birbnetes/model-service
birbnetes/model-service
4
0
Fork 0
Code Issues Releases Wiki Activity

Added more traces
All checks were successful
continuous-integration/drone/push Build is passing
Details

Browse Source
This commit is contained in:
Pünkösd Marcell 2021-08-05 17:54:11 +02:00
parent 17927b472d
commit d807bc6b00
4 changed files with 304 additions and 162 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
model_service/views/cnn_view.py
View File

@ -6,6 +6,7 @@ from minio.error import NoSuchKey
from schemas import AIModelSchema, DefaultSchema, InfoSchema
from marshmallow.exceptions import ValidationError
from utils import multipart_required, storage, ensure_buckets
import opentracing
class CNNView(FlaskView):
@ -21,34 +22,48 @@ class CNNView(FlaskView):
def post(self):
# get important data from the request
with opentracing.tracer.start_active_span('parseAndValidate'):
try:
info = self.info_schema.loads(request.form.get('info'))
except ValidationError as e:
abort(400, str(e))
return abort(400, str(e))
# check for conflict
m = AIModel.query.filter_by(id=info['id']).first()
if m:
abort(409)
return abort(409)
# get and validate file
model_file = request.files['modelFile']
if model_file.content_length <= 0:
abort(411, f"Content length for modelFile is not a positive integer or missing.")
return abort(411, f"Content length for modelFile is not a positive integer or missing.")
weights_file = request.files['weightsFile']
if weights_file.content_length <= 0:
abort(411, f"Content length for weightsFile is not a positive integer or missing.")
return abort(411, f"Content length for weightsFile is not a positive integer or missing.")
# create bucket if necessary
with opentracing.tracer.start_active_span('ensureBuckets'):
ensure_buckets()
# Create the entry in the db
with opentracing.tracer.start_active_span('sqlalchemy.create'):
m = AIModel(id=info['id'], type=AIModelType.cnn, target_class_name=info['target_class_name'])
db.session.add(m)
# Put files into MinIO
with opentracing.tracer.start_active_span('putObjectsInMinio'):
with opentracing.tracer.start_active_span(
'minio.putObject',
tags={
"bucket": current_app.config['MINIO_CNN_BUCKET_NAME'],
"object_name": self.MODEL_DIRECTORY + str(m.id),
"length": model_file.content_length,
"component": "model"
}
):
storage.connection.put_object(
current_app.config['MINIO_CNN_BUCKET_NAME'],
self.MODEL_DIRECTORY + str(m.id),
@ -56,7 +71,15 @@ class CNNView(FlaskView):
model_file.content_length,
content_type=model_file.content_type
)
with opentracing.tracer.start_active_span(
'minio.putObject',
tags={
"bucket": current_app.config['MINIO_CNN_BUCKET_NAME'],
"object_name": self.WEIGHTS_DIRECTORY + str(m.id),
"length": model_file.content_type,
"component": "weights"
}
):
storage.connection.put_object(
current_app.config['MINIO_CNN_BUCKET_NAME'],
self.WEIGHTS_DIRECTORY + str(m.id),
@ -65,23 +88,53 @@ class CNNView(FlaskView):
content_type=weights_file.content_type
)
db.session.add(m)
with opentracing.tracer.start_active_span('sqlalchemy.commit'):
db.session.commit()
return jsonify(self.aimodel_schema.dump(m)), 200
def delete(self, id_: str):
with opentracing.tracer.start_active_span(
'sqlalchemy.select',
tags={"aimodel_type": AIModelType.cnn, "id": id_}
):
if id_ == "$default":
default = Default.query.filter_by(type=AIModelType.cnn).first_or_404()
m = default.aimodel
else:
m = AIModel.query.filter_by(type=AIModelType.cnn, id=id_).first_or_404()
storage.connection.remove_object(current_app.config['MINIO_CNN_BUCKET_NAME'], self.WEIGHTS_DIRECTORY + str(m.id))
storage.connection.remove_object(current_app.config['MINIO_CNN_BUCKET_NAME'], self.MODEL_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span('removeFromMinio'):
with opentracing.tracer.start_active_span(
'minio.removeObject',
tags={
"bucket": current_app.config['MINIO_CNN_BUCKET_NAME'],
"object_name": self.WEIGHTS_DIRECTORY + str(m.id),
"component": "weights"
}
):
storage.connection.remove_object(current_app.config['MINIO_CNN_BUCKET_NAME'],
self.WEIGHTS_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span(
'minio.removeObject',
tags={
"bucket": current_app.config['MINIO_CNN_BUCKET_NAME'],
"object_name": self.MODEL_DIRECTORY + str(m.id),
"component": "model"
}
):
storage.connection.remove_object(current_app.config['MINIO_CNN_BUCKET_NAME'],
self.MODEL_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span(
'sqlalchemy.delete',
tags={"aimodel_type": AIModelType.cnn, "id": id_}
):
db.session.delete(m)
with opentracing.tracer.start_active_span('sqlalchemy.commit'):
db.session.commit()
return '', 204
@ -89,6 +142,10 @@ class CNNView(FlaskView):
@route('<id_>/file')
def get_file(self, id_: str):
with opentracing.tracer.start_active_span(
'sqlalchemy.select',
tags={"aimodel_type": AIModelType.cnn, "id": id_}
):
if id_ == "$default":
default = Default.query.filter_by(type=AIModelType.cnn).first_or_404()
m = default.aimodel
@ -97,12 +154,23 @@ class CNNView(FlaskView):
if "weights" in request.args:
path = self.WEIGHTS_DIRECTORY + str(m.id)
component = "weights"
else:
path = self.MODEL_DIRECTORY + str(m.id)
component = "model"
with opentracing.tracer.start_active_span(
'minio.getObject',
tags={
"bucket": current_app.config['MINIO_CNN_BUCKET_NAME'],
"object_name": path,
"component": component
}
):
# Note: this only initiates the download, the file download itself is streamed
try:
data = storage.connection.get_object(current_app.config['MINIO_CNN_BUCKET_NAME'], path)
except NoSuchKey:
abort(500, "The ID is stored in the database but not int the Object Store")
return abort(500, "The ID is stored in the database but not int the Object Store")
return Response(data.stream(), mimetype=data.headers['Content-type'])

29
model_service/views/root_view.py
View File

@ -9,6 +9,8 @@ from model import db, AIModel, AIModelType, Default
from schemas import AIModelSchema, DefaultSchema
import opentracing
class RootView(FlaskView):
route_base = '/'
@ -21,7 +23,9 @@ class RootView(FlaskView):
## Shared stuff goes here
def index(self):
with opentracing.tracer.start_active_span('sqlalchemy.select'):
models = AIModel.query.all()
return jsonify(self.aimodels_schema.dump(models))
@route('/<type_>')
@ -29,9 +33,11 @@ class RootView(FlaskView):
try:
aimodel_type = AIModelType[type_]
except KeyError:
abort(404, "Unknown type")
return abort(404, "Unknown type")
with opentracing.tracer.start_active_span('sqlalchemy.select', tags={"aimodel_type": aimodel_type}):
models = AIModel.query.filter_by(type=aimodel_type).all()
return jsonify(self.aimodels_schema.dump(models)), 200
@route('/<type_>/<id_>')
@ -39,15 +45,24 @@ class RootView(FlaskView):
try:
aimodel_type = AIModelType[type_]
except KeyError:
abort(404, "Unknown type")
return abort(404, "Unknown type")
with opentracing.tracer.start_active_span(
'sqlalchemy.select',
tags={"aimodel_type": aimodel_type, "id": id_}
):
if id_ == "$default":
default = Default.query.filter_by(type=aimodel_type).first_or_404()
m = default.aimodel
else:
m = AIModel.query.filter_by(type=aimodel_type, id=id_).first_or_404()
# Append download links
with opentracing.tracer.start_active_span(
'compileResponseDict',
tags={"id": id_}
):
details = self.aimodel_schema.dump(m)
# Vagy ez, vagy visszateszem a saját view-jébe és duplikálva lesz az egész
@ -75,18 +90,24 @@ class RootView(FlaskView):
try:
aimodel_type = AIModelType[type_]
except KeyError:
abort(404, "Unknown type")
return abort(404, "Unknown type")
try:
req = self.default_schema.load(request.json)
except ValidationError as e:
abort(400, str(e))
return abort(400, str(e))
with opentracing.tracer.start_active_span('sqlalchemy.select'):
m = AIModel.query.filter_by(type=aimodel_type, id=req['id']).first_or_404()
with opentracing.tracer.start_active_span('sqlalchemy.delete'):
Default.query.filter_by(type=aimodel_type).delete()
with opentracing.tracer.start_active_span('sqlalchemy.create'):
new_default = Default(type=aimodel_type, aimodel=m)
db.session.add(new_default)
with opentracing.tracer.start_active_span('sqlalchemy.commit'):
db.session.commit()
return '', 204

71
model_service/views/svm_view.py
View File

@ -9,6 +9,7 @@ from schemas import AIModelSchema, InfoSchema
from marshmallow.exceptions import ValidationError
from utils import storage, ensure_buckets, multipart_required
from pyAudioAnalysis.audioTrainTest import load_model
import opentracing
class SVMView(FlaskView):
@ -24,31 +25,34 @@ class SVMView(FlaskView):
def post(self):
# get important data from the request
with opentracing.tracer.start_active_span('parseAndValidate'):
try:
info = self.info_schema.loads(request.form.get('info'))
except ValidationError as e:
abort(400, str(e))
return abort(400, str(e))
# check for conflict
m = AIModel.query.filter_by(id=info['id']).first()
if m:
abort(409)
return abort(409)
# get and validate file
model_file = request.files['modelFile']
if model_file.content_length <= 0:
abort(411, f"Content length for modelFile is not a positive integer or missing.")
return abort(411, f"Content length for modelFile is not a positive integer or missing.")
means_file = request.files['meansFile']
if means_file.content_length <= 0:
abort(411, f"Content length for meansFile is not a positive integer or missing.")
return abort(411, f"Content length for meansFile is not a positive integer or missing.")
# create bucket if necessary
with opentracing.tracer.start_active_span('ensureBuckets'):
ensure_buckets()
# Temporarily save the file, because pyAudioAnalysis can only read files
with opentracing.tracer.start_active_span('tempfile.save'):
temp_model_handle, temp_model_filename = tempfile.mkstemp()
temp_means_filename = temp_model_filename + "MEANS"
@ -57,13 +61,14 @@ class SVMView(FlaskView):
model_file.save(temp_model_filename)
means_file.save(temp_means_filename)
with opentracing.tracer.start_active_span('pyAudioAnalysis.readModel'):
try:
_, _, _, classes, mid_window, mid_step, short_window, short_step, compute_beat \
= load_model(temp_model_filename)
if info['target_class_name'] not in classes:
abort(400, f"This model does not have a class named {info['target_class_name']}")
return abort(400, f"This model does not have a class named {info['target_class_name']}")
# Because of pyAudiomeme the files already saved, so we just use the file uploader functions
storage.connection.fput_object(
@ -82,7 +87,9 @@ class SVMView(FlaskView):
os.remove(temp_model_filename)
os.remove(temp_means_filename)
with opentracing.tracer.start_active_span('sqlalchemy.create'):
m = AIModel(id=info['id'], type=AIModelType.svm, target_class_name=info['target_class_name'])
db.session.add(m)
d = SVMDetails(
aimodel=m,
@ -92,25 +99,55 @@ class SVMView(FlaskView):
short_step=short_step,
compute_beat=compute_beat
)
db.session.add(m)
db.session.add(d)
with opentracing.tracer.start_active_span('sqlalchemy.commit'):
db.session.commit()
return jsonify(self.aimodel_schema.dump(m)), 200
def delete(self, id_: str):
with opentracing.tracer.start_active_span(
'sqlalchemy.select',
tags={"aimodel_type": AIModelType.svm, "id": id_}
):
if id_ == "$default":
default = Default.query.filter_by(type=AIModelType.svm).first_or_404()
m = default.aimodel
else:
m = AIModel.query.filter_by(type=AIModelType.svm, id=id_).first_or_404()
storage.connection.remove_object(current_app.config['MINIO_SVM_BUCKET_NAME'], self.MEANS_DIRECTORY + str(m.id))
storage.connection.remove_object(current_app.config['MINIO_SVM_BUCKET_NAME'], self.MODEL_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span('removeFromMinio'):
with opentracing.tracer.start_active_span(
'minio.removeObject',
tags={
"bucket": current_app.config['MINIO_SVM_BUCKET_NAME'],
"object_name": self.MEANS_DIRECTORY + str(m.id),
"component": "means"
}
):
storage.connection.remove_object(current_app.config['MINIO_SVM_BUCKET_NAME'],
self.MEANS_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span(
'minio.removeObject',
tags={
"bucket": current_app.config['MINIO_SVM_BUCKET_NAME'],
"object_name": self.MODEL_DIRECTORY + str(m.id),
"component": "model"
}
):
storage.connection.remove_object(current_app.config['MINIO_SVM_BUCKET_NAME'],
self.MODEL_DIRECTORY + str(m.id))
with opentracing.tracer.start_active_span(
'sqlalchemy.delete',
tags={"aimodel_type": AIModelType.svm, "id": id_}
):
db.session.delete(m)
with opentracing.tracer.start_active_span('sqlalchemy.commit'):
db.session.commit()
return '', 204
@ -119,6 +156,10 @@ class SVMView(FlaskView):
@route('<id_>/file')
def get_file(self, id_: str):
with opentracing.tracer.start_active_span(
'sqlalchemy.select',
tags={"aimodel_type": AIModelType.svm, "id": id_}
):
if id_ == "$default":
default = Default.query.filter_by(type=AIModelType.svm).first_or_404()
m = default.aimodel
@ -127,12 +168,22 @@ class SVMView(FlaskView):
if "means" in request.args:
path = self.MEANS_DIRECTORY + str(m.id)
component = "means"
else:
path = self.MODEL_DIRECTORY + str(m.id)
component = "model"
with opentracing.tracer.start_active_span(
'minio.getObject',
tags={
"bucket": current_app.config['MINIO_SVM_BUCKET_NAME'],
"object_name": path,
"component": component
}
):
try:
data = storage.connection.get_object(current_app.config['MINIO_SVM_BUCKET_NAME'], path)
except NoSuchKey:
abort(500, "The ID is stored in the database but not int the Object Store")
return abort(500, "The ID is stored in the database but not int the Object Store")
return Response(data.stream(), mimetype=data.headers['Content-type'])

10
requirements.txt
View File

@ -1,11 +1,11 @@
requests
blinker
Flask
marshmallow
Flask~=1.1.2
marshmallow~=3.7.1
Flask-Classful
Flask-SQLAlchemy
SQLAlchemy-Utils
SQLAlchemy
SQLAlchemy~=1.3.19
marshmallow-sqlalchemy~=0.26.1
marshmallow-enum
psycopg2-binary
@ -13,6 +13,7 @@ minio~=6.0.0
flask_minio
sentry-sdk
py-healthcheck
Werkzeug~=1.0.1
cython
@ -25,7 +26,7 @@ joblib==0.14.1
kiwisolver==1.2.0
matplotlib==3.2.1
numpy==1.18.2
pyAudioAnalysis==0.3.0
pyAudioAnalysis==0.3.6
pydub==0.23.1
pyparsing==2.4.6
python-dateutil==2.8.1
@ -36,3 +37,4 @@ six==1.14.0
jaeger-client
Flask-Opentracing
opentracing~=2.4.0