def new_application(app_name, fields, s3_bucket):
ok, message = fields_check(fields)
if not ok:
raise ArgsCheckError(message, "")
try:
# check application exist
if MongoIns.search_by_name(APPLICATION_COLLECTION_NAME, app_name):
raise ExistError(f"application <{app_name}> had exist", "")
except ExistError:
raise
try:
for _, value in fields.items():
if value.get("type") == "pipeline":
pipe = MongoIns.search_by_name(PIPELINE_COLLECTION_NAME,
value.get("value"))[0]
ei = identity(
pipe.get("encoder").get("instance").get("endpoint"))
name = f"{app_name}_{pipe.get('encoder').get('instance').get('name').replace('phantoscope_', '')}"
MilvusIns.new_milvus_collection(name, int(ei["dimension"]),
1024, "l2")
# create a application entity collection
MongoIns.new_mongo_collection(f"{app_name}_entity")
S3Ins.new_s3_buckets(s3_bucket)
# create milvus collections
app = Application(name=app_name, fields=fields, bucket=s3_bucket)
app.metadata = app._metadata()
MongoIns.insert_documents(APPLICATION_COLLECTION_NAME, app.to_dict())
return app
except Exception as e:
logger.error("error happen during create app: %s",
str(e),
exc_info=True)
raise e
def new_pipeline(name,
input,
index_file_size,
processors,
encoder,
description=None):
try:
encoder = operator_detail(encoder)
pipe = Pipeline(name=name,
input=input,
output=encoder.output,
dimension=encoder.dimension,
index_file_size=index_file_size,
metric_type=encoder.metric_type,
description=description,
processors=processors.split(","),
encoder=encoder.name)
if pipeline_ilegal(pipe):
return PipelineIlegalError("Pipeline ilegal check error", "")
milvus_collection_name = f"{name}_{encoder.name}"
MilvusIns.new_milvus_collection(milvus_collection_name,
encoder.dimension, index_file_size,
encoder.metric_type)
return pipe.save()
except Exception as e:
print(e)
logger.error(e)
return e
def delete_milvus_collections_by_fields(app):
for _, field in app['fields'].items():
if field["type"] == "pipeline":
pipe = MongoIns.search_by_name(PIPELINE_COLLECTION_NAME,
field.get("value"))[0]
name = f"{app.get('name')}_{pipe.get('encoder').get('instance').get('name').replace('phantoscope_', '')}"
MilvusIns.del_milvus_collection(name)
def delete_entity(app_name, entity_id):
try:
mongo_ins_name = f"{app_name}_entity"
entity = MongoIns.search_by_id(mongo_ins_name, entity_id)
if not entity.count():
raise NotExistError("Entity %s not exist" % entity_id,
"NotExistError")
for item in entity:
en = new_mapping_ins(item)
for name, fields in en._docs.items():
# delete s3 object
bucket_name = fields.get("url").split("/")[-2]
object_name = fields.get("url").split("/")[-1]
S3Ins.del_object(bucket_name, object_name)
# delete vector from milvus
vids = fields.get("ids")
app = application_detail(app_name)
pipe_name = app.fields[name]["value"]
pipe = pipeline_detail(pipe_name)
instance_name = pipe.encoder.get("instance")
MilvusIns.del_vectors(f"{app_name}_{name}_{instance_name}",
vids)
# delete from mongodb
MongoIns.delete_by_id(mongo_ins_name, entity_id)
logger.info("delete entity %s in application %s", entity_id,
app_name)
return en
except Exception as e:
logger.error(e)
raise e
def delete_milvus_collections_by_fields(app):
for _, field in app.fields.items():
if field["type"] == "pipeline":
pipe = pipeline_detail(field["value"])
name = pipe.encoder.get("name")
instance_name = pipe.encoder.get("instance")
MilvusIns.del_milvus_collection(
f"{app.name}_{name}_{instance_name}")
def create_milvus_collections_by_fields(app):
for field in search_fields(app.fields):
if field.type == "pipeline":
pipe = pipeline_detail(field.value)
name = pipe.encoder.get("name")
instance_name = pipe.encoder.get("instance")
encoder = operator_detail(name)
instance = encoder.inspect_instance(instance_name)
ei = identity(instance.endpoint)
MilvusIns.new_milvus_collection(
f"{app.name}_{name}_{instance_name}", int(ei["dimension"]),
1024, "l2")
def search(name, fields={}, topk=10, nprobe=16):
res = []
try:
app = application_detail(name)
accept_fields = [x for x, y in app.fields.items() if y.get('type') != "object"]
pipeline_fields = {x: y['pipeline'] for x, y in app.fields.items() if y.get('type') == "object"}
for k, _ in fields.items():
if k not in accept_fields and k not in pipeline_fields:
raise RequestError(f"fields {k} not in application", "")
for n, p in pipeline_fields.items():
pipe = pipeline_detail(p)
value = fields.get(n)
file_data = value.get('data')
url = value.get('url')
if not file_data and not url:
raise RequestError("can't find data or url from request", "")
vectors = run_pipeline(pipe, data=file_data, url=url)
if not vectors:
raise NoneVectorError("can't encode data by encoder, check input or encoder", "")
milvus_collection_name = f"{pipe.name}_{pipe.encoder}"
vids = MilvusIns.search_vectors(milvus_collection_name, vectors, topk=topk, nprobe=nprobe)
# here add scoreling function
dbs = search_ids_from_mapping([x.id for x in vids[0]])
for db in dbs:
m = new_mapping_ins(id=db.id, app_name=db.app_name,
image_url=db.image_url, fields=db.fields)
res.append(m)
return res
except Exception as e:
raise e
def upload(name, **kwargs):
try:
app = application_detail(name)
if not app:
raise NotExistError("application not exist", "application %s not exist" % name)
bucket_name = app.buckets.split(",")[0]
accept_fields = [x for x, y in app.fields.items() if y.get('type') != "object"]
pipeline_fields = {x: y['pipeline'] for x, y in app.fields.items() if y.get('type') == "object"}
new_fields = app.fields.copy()
for k, v in kwargs.items():
if k in accept_fields:
new_fields[k]['value'] = v
res = []
for k, _ in kwargs.get('fields').items():
if k not in accept_fields and k not in pipeline_fields:
raise RequestError(f"fields {k} not in application", "")
for n, p in pipeline_fields.items():
pipe = pipeline_detail(p)
if not pipe:
raise NotExistError("pipeline not exist", "pipeline %s not exist" % p)
value = kwargs['fields'].get(n)
file_data = value.get('data')
url = value.get('url')
if not file_data and not url:
raise RequestError("can't find data or url from request", "")
file_name = "{}-{}".format(name, uuid.uuid4().hex)
file_path = save_tmp_file(file_name, file_data, url)
# begin to timing
start = time.time()
S3Ins.upload2bucket(bucket_name, file_path, file_name)
upload_time = time.time()
logger.debug("[timing] upload image to bucket costs: {:.3f}s".format(upload_time - start))
vectors = run_pipeline(pipe, data=file_data, url=url)
pipeline_time = time.time()
logger.debug("[timing] run pipeline costs: {:.3f}s".format(pipeline_time - upload_time))
milvus_collection_name = f"{pipe.name}_{pipe.encoder}"
vids = MilvusIns.insert_vectors(milvus_collection_name, vectors)
insert_time = time.time()
logger.debug("[timing] insert to milvus costs: {:.3f}s".format(insert_time - pipeline_time))
for vid in vids:
m = DB(id=vid, app_name=name,
image_url=gen_url(bucket_name, file_name),
fields=new_fields)
add_mapping_data(m)
res.append(new_mapping_ins(id=vid, app_name=name,
image_url=gen_url(bucket_name, file_name),
fields=new_fields))
final_time = time.time()
logger.debug("[timing] prepare result costs: {:.3f}s".format(final_time - insert_time))
return res
except Exception as e:
print(e)
return e
def upload(name, **kwargs):
try:
app = application_detail(name)
if not app:
raise NotExistError("application not exist",
"application %s not exist" % name)
bucket_name = app.buckets.split(",")[0]
accept_fields = [
x for x, y in app.fields.items() if y.get('type') != "object"
]
pipeline_fields = {
x: y['pipeline']
for x, y in app.fields.items() if y.get('type') == "object"
}
new_fields = app.fields.copy()
for k, v in kwargs.items():
if k in accept_fields:
new_fields[k]['value'] = v
res = []
for k, _ in kwargs.get('fields').items():
if k not in accept_fields and k not in pipeline_fields:
raise RequestError(f"fields {k} not in application", "")
for n, p in pipeline_fields.items():
pipe = pipeline_detail(p)
if not pipe:
raise NotExistError("pipeline not exist",
"pipeline %s not exist" % p)
value = kwargs['fields'].get(n)
file_data = value.get('data')
url = value.get('url')
if not file_data and not url:
raise RequestError("can't find data or url from request", "")
file_name = "{}-{}".format(name, uuid.uuid4().hex)
file_path = save_tmp_file(file_name, file_data, url)
S3Ins.upload2bucket(bucket_name, file_path, file_name)
vectors = run_pipeline(pipe, data=file_data, url=url)
if not vectors:
raise NoneVectorError(
"can't encode data by encoder, check input or encoder", "")
milvus_collection_name = f"{pipe.name}_{pipe.encoder}"
vids = MilvusIns.insert_vectors(milvus_collection_name, vectors)
for vid in vids:
m = DB(id=vid,
app_name=name,
image_url=gen_url(bucket_name, file_name),
fields=new_fields)
add_mapping_data(m)
res.append(
new_mapping_ins(id=vid,
app_name=name,
image_url=gen_url(bucket_name, file_name),
fields=new_fields))
return res
except Exception as e:
print(e)
return e
def delete_entity(app_name, entity_name):
try:
entity = search_from_mapping(entity_name)
if not entity:
raise NotExistError("Entity %s not exist" % entity_name,
"NotExistError")
MilvusIns.del_vectors(app_name, [int(entity_name)])
bucket_name = entity.image_url.split("/")[-2]
object_name = entity.image_url.split("/")[-1]
S3Ins.del_object(bucket_name, object_name)
del_mapping(entity_name)
logger.info("delete entity %s in application %s", entity_name,
app_name)
return new_mapping_ins(id=entity.id,
app_name=entity.app_name,
image_url=entity.image_url,
fields=entity.fields)
except Exception as e:
logger.error(e)
return e
def upload(name, **kwargs):
try:
app = application_detail(name)
if not app:
raise NotExistError("application not exist", "application %s not exist" % name)
bucket_name = app.buckets.split(",")[0]
accept_fields = [x for x, y in app.fields.items() if y.get('type') != "pipeline"]
pipeline_fields = {x: y['value'] for x, y in app.fields.items() if y.get('type') == "pipeline"}
new_fields = app.fields.copy()
for k, v in kwargs.items():
if k in accept_fields:
new_fields[k]['value'] = v
res = []
for k, _ in kwargs.get('fields').items():
if k not in accept_fields and k not in pipeline_fields:
raise RequestError(f"fields {k} not in application", "")
docs = {}
valid_field_flag = False
for n, p in pipeline_fields.items():
pipe = pipeline_detail(p)
if not pipe:
raise NotExistError("pipeline not exist", "pipeline %s not exist" % p)
value = kwargs['fields'].get(n)
if not value:
continue
valid_field_flag = True
file_data = value.get('data')
url = value.get('url')
if not file_data and not url:
raise RequestError("can't find data or url from request", "")
file_name = "{}-{}".format(name, uuid.uuid4().hex)
file_path = save_tmp_file(file_name, file_data, url)
S3Ins.upload2bucket(bucket_name, file_path, file_name)
vectors = run_pipeline(pipe, data=file_data, url=url)
if not vectors:
raise NoneVectorError("can't encode data by encoder, check input or encoder", "")
milvus_collection_name = f"{app.name}_{pipe.encoder['name']}_{pipe.encoder['instance']}"
vids = MilvusIns.insert_vectors(milvus_collection_name, vectors)
docs[n] = {"ids": vids, "url": gen_url(bucket_name, file_name)}
doc_id = MongoIns.insert_documents(f"{app.name}_entity", docs)
res.append(new_mapping_ins(docs))
if not valid_field_flag:
raise RequestError("none valid field exist", "")
return res
except Exception as e:
err_msg = f"Unexpected error happen when upload: {str(e)}"
logger.error(err_msg, exc_info=True)
raise UnexpectedError(err_msg, e)
def search_and_score(milvus_collection_name, mongo_name, field_name, vectors,
topk, nprobe, inner_score_mode: str):
"""
search vectors from milvus and score by inner field score mode
:param milvus_collection_name: collection name will be search
:param mongo_name: mongo collection name will be selected from
:param field_name: field name for searching from mongodb
:param vectors: vectors which will be searched in milvus
:param topk: milvus topk number
:param nprobe: milvus nprobe number
:param inner_score_mode:
:return: image id of entity
"""
result_dbs = []
MAX_TOPK = 2048
magic_number = 60
increase_rate = 0.1
query_topk = topk + magic_number
end_flag = False
try:
inner_score_mode = InnerFieldScoreMode(inner_score_mode)
except Exception as e:
raise WrongInnerFieldModeError("Unsupported inner field mode", e)
while (len(result_dbs) < topk) and (not end_flag):
# check query topk max value
query_topk = min(query_topk, MAX_TOPK)
vids = MilvusIns.search_vectors(milvus_collection_name,
vectors,
topk=query_topk,
nprobe=nprobe)
if len(vids) == 0:
raise NoneVectorError("milvus search result is None", "")
# filter -1 and if exist -1 or len(vids) < topk
if (-1 in vids.id_array[0]) or len(vids[0]) < query_topk:
end_flag = True
# inner field score function here
res_vids = get_inner_field_score_result(vids, query_topk,
inner_score_mode)
if len(res_vids) < topk:
if query_topk < MAX_TOPK:
# calc a new query_topk and needn't to query from mysql
query_topk += math.ceil(query_topk * increase_rate)
increase_rate *= 2
if not end_flag:
continue
end_flag = True
result_dbs = MongoIns.search_by_vector_id(mongo_name, field_name,
res_vids)
# calc a new query_topk if len(result_dbs) < topk
query_topk += math.ceil(query_topk * increase_rate)
return result_dbs[:topk]
def delete_pipeline(name):
try:
p = del_pipeline(name)
if not p:
raise NotExistError("pipeline %s is not exist" % name, "")
p = p[0]
milvus_collection_name = f"{name}_{p.encoder}"
MilvusIns.del_milvus_collection(milvus_collection_name)
pipe = Pipeline(name=p.name,
input=p.input,
output=p.output,
dimension=p.dimension,
index_file_size=p.index_file_size,
metric_type=p.metric_type,
description=p.description,
processors=p.processors.split(","),
encoder=p.encoder)
return pipe
except Exception as e:
logger.error(e)
return e
def upload(name, **kwargs):
try:
app = application_detail(name)
if not app:
raise NotExistError("application not exist", "application %s not exist" % name)
bucket_name = app.buckets.split(",")[0]
accept_fields = [x for x, y in app.fields.items() if y.get('type') != "pipeline"]
pipeline_fields = {x: y['value'] for x, y in app.fields.items() if y.get('type') == "pipeline"}
new_fields = app.fields.copy()
for k, v in kwargs.items():
if k in accept_fields:
new_fields[k]['value'] = v
res = []
for k, _ in kwargs.get('fields').items():
if k not in accept_fields and k not in pipeline_fields:
raise RequestError(f"fields {k} not in application", "")
docs = {}
for n, p in pipeline_fields.items():
pipe = pipeline_detail(p)
if not pipe:
raise NotExistError("pipeline not exist", "pipeline %s not exist" % p)
value = kwargs['fields'].get(n)
file_data = value.get('data')
url = value.get('url')
if not file_data and not url:
raise RequestError("can't find data or url from request", "")
file_name = "{}-{}".format(name, uuid.uuid4().hex)
file_path = save_tmp_file(file_name, file_data, url)
S3Ins.upload2bucket(bucket_name, file_path, file_name)
vectors = run_pipeline(pipe, data=file_data, url=url)
milvus_collection_name = f"{app.name}_{pipe.encoder['name']}_{pipe.encoder['instance']}"
vids = MilvusIns.insert_vectors(milvus_collection_name, vectors)
docs[n] = {"ids": vids, "url": gen_url(bucket_name, file_name)}
doc_id = MongoIns.insert_documents(f"{app.name}_entity", docs)
res.append(new_mapping_ins(docs))
fields=new_fields))
return res
def test_new_collection(self):
"""create new collection"""
rv = MilvusIns.new_milvus_collection(self.name, self.dimension,
self.index_file_size,
self.metric_type)
assert rv == None
def test_insert_vectors(self):
"""test insert vectors"""
vectors = [[random.random() for _ in range(self.dimension)]
for _ in range(20)]
rv = MilvusIns.insert_vectors(self.name, vectors)
assert len(rv) == 20
def test_search_vectors(self):
"""test search vectors"""
q_records = [[random.random() for _ in range(self.dimension)]]
rv = MilvusIns.search_vectors(self.name, q_records, 10, 16)
def test_del_milvus_collection(self):
"""drop collection"""
rv = MilvusIns.del_milvus_collection(self.name)
assert rv == None
