-u http://url.com -t {} -a sdf -p w".format(
AUTH_BASIC)
line = " ".join([command, name, language, connection_string])
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://url.com", initialize_auth(args)), False,
{"kind": "pyspark"})
# Skip and scala - upper case
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l scala"
connection_string = "--url http://location:port"
line = " ".join([command, name, language, connection_string, "-k"])
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
args.auth = NO_AUTH
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://location:port", initialize_auth(args)), True,
{"kind": "spark"})
def logs(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.logs, line)
self._assure_cell_body_is_empty(KernelMagics.logs.__name__, cell)
if self.session_started:
out = self.spark_controller.get_logs()
self.ipython_display.write(out)
else:
self.ipython_display.write(u"No logs yet.")
def logs(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.logs, line)
self._assure_cell_body_is_empty(KernelMagics.logs.__name__, cell)
if self.session_started:
out = self.spark_controller.get_logs()
self.ipython_display.write(out)
else:
self.ipython_display.write(u"No logs yet.")
def spark(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.spark, line)
if self._do_not_call_start_session(u""):
(success, out) = self.spark_controller.run_command(Command(cell))
if success:
self.ipython_display.write(out)
else:
self.ipython_display.send_error(out)
else:
return None
def spark(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.spark, line)
if self._do_not_call_start_session(u""):
(success, out) = self.spark_controller.run_command(Command(cell))
if success:
self.ipython_display.write(out)
else:
self.ipython_display.send_error(out)
else:
return None
def info(self, line, cell=u"", local_ns=None):
parse_argstring_or_throw(self.info, line)
self._assure_cell_body_is_empty(KernelMagics.info.__name__, cell)
if self.session_started:
current_session_id = self.spark_controller.get_session_id_for_client(self.session_name)
else:
current_session_id = None
self.ipython_display.html(u"Current session configs: <tt>{}</tt><br>".format(conf.get_session_properties(self.language)))
info_sessions = self.spark_controller.get_all_sessions_endpoint(self.endpoint)
self._print_endpoint_info(info_sessions, current_session_id)
def info(self, line, cell=u"", local_ns=None):
parse_argstring_or_throw(self.info, line)
self._assure_cell_body_is_empty(KernelMagics.info.__name__, cell)
if self.session_started:
current_session_id = self.spark_controller.get_session_id_for_client(self.session_name)
else:
current_session_id = None
self.ipython_display.html(u"Current session configs: <tt>{}</tt><br>".format(conf.get_session_properties(self.language)))
info_sessions = self.spark_controller.get_all_sessions_endpoint(self.endpoint)
self._print_endpoint_info(info_sessions, current_session_id)
def test_parse_argstring_or_throw():
parse_argstring = MagicMock(side_effect=UsageError('OOGABOOGABOOGA'))
try:
parse_argstring_or_throw(MagicMock(), MagicMock(), parse_argstring=parse_argstring)
assert False
except BadUserDataException as e:
assert_equals(str(e), str(parse_argstring.side_effect))
parse_argstring = MagicMock(side_effect=ValueError('AN UNKNOWN ERROR HAPPENED'))
try:
parse_argstring_or_throw(MagicMock(), MagicMock(), parse_argstring=parse_argstring)
assert False
except ValueError as e:
assert_is(e, parse_argstring.side_effect)
def test_parse_argstring_or_throw():
parse_argstring = MagicMock(side_effect=UsageError('OOGABOOGABOOGA'))
try:
parse_argstring_or_throw(MagicMock(), MagicMock(), parse_argstring=parse_argstring)
assert False
except BadUserDataException as e:
assert_equals(str(e), str(parse_argstring.side_effect))
parse_argstring = MagicMock(side_effect=ValueError('AN UNKNOWN ERROR HAPPENED'))
try:
parse_argstring_or_throw(MagicMock(), MagicMock(), parse_argstring=parse_argstring)
assert False
except ValueError as e:
assert_is(e, parse_argstring.side_effect)
def test_cleanup_endpoint_command_parses():
mock_method = MagicMock()
spark_controller.cleanup_endpoint = mock_method
line = "cleanup -u endp"
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
args.auth = NO_AUTH
mock_method.assert_called_once_with(Endpoint("endp",
initialize_auth(args)))
line = "cleanup -u endp -a user -p passw -t {}".format(AUTH_BASIC)
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
mock_method.assert_called_with(Endpoint("endp", initialize_auth(args)))
def sql(self, line, cell="", local_ns=None):
if self._do_not_call_start_session(""):
args = parse_argstring_or_throw(self.sql, line)
return self.execute_sqlquery(cell, args.samplemethod, args.maxrows, args.samplefraction,
None, args.output, args.quiet)
else:
return
def sql(self, line, cell="", local_ns=None):
if self._do_not_call_start_session(""):
args = parse_argstring_or_throw(self.sql, line)
return self.execute_sqlquery(cell, args.samplemethod, args.maxrows, args.samplefraction,
None, args.output, args.quiet)
else:
return
def spark(self, line, cell="", local_ns=None):
if self._do_not_call_start_session(u""):
args = parse_argstring_or_throw(self.spark, line)
self.execute_spark(cell, args.output, args.samplemethod,
args.maxrows, args.samplefraction, None)
else:
return
def delete(self, line, cell="", local_ns=None):
self._assure_cell_body_is_empty(KernelMagics.delete.__name__, cell)
args = parse_argstring_or_throw(self.delete, line)
session = args.session
if args.session is None:
self.ipython_display.send_error(
u'You must provide a session ID (-s argument).')
return
if args.force:
id = self.spark_controller.get_session_id_for_client(
self.session_name)
if session == id:
self.ipython_display.send_error(
u"Cannot delete this kernel's session ({}). Specify a different session,"
u" shutdown the kernel to delete this session, or run %cleanup to "
u"delete all sessions for this endpoint.".format(id))
return
self.spark_controller.delete_session_by_id(self.endpoint, session)
else:
self.ipython_display.send_error(
u"Include the -f parameter if you understand that all statements executed "
u"in this session will be lost.")
def _do_not_call_change_endpoint(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_endpoint,
line)
if self.session_started:
error = u"Cannot change the endpoint if a session has been started."
raise BadUserDataException(error)
auth = initialize_auth(args=args)
self.endpoint = Endpoint(args.url, auth)
def spark(self, line, cell="", local_ns=None):
if not self._do_not_call_start_session(u""):
return
args = parse_argstring_or_throw(self.spark, line)
coerce = get_coerce_value(args.coerce)
self.execute_spark(cell, args.output, args.samplemethod, args.maxrows,
args.samplefraction, None, coerce)
def _do_not_call_change_endpoint(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_endpoint, line)
username = args.username
password = args.password
server = args.server
if self.session_started:
error = u"Cannot change the endpoint if a session has been started."
raise BadUserDataException(error)
self.endpoint = Endpoint(server, username, password)
def _do_not_call_init_livy_session(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_init_livy_session,
line)
language = args.language.lower()
if language not in LANGS_SUPPORTED:
self.ipython_display.send_error(
u"'{}' language not supported in kernel magics.".format(
language))
return
# init livy session
self.init_livy_session(language)
def send_to_spark(self, line, cell=u"", local_ns=None):
self._assure_cell_body_is_empty(KernelMagics.send_to_spark.__name__, cell)
args = parse_argstring_or_throw(self.send_to_spark, line)
if not args.input:
raise BadUserDataException("-i param not provided.")
if self._do_not_call_start_session(""):
self.do_send_to_spark(cell, args.input, args.vartype, args.varname, args.maxrows, None)
else:
return
def _do_not_call_change_endpoint(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_endpoint, line)
username = args.username
password = args.password
server = args.server
if self.session_started:
error = u"Cannot change the endpoint if a session has been started."
raise BadUserDataException(error)
self.endpoint = Endpoint(server, username, password)
def cleanup(self, line, cell="", local_ns=None):
self._assure_cell_body_is_empty(KernelMagics.cleanup.__name__, cell)
args = parse_argstring_or_throw(self.cleanup, line)
if args.force:
self._do_not_call_delete_session(u"")
self.spark_controller.cleanup_endpoint(self.endpoint)
else:
self.ipython_display.send_error(u"When you clean up the endpoint, all sessions will be lost, including the "
u"one used for this notebook. Include the -f parameter if that's your "
u"intention.")
return
def cleanup(self, line, cell="", local_ns=None):
self._assure_cell_body_is_empty(KernelMagics.cleanup.__name__, cell)
args = parse_argstring_or_throw(self.cleanup, line)
if args.force:
self._do_not_call_delete_session(u"")
self.spark_controller.cleanup_endpoint(self.endpoint)
else:
self.ipython_display.send_error(u"When you clean up the endpoint, all sessions will be lost, including the "
u"one used for this notebook. Include the -f parameter if that's your "
u"intention.")
return
def test_add_sessions_command_raises_google_no_account():
with patch('google.auth.default', return_value=(creds, 'project'), \
autospec=True):
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
connection_string = "-u http://url.com -t Google"
line = " ".join([command, name, language, connection_string])
magic.spark(line)
args = parse_argstring_or_throw(DataprocMagics.spark, line)
initialize_auth(args)
def _do_not_call_change_language(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_language, line)
language = args.language.lower()
if language not in LANGS_SUPPORTED:
self.ipython_display.send_error(u"'{}' language not supported in kernel magics.".format(language))
return
if self.session_started:
self.ipython_display.send_error(u"Cannot change the language if a session has been started.")
return
self.language = language
self.refresh_configuration()
def _do_not_call_change_language(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_language, line)
language = args.language.lower()
if language not in LANGS_SUPPORTED:
self.ipython_display.send_error(u"'{}' language not supported in kernel magics.".format(language))
return
if self.session_started:
self.ipython_display.send_error(u"Cannot change the language if a session has been started.")
return
self.language = language
self.refresh_configuration()
def test_delete_sessions_command_parses():
mock_method = MagicMock()
spark_controller.delete_session_by_name = mock_method
command = "delete -s name"
magic.spark(command)
mock_method.assert_called_once_with("name")
command = "delete -u URL -t {} -a username -p password -i 4".format(
AUTH_BASIC)
mock_method = MagicMock()
spark_controller.delete_session_by_id = mock_method
magic.spark(command)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, command)
mock_method.assert_called_once_with(Endpoint("URL", initialize_auth(args)),
4)
def test_add_sessions_command_parses_kerberos():
# Do not skip and python
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
connection_string = "-u http://url.com -t {}".format('Kerberos')
line = " ".join([command, name, language, connection_string])
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
auth_instance = initialize_auth(args)
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://url.com", initialize_auth(args)), False,
{"kind": "pyspark"})
assert_equals(auth_instance.url, "http://url.com")
def test_add_sessions_command_exception():
# Do not skip and python
add_sessions_mock = MagicMock(side_effect=BadUserDataException('hehe'))
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
connection_string = "-u http://url.com -t {} -a sdf -p w".format(
AUTH_BASIC)
line = " ".join([command, name, language, connection_string])
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://url.com", initialize_auth(args)), False,
{"kind": "pyspark"})
ipython_display.send_error.assert_called_once_with(
EXPECTED_ERROR_MSG.format(add_sessions_mock.side_effect))
def _do_not_call_change_endpoint(self, line, cell="", local_ns=None):
args = parse_argstring_or_throw(self._do_not_call_change_endpoint,
line)
username = args.username
password = args.password
server = args.server
auth = args.auth
mutualauth = args.krb_mutual_auth
kerberoshostname = args.krb_host_override
if self.session_started:
error = u"Cannot change the endpoint if a session has been started."
raise BadUserDataException(error)
self.endpoint = Endpoint(server,
auth,
username,
password,
krb_mutual_auth=mutualauth,
krb_host_override=kerberoshostname)
def configure(self, line, cell="", local_ns=None):
try:
dictionary = json.loads(cell)
except ValueError:
self.ipython_display.send_error(u"Could not parse JSON object from input '{}'".format(cell))
return
args = parse_argstring_or_throw(self.configure, line)
if self.session_started:
if not args.force:
self.ipython_display.send_error(u"A session has already been started. If you intend to recreate the "
u"session with new configurations, please include the -f argument.")
return
else:
self._do_not_call_delete_session(u"")
self._override_session_settings(dictionary)
self._do_not_call_start_session(u"")
else:
self._override_session_settings(dictionary)
self.info(u"")
def configure(self, line, cell="", local_ns=None):
try:
dictionary = json.loads(cell)
except ValueError:
self.ipython_display.send_error(u"Could not parse JSON object from input '{}'".format(cell))
return
args = parse_argstring_or_throw(self.configure, line)
if self.session_started:
if not args.force:
self.ipython_display.send_error(u"A session has already been started. If you intend to recreate the "
u"session with new configurations, please include the -f argument.")
return
else:
self._do_not_call_delete_session(u"")
self._override_session_settings(dictionary)
self._do_not_call_start_session(u"")
else:
self._override_session_settings(dictionary)
self.info(u"")
def test_add_sessions_command_parses_google_default_credentials():
with patch('google.auth.default', return_value=(creds, 'project'), \
autospec=True):
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
account = "-g default-credentials"
connection_string = "-u http://url.com -t Google"
line = " ".join([command, name, language, connection_string, account])
magic.spark(line)
args = parse_argstring_or_throw(DataprocMagics.spark, line)
auth_instance = initialize_auth(args)
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://url.com", initialize_auth(args)), False,
{"kind": "pyspark"})
assert_equals(auth_instance.url, "http://url.com")
isinstance(auth_instance, GoogleAuth)
assert_equals(auth_instance.active_credentials, 'default-credentials')
def delete(self, line, cell="", local_ns=None):
self._assure_cell_body_is_empty(KernelMagics.delete.__name__, cell)
args = parse_argstring_or_throw(self.delete, line)
session = args.session
if args.session is None:
self.ipython_display.send_error(u'You must provide a session ID (-s argument).')
return
if args.force:
id = self.spark_controller.get_session_id_for_client(self.session_name)
if session == id:
self.ipython_display.send_error(u"Cannot delete this kernel's session ({}). Specify a different session,"
u" shutdown the kernel to delete this session, or run %cleanup to "
u"delete all sessions for this endpoint.".format(id))
return
self.spark_controller.delete_session_by_id(self.endpoint, session)
else:
self.ipython_display.send_error(u"Include the -f parameter if you understand that all statements executed "
u"in this session will be lost.")
def test_add_sessions_command_parses_google_user_credentials():
with patch('sparkmagic.auth.google.list_credentialed_user_accounts', \
return_value=mock_credentialed_accounts_valid_accounts), patch('subprocess.check_output',\
return_value=AUTH_DESCRIBE_USER):
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l python"
account = "-g [email protected]"
connection_string = "-u http://url.com -t Google"
line = " ".join([command, name, language, connection_string, account])
magic.spark(line)
args = parse_argstring_or_throw(DataprocMagics.spark, line)
auth_instance = initialize_auth(args)
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://url.com", initialize_auth(args)), False,
{"kind": "pyspark"})
assert_equals(auth_instance.url, "http://url.com")
isinstance(auth_instance, GoogleAuth)
assert_equals(auth_instance.active_credentials, '*****@*****.**')
def test_add_sessions_command_extra_properties():
conf.override_all({})
magic.spark("config", "{\"extra\": \"yes\"}")
assert conf.session_configs() == {"extra": "yes"}
add_sessions_mock = MagicMock()
spark_controller.add_session = add_sessions_mock
command = "add"
name = "-s name"
language = "-l scala"
connection_string = "-u http://livyendpoint.com"
line = " ".join([command, name, language, connection_string])
magic.spark(line)
args = parse_argstring_or_throw(RemoteSparkMagics.spark, line)
args.auth = NO_AUTH
add_sessions_mock.assert_called_once_with(
"name", Endpoint("http://livyendpoint.com", initialize_auth(args)),
False, {
"kind": "spark",
"extra": "yes"
})
conf.override_all({})
def spark(self, line, cell="", local_ns=None):
"""Magic to execute spark remotely.
This magic allows you to create a Livy Scala or Python session against a Livy endpoint. Every session can
be used to execute either Spark code or SparkSQL code by executing against the SQL context in the session.
When the SQL context is used, the result will be a Pandas dataframe of a sample of the results.
If invoked with no subcommand, the cell will be executed against the specified session.
Subcommands
-----------
info
Display the available Livy sessions and other configurations for sessions.
add
Add a Livy session given a session name (-s), language (-l), and endpoint credentials.
The -k argument, if present, will skip adding this session if it already exists.
e.g. `%spark add -s test -l python -u https://sparkcluster.net/livy -t Kerberos -a u -p -k`
config
Override the livy session properties sent to Livy on session creation. All session creations will
contain these config settings from then on.
Expected value is a JSON key-value string to be sent as part of the Request Body for the POST /sessions
endpoint in Livy.
e.g. `%%spark config`
`{"driverMemory":"1000M", "executorCores":4}`
run
Run Spark code against a session.
e.g. `%%spark -s testsession` will execute the cell code against the testsession previously created
e.g. `%%spark -s testsession -c sql` will execute the SQL code against the testsession previously created
e.g. `%%spark -s testsession -c sql -o my_var` will execute the SQL code against the testsession
previously created and store the pandas dataframe created in the my_var variable in the
Python environment.
logs
Returns the logs for a given session.
e.g. `%spark logs -s testsession` will return the logs for the testsession previously created
delete
Delete a Livy session.
e.g. `%spark delete -s defaultlivy`
cleanup
Delete all Livy sessions created by the notebook. No arguments required.
e.g. `%spark cleanup`
"""
usage = "Please look at usage of %spark by executing `%spark?`."
user_input = line
args = parse_argstring_or_throw(self.spark, user_input)
subcommand = args.command[0].lower()
if args.auth is None:
args.auth = conf.get_auth_value(args.user, args.password)
else:
args.auth = args.auth
# info
if subcommand == "info":
if args.url is not None:
endpoint = Endpoint(args.url, args.auth, args.user, args.password)
info_sessions = self.spark_controller.get_all_sessions_endpoint_info(endpoint)
self._print_endpoint_info(info_sessions)
else:
self._print_local_info()
# config
elif subcommand == "config":
conf.override(conf.session_configs.__name__, json.loads(cell))
#conf file
elif subcommand == "encoded":
language = args.language
session = args.session
conf_json = json.loads(base64.b64decode(args.encodedconf).decode('utf-8'))
lang_args = conf_json['kernel_{}_credentials'.format(language)]
url = lang_args['url']
auth = lang_args['auth']
username = lang_args['username']
password = lang_args['base64_password']
conf.override_all(conf_json)
properties = conf.get_session_properties(language)
if url is not None:
endpoint = Endpoint(url, auth, username, password)
info_sessions = self.spark_controller.get_all_sessions_endpoint_info(endpoint)
if session in info_sessions:
print("found session")
else:
self.spark_controller.add_session(session, endpoint, True, properties)
coerce = get_coerce_value(args.coerce)
if args.context == CONTEXT_NAME_SPARK:
return self.execute_spark(cell, args.output, args.samplemethod,
args.maxrows, args.samplefraction, session, coerce)
elif args.context == CONTEXT_NAME_SQL:
return self.execute_sqlquery(cell, args.samplemethod, args.maxrows, args.samplefraction,
session, args.output, args.quiet, coerce)
else:
self.ipython_display.send_error("Context '{}' not found".format(args.context))
#check if session already exists
# add
elif subcommand == "add":
if args.url is None:
self.ipython_display.send_error("Need to supply URL argument (e.g. -u https://example.com/livyendpoint)")
return
name = args.session
language = args.language
endpoint = Endpoint(args.url, args.auth, args.user, args.password)
skip = args.skip
properties = conf.get_session_properties(language)
self.spark_controller.add_session(name, endpoint, skip, properties)
# delete
elif subcommand == "delete":
if args.session is not None:
self.spark_controller.delete_session_by_name(args.session)
elif args.url is not None:
if args.id is None:
self.ipython_display.send_error("Must provide --id or -i option to delete session at endpoint from URL")
return
endpoint = Endpoint(args.url, args.auth, args.user, args.password)
session_id = args.id
self.spark_controller.delete_session_by_id(endpoint, session_id)
else:
self.ipython_display.send_error("Subcommand 'delete' requires a session name or a URL and session ID")
# cleanup
elif subcommand == "cleanup":
if args.url is not None:
endpoint = Endpoint(args.url, args.auth, args.user, args.password)
self.spark_controller.cleanup_endpoint(endpoint)
else:
self.spark_controller.cleanup()
# logs
elif subcommand == "logs":
self.ipython_display.write(self.spark_controller.get_logs(args.session))
# run
elif len(subcommand) == 0:
coerce = get_coerce_value(args.coerce)
if args.context == CONTEXT_NAME_SPARK:
return self.execute_spark(cell, args.output, args.samplemethod,
args.maxrows, args.samplefraction, args.session, coerce)
elif args.context == CONTEXT_NAME_SQL:
return self.execute_sqlquery(cell, args.samplemethod, args.maxrows, args.samplefraction,
args.session, args.output, args.quiet, coerce)
else:
self.ipython_display.send_error("Context '{}' not found".format(args.context))
# error
else:
self.ipython_display.send_error("Subcommand '{}' not found. {}".format(subcommand, usage))
def spark(self, line, cell="", local_ns=None):
"""Magic to execute spark remotely.
This magic allows you to create a Livy Scala or Python session against a Livy endpoint. Every session can
be used to execute either Spark code or SparkSQL code by executing against the SQL context in the session.
When the SQL context is used, the result will be a Pandas dataframe of a sample of the results.
If invoked with no subcommand, the cell will be executed against the specified session.
Subcommands
-----------
info
Display the available Livy sessions and other configurations for sessions.
add
Add a Livy session given a session name (-s), language (-l), and endpoint credentials.
The -k argument, if present, will skip adding this session if it already exists.
e.g. `%spark add -s test -l python -u https://sparkcluster.net/livy -a u -p -k`
config
Override the livy session properties sent to Livy on session creation. All session creations will
contain these config settings from then on.
Expected value is a JSON key-value string to be sent as part of the Request Body for the POST /sessions
endpoint in Livy.
e.g. `%%spark config`
`{"driverMemory":"1000M", "executorCores":4}`
run
Run Spark code against a session.
e.g. `%%spark -s testsession` will execute the cell code against the testsession previously created
e.g. `%%spark -s testsession -c sql` will execute the SQL code against the testsession previously created
e.g. `%%spark -s testsession -c sql -o my_var` will execute the SQL code against the testsession
previously created and store the pandas dataframe created in the my_var variable in the
Python environment.
logs
Returns the logs for a given session.
e.g. `%spark logs -s testsession` will return the logs for the testsession previously created
delete
Delete a Livy session.
e.g. `%spark delete -s defaultlivy`
cleanup
Delete all Livy sessions created by the notebook. No arguments required.
e.g. `%spark cleanup`
"""
usage = "Please look at usage of %spark by executing `%spark?`."
user_input = line
args = parse_argstring_or_throw(self.spark, user_input)
subcommand = args.command[0].lower()
# info
if subcommand == "info":
if args.url is not None:
endpoint = Endpoint(args.url, args.user, args.password)
info_sessions = self.spark_controller.get_all_sessions_endpoint_info(endpoint)
self._print_endpoint_info(info_sessions)
else:
self._print_local_info()
# config
elif subcommand == "config":
conf.override(conf.session_configs.__name__, json.loads(cell))
# add
elif subcommand == "add":
if args.url is None:
self.ipython_display.send_error("Need to supply URL argument (e.g. -u https://example.com/livyendpoint)")
return
name = args.session
language = args.language
endpoint = Endpoint(args.url, args.user, args.password)
skip = args.skip
properties = conf.get_session_properties(language)
self.spark_controller.add_session(name, endpoint, skip, properties)
# delete
elif subcommand == "delete":
if args.session is not None:
self.spark_controller.delete_session_by_name(args.session)
elif args.url is not None:
if args.id is None:
self.ipython_display.send_error("Must provide --id or -i option to delete session at endpoint from URL")
return
endpoint = Endpoint(args.url, args.user, args.password)
session_id = args.id
self.spark_controller.delete_session_by_id(endpoint, session_id)
else:
self.ipython_display.send_error("Subcommand 'delete' requires a session name or a URL and session ID")
# cleanup
elif subcommand == "cleanup":
if args.url is not None:
endpoint = Endpoint(args.url, args.user, args.password)
self.spark_controller.cleanup_endpoint(endpoint)
else:
self.spark_controller.cleanup()
# logs
elif subcommand == "logs":
self.ipython_display.write(self.spark_controller.get_logs(args.session))
# run
elif len(subcommand) == 0:
if args.context == CONTEXT_NAME_SPARK:
(success, out) = self.spark_controller.run_command(Command(cell), args.session)
if success:
self.ipython_display.write(out)
else:
self.ipython_display.send_error(out)
elif args.context == CONTEXT_NAME_SQL:
return self.execute_sqlquery(cell, args.samplemethod, args.maxrows, args.samplefraction,
args.session, args.output, args.quiet)
else:
self.ipython_display.send_error("Context '{}' not found".format(args.context))
# error
else:
self.ipython_display.send_error("Subcommand '{}' not found. {}".format(subcommand, usage))
def help(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.help, line)
self._assure_cell_body_is_empty(KernelMagics.help.__name__, cell)
help_html = u"""
<table>
<tr>
<th>Magic</th>
<th>Example</th>
<th>Explanation</th>
</tr>
<tr>
<td>info</td>
<td>%%info</td>
<td>Outputs session information for the current Livy endpoint.</td>
</tr>
<tr>
<td>cleanup</td>
<td>%%cleanup -f</td>
<td>Deletes all sessions for the current Livy endpoint, including this notebook's session. The force flag is mandatory.</td>
</tr>
<tr>
<td>delete</td>
<td>%%delete -f -s 0</td>
<td>Deletes a session by number for the current Livy endpoint. Cannot delete this kernel's session.</td>
</tr>
<tr>
<td>logs</td>
<td>%%logs</td>
<td>Outputs the current session's Livy logs.</td>
</tr>
<tr>
<td>configure</td>
<td>%%configure -f<br/>{"executorMemory": "1000M", "executorCores": 4}</td>
<td>Configure the session creation parameters. The force flag is mandatory if a session has already been
created and the session will be dropped and recreated.<br/>Look at <a href="https://github.com/cloudera/livy#request-body">
Livy's POST /sessions Request Body</a> for a list of valid parameters. Parameters must be passed in as a JSON string.</td>
</tr>
<tr>
<td>spark</td>
<td>%%spark -o df<br/>df = spark.read.parquet('...</td>
<td>Executes spark commands.
Parameters:
<ul>
<li>-o VAR_NAME: The Spark dataframe of name VAR_NAME will be available in the %%local Python context as a
<a href="http://pandas.pydata.org/">Pandas</a> dataframe with the same name.</li>
<li>-m METHOD: Sample method, either <tt>take</tt> or <tt>sample</tt>.</li>
<li>-n MAXROWS: The maximum number of rows of a dataframe that will be pulled from Livy to Jupyter.
If this number is negative, then the number of rows will be unlimited.</li>
<li>-r FRACTION: Fraction used for sampling.</li>
</ul>
</td>
</tr>
<tr>
<td>sql</td>
<td>%%sql -o tables -q<br/>SHOW TABLES</td>
<td>Executes a SQL query against the variable sqlContext (Spark v1.x) or spark (Spark v2.x).
Parameters:
<ul>
<li>-o VAR_NAME: The result of the SQL query will be available in the %%local Python context as a
<a href="http://pandas.pydata.org/">Pandas</a> dataframe.</li>
<li>-q: The magic will return None instead of the dataframe (no visualization).</li>
<li>-m, -n, -r are the same as the %%spark parameters above.</li>
</ul>
</td>
</tr>
<tr>
<td>local</td>
<td>%%local<br/>a = 1</td>
<td>All the code in subsequent lines will be executed locally. Code must be valid Python code.</td>
</tr>
<tr>
<td>send_to_spark</td>
<td>%%send_to_spark -i variable -t str -n var</td>
<td>Sends a variable from local output to spark cluster.
<br/>
Parameters:
<ul>
<li>-i VAR_NAME: Local Pandas DataFrame(or String) of name VAR_NAME will be available in the %%spark context as a
Spark dataframe(or String) with the same name.</li>
<li>-t TYPE: Specifies the type of variable passed as -i. Available options are:
`str` for string and `df` for Pandas DataFrame. Optional, defaults to `str`.</li>
<li>-n NAME: Custom name of variable passed as -i. Optional, defaults to -i variable name.</li>
<li>-m MAXROWS: Maximum amount of Pandas rows that will be sent to Spark. Defaults to 2500.</li>
</ul>
</td>
</tr>
</table>
"""
self.ipython_display.html(help_html)
def spark(self, line, cell="", local_ns=None):
"""Magic to execute spark remotely.
This magic allows you to create a Livy Scala or Python session against a Livy endpoint.
Every session can be used to execute either Spark code or SparkSQL code by executing
against the SQL context in the session. When the SQL context is used, the result will
be a Pandas dataframe of a sample of the results.
If invoked with no subcommand, the cell will be executed against the specified session.
Subcommands
-----------
info
Display the available Livy sessions and other configurations for sessions with None,
Basic, or Kerberos auth.
add
Add a Livy session given a session name (-s), language (-l), and endpoint
credentials. The -k argument, if present, will skip adding this session if it already
exists.
e.g. `%spark add -s test -l python -u https://sparkcluster.net/livy -t Kerberos -a u -p -k`
config
Override the livy session properties sent to Livy on session creation. All session
creations will contain these config settings from then on.
Expected value is a JSON key-value string to be sent as part of the Request Body
for the POST /sessions endpoint in Livy.
e.g. `%%spark config`
`{"driverMemory":"1000M", "executorCores":4}`
run
Run Spark code against a session.
e.g. `%%spark -s testsession` will execute the cell code against the testsession
previously created
e.g. `%%spark -s testsession -c sql` will execute the SQL code against the
testsession previously created
e.g. `%%spark -s testsession -c sql -o my_var` will execute the SQL code against
the testsession previously created and store the pandas dataframe created in the
my_var variable in the Python environment.
logs
Returns the logs for a given session.
e.g. `%spark logs -s testsession` will return the logs for the testsession
previously created
delete
Delete a Livy session.
e.g. `%spark delete -s defaultlivy`
cleanup
Delete all Livy sessions created by the notebook. No arguments required.
e.g. `%spark cleanup`
"""
user_input = line
args = parse_argstring_or_throw(self.spark, user_input)
subcommand = args.command[0].lower()
if args.auth == "Google" and subcommand == "add":
if args.url is None:
self.ipython_display.send_error(
"Need to supply URL argument (e.g. -u https://example.com/livyendpoint)"
)
return
name = args.session
language = args.language
endpoint = Endpoint(args.url, initialize_auth(args))
self.endpoints[args.url] = endpoint
# convert Endpoints in self.endpoints into list of dictionaries, each storing an
# Endpoints writeable attributes
stored_endpoints = [
SerializableEndpoint(endpoint).__dict__ for endpoint in self.endpoints.values()
]
self.db['autorestore/' + 'stored_endpoints'] = stored_endpoints
skip = args.skip
properties = conf.get_session_properties(language)
self.spark_controller.add_session(name, endpoint, skip, properties)
# session_id_to_name dict is necessary to restore session name across notebook sessions
# since the livy server does not store the name.
session_id_to_name = get_session_id_to_name(self.db, self.ipython_display)
# add session id -> name to session_id_to_name dict
session_id_to_name[self.spark_controller.session_manager.get_session(name).id] = name
self.db['autorestore/' + 'session_id_to_name'] = session_id_to_name
elif subcommand == "info":
if args.url is not None and args.id is not None:
endpoint = Endpoint(args.url, initialize_auth(args))
info_sessions = self.spark_controller.get_all_sessions_endpoint_info(endpoint)
self._print_endpoint_info(info_sessions, args.id)
else:
self._print_local_info()
else:
self.__remotesparkmagics.spark(line, cell, local_ns=None)
def help(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.help, line)
self._assure_cell_body_is_empty(KernelMagics.help.__name__, cell)
help_html = u"""
<table>
<tr>
<th>Magic</th>
<th>Example</th>
<th>Explanation</th>
</tr>
<tr>
<td>info</td>
<td>%%info</td>
<td>Outputs session information for the current Livy endpoint.</td>
</tr>
<tr>
<td>cleanup</td>
<td>%%cleanup -f</td>
<td>Deletes all sessions for the current Livy endpoint, including this notebook's session. The force flag is mandatory.</td>
</tr>
<tr>
<td>delete</td>
<td>%%delete -f -s 0</td>
<td>Deletes a session by number for the current Livy endpoint. Cannot delete this kernel's session.</td>
</tr>
<tr>
<td>logs</td>
<td>%%logs</td>
<td>Outputs the current session's Livy logs.</td>
</tr>
<tr>
<td>configure</td>
<td>%%configure -f<br/>{"executorMemory": "1000M", "executorCores": 4}</td>
<td>Configure the session creation parameters. The force flag is mandatory if a session has already been
created and the session will be dropped and recreated.<br/>Look at <a href="https://github.com/cloudera/livy#request-body">
Livy's POST /sessions Request Body</a> for a list of valid parameters. Parameters must be passed in as a JSON string.</td>
</tr>
<tr>
<td>sql</td>
<td>%%sql -o tables -q<br/>SHOW TABLES</td>
<td>Executes a SQL query against the sqlContext.
Parameters:
<ul>
<li>-o VAR_NAME: The result of the query will be available in the %%local Python context as a
<a href="http://pandas.pydata.org/">Pandas</a> dataframe.</li>
<li>-q: The magic will return None instead of the dataframe (no visualization).</li>
<li>-m METHOD: Sample method, either <tt>take</tt> or <tt>sample</tt>.</li>
<li>-n MAXROWS: The maximum number of rows of a SQL query that will be pulled from Livy to Jupyter.
If this number is negative, then the number of rows will be unlimited.</li>
<li>-r FRACTION: Fraction used for sampling.</li>
</ul>
</td>
</tr>
<tr>
<td>local</td>
<td>%%local<br/>a = 1</td>
<td>All the code in subsequent lines will be executed locally. Code must be valid Python code.</td>
</tr>
</table>
"""
self.ipython_display.html(help_html)
def help(self, line, cell="", local_ns=None):
parse_argstring_or_throw(self.help, line)
self._assure_cell_body_is_empty(KernelMagics.help.__name__, cell)
help_html = u"""
<table>
<tr>
<th>Magic</th>
<th>Example</th>
<th>Explanation</th>
</tr>
<tr>
<td>info</td>
<td>%%info</td>
<td>Outputs session information for the current Livy endpoint.</td>
</tr>
<tr>
<td>cleanup</td>
<td>%%cleanup -f</td>
<td>Deletes all sessions for the current Livy endpoint, including this notebook's session. The force flag is mandatory.</td>
</tr>
<tr>
<td>delete</td>
<td>%%delete -f -s 0</td>
<td>Deletes a session by number for the current Livy endpoint. Cannot delete this kernel's session.</td>
</tr>
<tr>
<td>logs</td>
<td>%%logs</td>
<td>Outputs the current session's Livy logs.</td>
</tr>
<tr>
<td>configure</td>
<td>%%configure -f<br/>{"executorMemory": "1000M", "executorCores": 4}</td>
<td>Configure the session creation parameters. The force flag is mandatory if a session has already been
created and the session will be dropped and recreated.<br/>Look at <a href="https://github.com/cloudera/livy#request-body">
Livy's POST /sessions Request Body</a> for a list of valid parameters. Parameters must be passed in as a JSON string.</td>
</tr>
<tr>
<td>sql</td>
<td>%%sql -o tables -q<br/>SHOW TABLES</td>
<td>Executes a SQL query against the sqlContext.
Parameters:
<ul>
<li>-o VAR_NAME: The result of the query will be available in the %%local Python context as a
<a href="http://pandas.pydata.org/">Pandas</a> dataframe.</li>
<li>-q: The magic will return None instead of the dataframe (no visualization).</li>
<li>-m METHOD: Sample method, either <tt>take</tt> or <tt>sample</tt>.</li>
<li>-n MAXROWS: The maximum number of rows of a SQL query that will be pulled from Livy to Jupyter.
If this number is negative, then the number of rows will be unlimited.</li>
<li>-r FRACTION: Fraction used for sampling.</li>
</ul>
</td>
</tr>
<tr>
<td>local</td>
<td>%%local<br/>a = 1</td>
<td>All the code in subsequent lines will be executed locally. Code must be valid Python code.</td>
</tr>
</table>
"""
self.ipython_display.html(help_html)
