def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
command = link.get_command()
try:
from dataiku.doctor import commands
COMMANDS = {
command_name: json_api(command_method)
for (command_name, command_method) in commands._list_commands()
}
task = command["command"]
arg = command.get("arg", "")
logging.info("Running analysis command: %s" % task)
if task not in COMMANDS:
raise ValueError("Command %s is unknown." % task)
else:
api_command = COMMANDS[task]
ret = api_command(arg)
link.send_json(ret)
# send end of stream
link.send_string('')
except:
link.send_string('') # mark failure
traceback.print_exc()
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
try:
# retrieve the initialization info and initiate serving
command = link.get_command()
function_name = command.get('functionName')
code_file = command.get('codeFilePath')
data_folders = command.get('resourceFolderPaths', [])
loaded_function = LoadedFunction(code_file, function_name,
data_folders)
logging.info("Predictor ready")
link.send_json({"ok": True})
stored_exception = None
# loop and process commands
while True:
request = link.read_json()
if request is None:
break
used_api_key = request.get("usedAPIKey", None)
if used_api_key is not None:
os.environ["DKU_CURRENT_REQUEST_USED_API_KEY"] = used_api_key
before = time.time()
response = loaded_function.predict(request["params"])
after = time.time()
link.send_json({
'ok': True,
'resp': response,
'execTimeUS': int(1000000 * (after - before))
})
if used_api_key is not None:
del os.environ["DKU_CURRENT_REQUEST_USED_API_KEY"]
# send end of stream
logging.info("Work done")
link.send_string('')
except:
logging.exception("Function user code failed")
link.send_string('') # send null to mark failure
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
try:
command = link.get_command()
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
root = command["root"]
# get the fs provider object
clazz = get_clazz_in_code(code, FSProvider)
arg_count = len(inspect.getargspec(clazz.__init__).args)
fsprovider = None
if arg_count == 1:
fsprovider = clazz()
elif arg_count == 2:
fsprovider = clazz(root)
elif arg_count == 3:
fsprovider = clazz(root, config)
elif arg_count == 4:
fsprovider = clazz(root, config, plugin_config)
else:
reason = "Wrong signature of the FSProvider subclass: %i args" % arg_count
link.send_json({'ok': False, 'reason': reason})
raise Exception(reason)
link.send_json({'ok': True})
# loop and process commands
closed = False
while not closed:
request = link.read_json()
if request is None:
break
closed, response = handle_request(request, fsprovider, closed)
link.send_json(response)
# send end of stream
link.send_string('')
except:
link.send_string('') # mark failure
traceback.print_exc()
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
command = link.get_command()
try:
# logging.info("Got %s" % json.dumps(command))
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
ctx = {"config": config, "plugin_config": plugin_config}
exec(code, ctx, ctx)
functions = [o for o in ctx.values() if inspect.isfunction(o)]
f = functions[0] if len(functions) == 1 else ctx.get('process', None)
if f is None:
raise Exception('No function "process" defined')
object_type = command['objectType']
full_name = command['fullName']
partition_id = command.get('partitionId', None)
if object_type == "DATASET":
obj_arg = Dataset(full_name)
elif object_type == "SAVED_MODEL":
obj_arg = Model(full_name)
elif object_type == "MANAGED_FOLDER":
obj_arg = Folder(full_name)
# work and get output
if command['command'] == 'compute':
result = compute_metric(obj_arg, partition_id, f)
elif command['command'] == 'check':
last_values = command.get("lastValues", {})
result = run_check(obj_arg, partition_id, last_values, f)
else:
raise Exception("Unknown command")
if result is None:
raise Exception("Code did not return a result")
link.send_json(result)
except:
traceback.print_exc()
link.send_string('') # send null to mark failure
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
try:
# retrieve the initialization info and initiate serving
command = link.get_command()
model_type = command.get('modelType')
code_file = command.get('codeFilePath')
data_folder = command.get('resourceFolderPath', None)
loaded_model = LoadedModel(model_type, code_file, data_folder)
logging.info("Predictor ready")
link.send_json({"ok":True})
stored_exception = None
# loop and process commands
while True:
request = link.read_json()
if request is None:
break
used_api_key = request.get("usedAPIKey", None)
if used_api_key is not None:
os.environ["DKU_CURRENT_REQUEST_USED_API_KEY"] = used_api_key
before = time.time()
response = handle_predict(loaded_model, request["obj"])
after = time.time()
response["execTimeUS"] = int(1000000 * (after-before))
link.send_json(response)
if used_api_key is not None:
del os.environ["DKU_CURRENT_REQUEST_USED_API_KEY"]
# send end of stream
logging.info("Work done")
link.send_string('')
except:
ex_type, ex, tb = sys.exc_info()
traceback.print_exc()
link.send_string('') # send null to mark failure
link.send_json({'errorType': str(ex_type), 'message':str(ex), 'traceback':traceback.extract_tb(tb)})
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
command = link.get_command()
try:
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
# get the exporter object
clazz = get_clazz_in_code(code, Exporter)
arg_count = len(inspect.getargspec(clazz.__init__).args)
exporter = None
if arg_count == 1:
exporter = clazz()
elif arg_count == 2:
exporter = clazz(config)
elif arg_count == 3:
exporter = clazz(config, plugin_config)
else:
raise Exception(
"Wrong signature of the Exporter subclass: %i args" %
arg_count)
# get task and dispatch work to exporter
task = command["task"]
if task == "export":
# schema is mandatory
with link.get_input() as input:
row_count = export_rows(
exporter, command["exportBehavior"], command["schema"],
input, command.get("destinationFilePath", None))
else:
raise Exception("Unexpected task %s" % task)
# send ack
link.send_json({'ok': True, 'count': row_count})
except:
link.send_string('') # mark failure
traceback.print_exc()
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
try:
# get work to do
command = link.get_command()
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
root = command["root"]
path = command["path"]
limit = command["limit"]
# get the fs provider object
clazz = get_clazz_in_code(code, FSProvider)
arg_count = len(inspect.getargspec(clazz.__init__).args)
fsprovider = None
if arg_count == 1:
fsprovider = clazz()
elif arg_count == 2:
fsprovider = clazz(root)
elif arg_count == 3:
fsprovider = clazz(root, config)
elif arg_count == 4:
fsprovider = clazz(root, config, plugin_config)
else:
reason = "Wrong signature of the FSProvider subclass: %i args" % arg_count
raise Exception(reason)
with link.get_output() as output:
fsprovider.read(path, output, limit)
# send end of stream
link.send_string('')
# send ack
link.send_json({'ok':True})
except:
link.send_string('') # end stream to send ack
traceback.print_exc()
error = get_json_friendly_error()
link.send_json({'ok':False, 'error':error})
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
try:
# retrieve the initialization info and initiate serving
command = link.get_command()
model_folder = command.get('modelFolder')
try:
conditional_outputs = load_from_filepath(osp.join(model_folder, "conditional_outputs.json"))
except Exception as e:
logging.exception("Can't load conditional outputs: " + str(e))
conditional_outputs = []
predictor = build_predictor_for_saved_model(model_folder, "PREDICTION", conditional_outputs)
logging.info("Predictor ready")
link.send_json({"ok":True})
stored_exception = None
# loop and process commands
while True:
request = link.read_json()
if request is None:
break
before = time.time()
response = handle_predict(predictor, request)
after = time.time()
response["execTimeUS"] = int(1000000 * (after-before))
link.send_json(response)
# send end of stream
logging.info("Work done")
link.send_string('')
except:
logging.exception("Prediction user code failed")
link.send_string('') # send null to mark failure
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
logging.info("Starting Webapp backend")
# Connect
link = JavaLink(port, secret)
link.connect()
logging.info("Webapp backend connected to DSS")
# get work to do
command = link.get_command()
try:
logging.info("Starting backend for web app: %s.%s" %
(command["projectKey"], command["webAppId"]))
@app.route('/__ping')
def ping():
return "pong"
# Execute user's code
exec(command["code"], globals(),
globals()) # in globals so that flask can find them
# Start the server
from werkzeug.serving import make_server
srv = make_server("127.0.0.1", 0, app)
myport = srv.server_port
link.send_json({'type': 'STARTED', "port": myport})
srv.serve_forever()
except:
logging.exception("Backend main loop failed")
link.send_json({'type': 'ERROR'})
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
command_is_streaming = None
# get work to do
command = link.get_command()
try:
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
# get the formatter object
clazz = get_clazz_in_code(code, Formatter)
arg_count = len(inspect.getargspec(clazz.__init__).args)
formatter = None
if arg_count == 1:
formatter = clazz()
elif arg_count == 2:
formatter = clazz(config)
elif arg_count == 3:
formatter = clazz(config, plugin_config)
else:
raise Exception(
"Wrong signature of the Formatter subclass: %i args" %
arg_count)
# get task and dispatch work to formatter
task = command["task"]
if task == "read":
# extract mode
command_is_streaming = True
with link.get_input() as input, link.get_output() as output:
row_count = extract_rows(formatter,
command.get("schema",
None), input, output)
# send end of stream
link.send_string('')
# send acknowledgment
link.send_json({'ok': True, 'count': row_count})
elif task == "write":
# format mode (schema is mandatory)
command_is_streaming = True
with link.get_input() as input, link.get_output() as output:
row_count = format_rows(formatter, command["schema"], input,
output)
# send end of stream
link.send_string('')
# send acknowledgment
link.send_json({'ok': True, 'count': row_count})
elif task == "schema":
# read schema mode
command_is_streaming = False
with link.get_input() as input:
schema = extract_schema(formatter, input)
if schema is not None:
link.send_json(schema)
else:
link.send_json({
'columns': [{
'name': '__dku_empty_schema__',
'type': 'string'
}]
})
else:
raise Exception("Unexpected task %s" % task)
except:
traceback.print_exc()
error = get_json_friendly_error()
link.send_string(
'') # send null to mark failure or to mark end of stream
if not command_is_streaming:
link.send_json(error)
else:
link.send_json({'ok': False, 'error': error})
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
command = link.get_command()
try:
project_key = command.get("projectKey", {})
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
cluster_id = command["clusterId"]
cluster_name = command["clusterName"]
# get the exporter object
clazz = get_clazz_in_code(code, Cluster)
arg_count = len(inspect.getargspec(clazz.__init__).args)
cluster = None
if arg_count == 5:
cluster = clazz(cluster_id, cluster_name, config, plugin_config)
else:
raise Exception("Wrong signature of the Cluster subclass: %i args but expected 5 (self, cluster_id, name, config, plugin_config)" % arg_count)
# work
call_name = command["type"]
data = command.get("data", {})
if call_name == 'start':
# init progress reporting if relevant
report_progress = get_progress_callback(cluster.get_start_progress_target, link)
arg_count = len(inspect.getargspec(cluster.start).args)
if arg_count == 1:
result = cluster.start()
elif arg_count == 2:
result = cluster.start(report_progress)
if isinstance(result, list) or isinstance(result, tuple):
cluster_settings = result[0]
cluster_data = result[1] if len(result) > 1 else {}
cluster_setup = {}
cluster_setup['hadoopSettings'] = cluster_settings.get('hadoop', None)
cluster_setup['hiveSettings'] = cluster_settings.get('hive', None)
cluster_setup['impalaSettings'] = cluster_settings.get('impala', None)
cluster_setup['sparkSettings'] = cluster_settings.get('spark', None)
cluster_setup['data'] = cluster_data
send_result_json(cluster_setup, link)
else:
raise Exception("start() didn't return an object of a valid type: %s" % type(result))
elif call_name == 'stop':
# init progress reporting if relevant
report_progress = get_progress_callback(cluster.get_stop_progress_target, link)
arg_count = len(inspect.getargspec(cluster.stop).args)
if arg_count == 2:
result = cluster.stop(data)
elif arg_count == 3:
result = cluster.stop(data, report_progress)
send_result_json({'ok':True}, link)
else:
if hasattr(cluster, call_name):
action_attr = getattr(cluster, call_name)
if inspect.ismethod(action_attr):
result = action_attr(data)
# convert to something that is legit for a JsonObject
if result is None:
result = {}
if not isinstance(result, dict):
result = {'result':result}
# send
link.send_json({'ok':True, 'response':result})
else:
raise Exception("Wrong call type : %s is not a method" % call_name)
else:
raise Exception("Wrong call type : %s" % call_name)
# send end of stream (data is expected as a stream)
link.send_string('')
except:
traceback.print_exc()
send_error(link)
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
try:
command = link.get_command()
config = command.get("config", {})
plugin_config = command.get("pluginConfig", {})
code = command["code"]
# get the connector object
clazz = get_clazz_in_code(code, Connector)
arg_count = len(inspect.getargspec(clazz.__init__).args)
connector = None
if arg_count == 1:
connector = clazz()
elif arg_count == 2:
connector = clazz(config)
elif arg_count == 3:
connector = clazz(config, plugin_config)
else:
raise Exception(
"Wrong signature of the Connector subclass: %i args" %
arg_count)
link.send_json({'ok': True})
stored_error = None
# loop and process commands
while True:
request = link.read_json()
if request is None:
break
response = None
task = request["task"]
logging.info("Processing task: %s" % task)
if task == "read_rows":
schema = request.get("schema", None)
partitioning = request.get("partitioning", None)
partition_id = request.get("partitionId", None)
limit = request.get("limit", None)
stored_error = None
try:
with link.get_output() as output:
read_rows(connector, schema, partitioning,
partition_id, limit, output)
except:
logging.exception("Connector send fail, storing exception")
stored_error = get_json_friendly_error()
link.send_string('')
elif task == "finish_read_session":
if stored_error is None:
link.send_json({"ok": True})
else:
link.send_json({"ok": False, "error": stored_error})
elif task == "write_rows":
schema = request.get("schema", None)
partitioning = request.get("partitioning", None)
partition_id = request.get("partitionId", None)
with link.get_input() as input:
write_rows(connector, schema, partitioning, partition_id,
input)
link.send_json({'ok': True})
elif task == "get_schema":
link.send_json({'schema': connector.get_read_schema()})
elif task == "get_partitioning_scheme":
link.send_json({'partitioning': connector.get_partitioning()})
elif task == "list_partitions":
partitioning = request.get("partitioning", None)
link.send_json(
{'partitions': connector.list_partitions(partitioning)})
elif task == "partition_exists":
partitioning = request.get("partitioning", None)
partition_id = request.get("partitionId", None)
link.send_json({
"exists":
connector.partition_exists(partitioning, partition_id)
})
elif task == "records_count":
partitioning = request.get("partitioning", None)
partition_id = request.get("partitionId", None)
link.send_json({
"count":
connector.get_records_count(partitioning, partition_id)
})
else:
raise Exception("Unexpected task %s" % task)
# send end of stream
logging.info("Work done")
link.send_string('')
except:
link.send_string('') # mark failure
traceback.print_exc()
link.send_json(get_json_friendly_error())
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
is_serving = False
# get work to do
try:
command = link.get_command()
plugin_config = command.get("pluginConfig", {})
code = command["code"]
# get the helper function
ctx = {}
python2_friendly_exec(code, ctx, ctx)
functions = [o for o in ctx.values() if inspect.isfunction(o)]
f = functions[0] if len(functions) == 1 else ctx.get('do', None)
if f is None:
raise Exception('No function "do" defined')
f_args_count = len(inspect.getargspec(f).args)
if f_args_count >= 5:
reason = "Too many arguments for the do() function : %i args" % f_args_count
raise Exception(reason)
link.send_json({'ok': True})
def call_do(payload, config, plugin_config, inputs):
result = None
if f_args_count == 0:
result = f()
if f_args_count == 1:
result = f(payload)
if f_args_count == 2:
result = f(payload, config)
if f_args_count == 3:
result = f(payload, config, plugin_config)
if f_args_count == 4:
result = f(payload, config, plugin_config, inputs)
return result
is_serving = True
# loop and process commands
while True:
request = link.read_json()
if request is None:
break
response = call_do(request.get('payload', None),
request.get('config', {}), plugin_config,
request.get('inputs', []))
if response is None:
raise Exception("Empty response to %s" % json.dumps(request))
link.send_json(response)
# send end of stream
link.send_string('')
except:
error = get_json_friendly_error()
link.send_string('') # mark failure
traceback.print_exc()
if not is_serving:
link.send_json(error)
else:
link.send_json({'ok': False, 'error': error})
finally:
# done
link.close()
def serve(port, secret):
link = JavaLink(port, secret)
# initiate connection
link.connect()
# get work to do
command = link.get_command()
try:
code = command["code"]
notebook = command["notebook"]
project_key = command.get("projectKey", None)
dataset_project_key = command.get("datasetProjectKey", None)
dataset_name = command.get("datasetName", None)
# get the expansion function
ctx = {}
python2_friendly_exec(code, ctx, ctx)
functions = [o for o in ctx.values() if inspect.isfunction(o)]
f = functions[0] if len(functions) == 1 else ctx.get('expand', None)
if f is None:
raise Exception('No function "expand" defined')
f_args_count = len(inspect.getargspec(f).args)
if f_args_count >= 5:
reason = "Too many arguments for the expand() function : %i args" % f_args_count
raise Exception(reason)
if f_args_count < 1:
reason = "Too few arguments for the expand() function : %i args" % f_args_count
raise Exception(reason)
def call_expand(notebook, project_key, dataset_project_key,
dataset_name):
result = None
if f_args_count == 1:
result = f(notebook)
if f_args_count == 2:
result = f(notebook, project_key)
if f_args_count == 3:
result = f(notebook, project_key, dataset_project_key)
if f_args_count == 4:
result = f(notebook, project_key, dataset_project_key,
dataset_name)
return result
# expand for real
expanded = call_expand(notebook, project_key, dataset_project_key,
dataset_name)
# send result
if isinstance(
expanded, unicode
): # we need to send a str (for python streams) and the java end is waiting for utf8
expanded = expanded.encode('utf8')
link.send_block(expanded)
# send end of stream
link.send_string('')
# send ack
link.send_json({'ok': True, 'count': len(expanded)})
except:
link.send_string('') # mark failure
traceback.print_exc()
link.send_json({'ok': False, 'error': get_json_friendly_error()})
finally:
# done
link.close()