def create_collection(self, dimension, data_type=DataType.FLOAT_VECTOR, auto_id=False,
collection_name=None, other_fields=None):
self._dimension = dimension
if not collection_name:
collection_name = self._collection_name
vec_field_name = utils.get_default_field_name(data_type)
fields = [
{"name": vec_field_name, "type": data_type, "params": {"dim": dimension}},
{"name": "id", "type": DataType.INT64, "is_primary": True}
]
if other_fields:
other_fields = other_fields.split(",")
for other_field_name in other_fields:
if other_field_name.startswith("int"):
field_type = DataType.INT64
elif other_field_name.startswith("float"):
field_type = DataType.FLOAT
elif other_field_name.startswith("double"):
field_type = DataType.DOUBLE
else:
raise Exception("Field name not supported")
fields.append({"name": other_field_name, "type": field_type})
create_param = {
"fields": fields,
"auto_id": auto_id}
try:
self._milvus.create_collection(collection_name, create_param)
logger.info("Create collection: <%s> successfully" % collection_name)
except Exception as e:
logger.error(str(e))
raise
def load_query_rand(self, nq_max=100):
# for ivf search
dimension = 128
top_k = random.randint(1, 100)
nq = random.randint(1, nq_max)
nprobe = random.randint(1, 100)
search_param = {"nprobe": nprobe}
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(nq)]
metric_type = random.choice(["l2", "ip"])
logger.info("%s, Search nq: %d, top_k: %d, nprobe: %d" % (self._collection_name, nq, top_k, nprobe))
vec_field_name = utils.get_default_field_name()
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors,
"metric_type": utils.metric_type_trans(metric_type),
"params": search_param}
}}
self.load_and_query(vector_query)
def create_collection(self,
dimension,
data_type=DataType.FLOAT_VECTOR,
auto_id=False,
collection_name=None,
other_fields=None):
self._dimension = dimension
if not collection_name:
collection_name = self._collection_name
vec_field_name = utils.get_default_field_name(data_type)
fields = [{
"name": vec_field_name,
"type": data_type,
"params": {
"dim": dimension
}
}]
if other_fields:
other_fields = other_fields.split(",")
if "int" in other_fields:
fields.append({
"name": utils.DEFAULT_INT_FIELD_NAME,
"type": DataType.INT64
})
if "float" in other_fields:
fields.append({
"name": utils.DEFAULT_FLOAT_FIELD_NAME,
"type": DataType.FLOAT
})
create_param = {"fields": fields, "auto_id": auto_id}
try:
self._milvus.create_collection(collection_name, create_param)
logger.info("Create collection: <%s> successfully" %
collection_name)
except Exception as e:
logger.error(str(e))
raise
def run(self, run_type, collection):
logger.debug(run_type)
logger.debug(collection)
collection_name = collection["collection_name"] if "collection_name" in collection else None
milvus_instance = MilvusClient(collection_name=collection_name, host=self.host)
# TODO: removed
# self.env_value = milvus_instance.get_server_config()
# ugly implemention
# self.env_value = utils.convert_nested(self.env_value)
# self.env_value.pop("logs")
# self.env_value.pop("network")
self.env_value = collection
if run_type == "insert_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
ni_per = collection["ni_per"]
build_index = collection["build_index"]
if milvus_instance.exists_collection():
milvus_instance.drop()
time.sleep(10)
index_info = {}
search_params = {}
vector_type = self.get_vector_type(data_type)
other_fields = collection["other_fields"] if "other_fields" in collection else None
milvus_instance.create_collection(dimension, data_type=vector_type,
other_fields=other_fields)
if build_index is True:
index_type = collection["index_type"]
index_param = collection["index_param"]
index_info = {
"index_type": index_type,
"index_param": index_param
}
index_field_name = utils.get_default_field_name(vector_type)
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
logger.debug(milvus_instance.describe_index())
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per)
flush_time = 0.0
if "flush" in collection and collection["flush"] == "no":
logger.debug("No manual flush")
else:
start_time = time.time()
milvus_instance.flush()
flush_time = time.time() - start_time
logger.debug(milvus_instance.count())
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name,
"other_fields": other_fields,
"ni_per": ni_per
}
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
search_params)
total_time = res["total_time"]
build_time = 0
if build_index is True:
logger.debug("Start build index for last file")
start_time = time.time()
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
build_time = time.time() - start_time
total_time = total_time + build_time
metric.metrics = {
"type": run_type,
"value": {
"total_time": total_time,
"qps": res["qps"],
"ni_time": res["ni_time"],
"flush_time": flush_time,
"build_time": build_time
}
}
report(metric)
elif run_type == "build_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
index_type = collection["index_type"]
index_param = collection["index_param"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
index_info = {
"index_type": index_type,
"index_param": index_param
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
search_params = {}
vector_type = self.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
start_time = time.time()
# drop index
logger.debug("Drop index")
milvus_instance.drop_index(index_field_name)
# start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
# TODO: need to check
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
logger.debug(milvus_instance.describe_index())
logger.debug(milvus_instance.count())
end_time = time.time()
# end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
search_params)
metric.metrics = {
"type": "build_performance",
"value": {
"build_time": round(end_time - start_time, 1),
}
}
report(metric)
elif run_type == "delete_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
ni_per = collection["ni_per"]
auto_flush = collection["auto_flush"] if "auto_flush" in collection else True
search_params = {}
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error(milvus_instance.show_collections())
logger.error("Table name: %s not existed" % collection_name)
return
length = milvus_instance.count()
logger.info(length)
index_info = milvus_instance.describe_index()
logger.info(index_info)
ids = [i for i in range(length)]
loops = int(length / ni_per)
milvus_instance.load_collection()
# TODO: remove
# start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
start_time = time.time()
# if auto_flush is False:
# milvus_instance.set_config("storage", "auto_flush_interval", BIG_FLUSH_INTERVAL)
for i in range(loops):
delete_ids = ids[i * ni_per: i * ni_per + ni_per]
logger.debug("Delete %d - %d" % (delete_ids[0], delete_ids[-1]))
milvus_instance.delete(delete_ids)
logger.debug("Table row counts: %d" % milvus_instance.count())
logger.debug("Table row counts: %d" % milvus_instance.count())
start_flush_time = time.time()
milvus_instance.flush()
end_flush_time = time.time()
end_time = time.time()
# end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.debug("Table row counts: %d" % milvus_instance.count())
# milvus_instance.set_config("storage", "auto_flush_interval", DEFAULT_FLUSH_INTERVAL)
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
search_params)
delete_time = round(end_time - start_time, 1)
metric.metrics = {
"type": "delete_performance",
"value": {
"delete_time": delete_time,
"qps": round(collection_size / delete_time, 1)
}
}
if auto_flush is False:
flush_time = round(end_flush_time - start_flush_time, 1)
metric.metrics["value"].update({"flush_time": flush_time})
report(metric)
elif run_type == "get_ids_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
ids_length_per_segment = collection["ids_length_per_segment"]
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
search_params = {}
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
for ids_num in ids_length_per_segment:
segment_num, get_ids = milvus_instance.get_rand_ids_each_segment(ids_num)
start_time = time.time()
get_res = milvus_instance.get_entities(get_ids)
total_time = time.time() - start_time
avg_time = total_time / segment_num
run_params = {"ids_num": ids_num}
logger.info(
"Segment num: %d, ids num per segment: %d, run_time: %f" % (segment_num, ids_num, total_time))
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info,
index_info, search_params, run_params=run_params)
metric.metrics = {
"type": run_type,
"value": {
"total_time": round(total_time, 1),
"avg_time": round(avg_time, 1)
}
}
report(metric)
elif run_type == "search_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
run_count = collection["run_count"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
# filter_query = collection["filter"] if "filter" in collection else None
filters = collection["filters"] if "filters" in collection else []
filter_query = []
search_params = collection["search_params"]
fields = self.get_fields(milvus_instance, collection_name)
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
"fields": fields
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
milvus_instance.load_collection()
logger.info("Start warm up query")
res = self.do_query(milvus_instance, collection_name, vec_field_name, [1], [1], 2,
search_param=search_params[0], filter_query=filter_query)
logger.info("End warm up query")
for search_param in search_params:
logger.info("Search param: %s" % json.dumps(search_param))
if not filters:
filters.append(None)
for filter in filters:
filter_param = []
if isinstance(filter, dict) and "range" in filter:
filter_query.append(eval(filter["range"]))
filter_param.append(filter["range"])
if isinstance(filter, dict) and "term" in filter:
filter_query.append(eval(filter["term"]))
filter_param.append(filter["term"])
logger.info("filter param: %s" % json.dumps(filter_param))
res = self.do_query(milvus_instance, collection_name, vec_field_name, top_ks, nqs, run_count,
search_param, filter_query=filter_query)
headers = ["Nq/Top-k"]
headers.extend([str(top_k) for top_k in top_ks])
logger.info("Search param: %s" % json.dumps(search_param))
utils.print_table(headers, nqs, res)
for index_nq, nq in enumerate(nqs):
for index_top_k, top_k in enumerate(top_ks):
search_param_group = {
"nq": nq,
"topk": top_k,
"search_param": search_param,
"filter": filter_param
}
search_time = res[index_nq][index_top_k]
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname,
collection_info, index_info, search_param_group)
metric.metrics = {
"type": "search_performance",
"value": {
"search_time": search_time
}
}
report(metric)
elif run_type == "locust_insert_stress":
pass
elif run_type in ["locust_search_performance", "locust_insert_performance", "locust_mix_performance"]:
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
ni_per = collection["ni_per"]
build_index = collection["build_index"]
if milvus_instance.exists_collection():
milvus_instance.drop()
time.sleep(10)
index_info = {}
search_params = {}
vector_type = self.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
milvus_instance.create_collection(dimension, data_type=vector_type, other_fields=None)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
if build_index is True:
index_type = collection["index_type"]
index_param = collection["index_param"]
index_info = {
"index_type": index_type,
"index_param": index_param
}
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
logger.debug(milvus_instance.describe_index())
if run_type in ["locust_search_performance", "locust_mix_performance"]:
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per)
if "flush" in collection and collection["flush"] == "no":
logger.debug("No manual flush")
else:
milvus_instance.flush()
if build_index is True:
logger.debug("Start build index for last file")
milvus_instance.create_index(index_field_name, index_type, metric_type, _async=True,
index_param=index_param)
logger.debug(milvus_instance.describe_index())
logger.debug("Table row counts: %d" % milvus_instance.count())
milvus_instance.load_collection()
logger.info("Start warm up query")
for i in range(2):
res = self.do_query(milvus_instance, collection_name, vec_field_name, [1], [1], 2,
search_param={"nprobe": 16})
logger.info("End warm up query")
real_metric_type = utils.metric_type_trans(metric_type)
### spawn locust requests
task = collection["task"]
connection_type = "single"
connection_num = task["connection_num"]
if connection_num > 1:
connection_type = "multi"
clients_num = task["clients_num"]
hatch_rate = task["hatch_rate"]
during_time = utils.timestr_to_int(task["during_time"])
task_types = task["types"]
run_params = {"tasks": {}, "clients_num": clients_num, "spawn_rate": hatch_rate, "during_time": during_time}
for task_type in task_types:
run_params["tasks"].update({task_type["type"]: task_type["weight"] if "weight" in task_type else 1})
# . collect stats
locust_stats = locust_user.locust_executor(self.host, self.port, collection_name,
connection_type=connection_type, run_params=run_params)
logger.info(locust_stats)
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
search_params)
metric.metrics = {
"type": run_type,
"value": locust_stats}
report(metric)
elif run_type == "search_ids_stability":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
search_params = collection["search_params"]
during_time = collection["during_time"]
ids_length = collection["ids_length"]
ids = collection["ids"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
g_top_k = int(collection["top_ks"].split("-")[1])
l_top_k = int(collection["top_ks"].split("-")[0])
g_id = int(ids.split("-")[1])
l_id = int(ids.split("-")[0])
g_id_length = int(ids_length.split("-")[1])
l_id_length = int(ids_length.split("-")[0])
milvus_instance.load_collection()
# start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
# logger.debug(start_mem_usage)
start_time = time.time()
while time.time() < start_time + during_time * 60:
search_param = {}
top_k = random.randint(l_top_k, g_top_k)
ids_num = random.randint(l_id_length, g_id_length)
ids_param = [random.randint(l_id_length, g_id_length) for _ in range(ids_num)]
for k, v in search_params.items():
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1]))
logger.debug("Query top-k: %d, ids_num: %d, param: %s" % (top_k, ids_num, json.dumps(search_param)))
result = milvus_instance.query_ids(top_k, ids_param, search_param=search_param)
# end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
{})
metric.metrics = {
"type": "search_ids_stability",
"value": {
"during_time": during_time,
}
}
report(metric)
# for sift/deep datasets
# TODO: enable
elif run_type == "accuracy":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
search_params = collection["search_params"]
# mapping to search param list
search_params = self.generate_combinations(search_params)
top_ks = collection["top_ks"]
nqs = collection["nqs"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
milvus_instance.load_collection()
true_ids_all = self.get_groundtruth_ids(collection_size)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
for search_param in search_params:
headers = ["Nq/Top-k"]
res = []
for nq in nqs:
for top_k in top_ks:
tmp_res = []
search_param_group = {
"nq": nq,
"topk": top_k,
"search_param": search_param,
"metric_type": metric_type
}
logger.info("Query params: %s" % json.dumps(search_param_group))
result_ids = self.do_query_ids(milvus_instance, collection_name, vec_field_name, top_k, nq,
search_param=search_param)
# mem_used = milvus_instance.get_mem_info()["memory_used"]
acc_value = self.get_recall_value(true_ids_all[:nq, :top_k].tolist(), result_ids)
logger.info("Query accuracy: %s" % acc_value)
tmp_res.append(acc_value)
# logger.info("Memory usage: %s" % mem_used)
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info,
index_info, search_param_group)
metric.metrics = {
"type": "accuracy",
"value": {
"acc": acc_value
}
}
report(metric)
# logger.info("Memory usage: %s" % mem_used)
res.append(tmp_res)
headers.extend([str(top_k) for top_k in top_ks])
logger.info("Search param: %s" % json.dumps(search_param))
utils.print_table(headers, nqs, res)
elif run_type == "ann_accuracy":
hdf5_source_file = collection["source_file"]
collection_name = collection["collection_name"]
index_types = collection["index_types"]
index_params = collection["index_params"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
search_params = collection["search_params"]
# mapping to search param list
search_params = self.generate_combinations(search_params)
# mapping to index param list
index_params = self.generate_combinations(index_params)
data_type, dimension, metric_type = parser.parse_ann_collection_name(collection_name)
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
dataset = utils.get_dataset(hdf5_source_file)
if milvus_instance.exists_collection(collection_name):
logger.info("Re-create collection: %s" % collection_name)
milvus_instance.drop()
time.sleep(DELETE_INTERVAL_TIME)
true_ids = np.array(dataset["neighbors"])
vector_type = self.get_vector_type_from_metric(metric_type)
vec_field_name = utils.get_default_field_name(vector_type)
real_metric_type = utils.metric_type_trans(metric_type)
# re-create collection
if milvus_instance.exists_collection(collection_name):
milvus_instance.drop()
time.sleep(DELETE_INTERVAL_TIME)
milvus_instance.create_collection(dimension, data_type=vector_type)
insert_vectors = self.normalize(metric_type, np.array(dataset["train"]))
if len(insert_vectors) != dataset["train"].shape[0]:
raise Exception("Row count of insert vectors: %d is not equal to dataset size: %d" % (
len(insert_vectors), dataset["train"].shape[0]))
logger.debug("The row count of entities to be inserted: %d" % len(insert_vectors))
# Insert batch once
# milvus_instance.insert(insert_vectors)
loops = len(insert_vectors) // INSERT_INTERVAL + 1
for i in range(loops):
start = i * INSERT_INTERVAL
end = min((i + 1) * INSERT_INTERVAL, len(insert_vectors))
if start < end:
tmp_vectors = insert_vectors[start:end]
ids = [i for i in range(start, end)]
if not isinstance(tmp_vectors, list):
entities = milvus_instance.generate_entities(tmp_vectors.tolist(), ids)
res_ids = milvus_instance.insert(entities, ids=ids)
else:
entities = milvus_instance.generate_entities(tmp_vectors, ids)
res_ids = milvus_instance.insert(entities, ids=ids)
assert res_ids == ids
milvus_instance.flush()
res_count = milvus_instance.count()
logger.info("Table: %s, row count: %d" % (collection_name, res_count))
if res_count != len(insert_vectors):
raise Exception("Table row count is not equal to insert vectors")
for index_type in index_types:
for index_param in index_params:
logger.debug("Building index with param: %s" % json.dumps(index_param))
if milvus_instance.get_config("cluster.enable") == "true":
milvus_instance.create_index(vec_field_name, index_type, metric_type, _async=True,
index_param=index_param)
else:
milvus_instance.create_index(vec_field_name, index_type, metric_type,
index_param=index_param)
logger.info(milvus_instance.describe_index())
logger.info("Start load collection: %s" % collection_name)
milvus_instance.load_collection()
logger.info("End load collection: %s" % collection_name)
index_info = {
"index_type": index_type,
"index_param": index_param
}
logger.debug(index_info)
warm_up = True
for search_param in search_params:
for nq in nqs:
query_vectors = self.normalize(metric_type, np.array(dataset["test"][:nq]))
if not isinstance(query_vectors, list):
query_vectors = query_vectors.tolist()
for top_k in top_ks:
search_param_group = {
"nq": len(query_vectors),
"topk": top_k,
"search_param": search_param,
"metric_type": metric_type
}
logger.debug(search_param_group)
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors,
"metric_type": real_metric_type,
"params": search_param}
}}
for i in range(2):
result = milvus_instance.query(vector_query)
warm_up = False
logger.info("End warm up")
result = milvus_instance.query(vector_query)
result_ids = milvus_instance.get_ids(result)
acc_value = self.get_recall_value(true_ids[:nq, :top_k].tolist(), result_ids)
logger.info("Query ann_accuracy: %s" % acc_value)
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname,
collection_info, index_info, search_param_group)
metric.metrics = {
"type": "ann_accuracy",
"value": {
"acc": acc_value
}
}
report(metric)
elif run_type == "search_stability":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
search_params = collection["search_params"]
during_time = collection["during_time"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
g_top_k = int(collection["top_ks"].split("-")[1])
g_nq = int(collection["nqs"].split("-")[1])
l_top_k = int(collection["top_ks"].split("-")[0])
l_nq = int(collection["nqs"].split("-")[0])
milvus_instance.load_collection()
# start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
# logger.debug(start_mem_usage)
start_row_count = milvus_instance.count()
logger.debug(milvus_instance.describe_index())
logger.info(start_row_count)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
real_metric_type = utils.metric_type_trans(metric_type)
start_time = time.time()
while time.time() < start_time + during_time * 60:
search_param = {}
top_k = random.randint(l_top_k, g_top_k)
nq = random.randint(l_nq, g_nq)
for k, v in search_params.items():
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1]))
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(nq)]
logger.debug("Query nq: %d, top-k: %d, param: %s" % (nq, top_k, json.dumps(search_param)))
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors[:nq],
"metric_type": real_metric_type,
"params": search_param}
}}
milvus_instance.query(vector_query)
# end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
{})
metric.metrics = {
"type": "search_stability",
"value": {
"during_time": during_time,
}
}
report(metric)
elif run_type == "loop_stability":
# init data
milvus_instance.clean_db()
pull_interval = collection["pull_interval"]
collection_num = collection["collection_num"]
concurrent = collection["concurrent"] if "concurrent" in collection else False
concurrent_num = collection_num
dimension = collection["dimension"] if "dimension" in collection else 128
insert_xb = collection["insert_xb"] if "insert_xb" in collection else 100000
index_types = collection["index_types"] if "index_types" in collection else ['ivf_sq8']
index_param = {"nlist": 256}
collection_names = []
milvus_instances_map = {}
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)]
ids = [i for i in range(insert_xb)]
# initialize and prepare
for i in range(collection_num):
name = utils.get_unique_name(prefix="collection_%d_" % i)
collection_names.append(name)
metric_type = random.choice(["l2", "ip"])
# default float_vector
milvus_instance = MilvusClient(collection_name=name, host=self.host)
milvus_instance.create_collection(dimension, other_fields=None)
index_type = random.choice(index_types)
field_name = utils.get_default_field_name()
milvus_instance.create_index(field_name, index_type, metric_type, index_param=index_param)
logger.info(milvus_instance.describe_index())
insert_vectors = utils.normalize(metric_type, insert_vectors)
entities = milvus_instance.generate_entities(insert_vectors, ids)
res_ids = milvus_instance.insert(entities, ids=ids)
milvus_instance.flush()
milvus_instances_map.update({name: milvus_instance})
logger.info(milvus_instance.describe_index())
# loop time unit: min -> s
pull_interval_seconds = pull_interval * 60
tasks = ["insert_rand", "query_rand", "flush"]
i = 1
while True:
logger.info("Loop time: %d" % i)
start_time = time.time()
while time.time() - start_time < pull_interval_seconds:
if concurrent:
threads = []
for name in collection_names:
task_name = random.choice(tasks)
task_run = getattr(milvus_instances_map[name], task_name)
t = threading.Thread(target=task_run, args=())
threads.append(t)
t.start()
for t in threads:
t.join()
# with concurrent.futures.ThreadPoolExecutor(max_workers=concurrent_num) as executor:
# future_results = {executor.submit(getattr(milvus_instances_map[mp[j][0]], mp[j][1])): j for j in range(concurrent_num)}
# for future in concurrent.futures.as_completed(future_results):
# future.result()
else:
tmp_collection_name = random.choice(collection_names)
task_name = random.choice(tasks)
logger.info(tmp_collection_name)
logger.info(task_name)
task_run = getattr(milvus_instances_map[tmp_collection_name], task_name)
task_run()
logger.debug("Restart server")
helm_utils.restart_server(self.service_name, namespace)
# new connection
# for name in collection_names:
# milvus_instance = MilvusClient(collection_name=name, host=self.host)
# milvus_instances_map.update({name: milvus_instance})
time.sleep(30)
i = i + 1
elif run_type == "stability":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
during_time = collection["during_time"]
operations = collection["operations"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error(milvus_instance.show_collections())
raise Exception("Table name: %s not existed" % collection_name)
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
# start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
start_row_count = milvus_instance.count()
logger.info(start_row_count)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
real_metric_type = utils.metric_type_trans(metric_type)
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(10000)]
if "insert" in operations:
insert_xb = operations["insert"]["xb"]
if "delete" in operations:
delete_xb = operations["delete"]["xb"]
if "query" in operations:
g_top_k = int(operations["query"]["top_ks"].split("-")[1])
l_top_k = int(operations["query"]["top_ks"].split("-")[0])
g_nq = int(operations["query"]["nqs"].split("-")[1])
l_nq = int(operations["query"]["nqs"].split("-")[0])
search_params = operations["query"]["search_params"]
i = 0
start_time = time.time()
while time.time() < start_time + during_time * 60:
i = i + 1
q = self.gen_executors(operations)
for name in q:
try:
if name == "insert":
insert_ids = random.sample(list(range(collection_size)), insert_xb)
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)]
entities = milvus_instance.generate_entities(insert_vectors, insert_ids)
milvus_instance.insert(entities, ids=insert_ids)
elif name == "delete":
delete_ids = random.sample(list(range(collection_size)), delete_xb)
milvus_instance.delete(delete_ids)
elif name == "query":
top_k = random.randint(l_top_k, g_top_k)
nq = random.randint(l_nq, g_nq)
search_param = {}
for k, v in search_params.items():
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1]))
logger.debug("Query nq: %d, top-k: %d, param: %s" % (nq, top_k, json.dumps(search_param)))
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors[:nq],
"metric_type": real_metric_type,
"params": search_param}
}}
result = milvus_instance.query(vector_query)
elif name in ["flush", "compact"]:
func = getattr(milvus_instance, name)
func()
logger.debug(milvus_instance.count())
except Exception as e:
logger.error(name)
logger.error(str(e))
raise
logger.debug("Loop time: %d" % i)
# end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
end_row_count = milvus_instance.count()
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info,
{})
metric.metrics = {
"type": "stability",
"value": {
"during_time": during_time,
"row_count_increments": end_row_count - start_row_count
}
}
report(metric)
elif run_type == "debug":
time.sleep(7200)
default_insert_vectors = [[random.random() for _ in range(128)] for _ in range(500000)]
interval = 50000
for loop in range(1, 7):
insert_xb = loop * interval
insert_vectors = default_insert_vectors[:insert_xb]
insert_ids = [i for i in range(insert_xb)]
entities = milvus_instance.generate_entities(insert_vectors, insert_ids)
for j in range(5):
milvus_instance.insert(entities, ids=insert_ids)
time.sleep(10)
else:
raise Exception("Run type not defined")
logger.debug("All test finished")
def run(self, run_type, collection):
logger.debug(run_type)
logger.debug(collection)
collection_name = collection["collection_name"] if "collection_name" in collection else None
milvus_instance = MilvusClient(collection_name=collection_name, host=self.host, port=self.port)
logger.info(milvus_instance.show_collections())
# TODO:
# self.env_value = milvus_instance.get_server_config()
# ugly implemention
# self.env_value = utils.convert_nested(self.env_value)
# self.env_value.pop("logs")
# self.env_value.pop("network")
# logger.info(self.env_value)
if run_type in ["insert_performance", "insert_flush_performance"]:
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
ni_per = collection["ni_per"]
build_index = collection["build_index"]
if milvus_instance.exists_collection():
milvus_instance.drop()
time.sleep(10)
vector_type = self.get_vector_type(data_type)
other_fields = collection["other_fields"] if "other_fields" in collection else None
milvus_instance.create_collection(dimension, data_type=vector_type, other_fields=other_fields)
if build_index is True:
index_type = collection["index_type"]
index_param = collection["index_param"]
index_field_name = utils.get_default_field_name(vector_type)
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per)
milvus_instance.flush()
logger.debug("Table row counts: %d" % milvus_instance.count())
if build_index is True:
logger.debug("Start build index for last file")
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
elif run_type == "delete_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
ni_per = collection["ni_per"]
auto_flush = collection["auto_flush"] if "auto_flush" in collection else True
if not milvus_instance.exists_collection():
logger.error(milvus_instance.show_collections())
logger.error("Table: %s not found" % collection_name)
return
length = milvus_instance.count()
ids = [i for i in range(length)]
loops = int(length / ni_per)
if auto_flush is False:
milvus_instance.set_config("storage", "auto_flush_interval", BIG_FLUSH_INTERVAL)
for i in range(loops):
delete_ids = ids[i*ni_per: i*ni_per+ni_per]
logger.debug("Delete %d - %d" % (delete_ids[0], delete_ids[-1]))
milvus_instance.delete(delete_ids)
logger.debug("Table row counts: %d" % milvus_instance.count())
logger.debug("Table row counts: %d" % milvus_instance.count())
milvus_instance.flush()
logger.debug("Table row counts: %d" % milvus_instance.count())
elif run_type == "build_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
index_type = collection["index_type"]
index_param = collection["index_param"]
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
vector_type = self.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
# drop index
logger.debug("Drop index")
milvus_instance.drop_index(index_field_name)
start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
start_time = time.time()
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
end_time = time.time()
logger.debug("Table row counts: %d" % milvus_instance.count())
end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.debug("Diff memory: %s, current memory usage: %s, build time: %s" % ((end_mem_usage - start_mem_usage), end_mem_usage, round(end_time - start_time, 1)))
elif run_type == "search_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
run_count = collection["run_count"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
search_params = collection["search_params"]
filter_query = []
filters = collection["filters"] if "filters" in collection else []
# pdb.set_trace()
# ranges = collection["range"] if "range" in collection else None
# terms = collection["term"] if "term" in collection else None
# if ranges:
# filter_query.append(eval(ranges))
# if terms:
# filter_query.append(eval(terms))
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
# for debugging
# time.sleep(3600)
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
logger.info(milvus_instance.count())
result = milvus_instance.describe_index()
logger.info(result)
milvus_instance.preload_collection()
mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.info(mem_usage)
for search_param in search_params:
logger.info("Search param: %s" % json.dumps(search_param))
filter_param = []
if not filters:
filters.append(None)
for filter in filters:
if isinstance(filter, dict) and "range" in filter:
filter_query.append(eval(filter["range"]))
filter_param.append(filter["range"])
if isinstance(filter, dict) and "term" in filter:
filter_query.append(eval(filter["term"]))
filter_param.append(filter["term"])
logger.info("filter param: %s" % json.dumps(filter_param))
res = self.do_query(milvus_instance, collection_name, vec_field_name, top_ks, nqs, run_count, search_param, filter_query)
headers = ["Nq/Top-k"]
headers.extend([str(top_k) for top_k in top_ks])
logger.info("Search param: %s" % json.dumps(search_param))
utils.print_table(headers, nqs, res)
mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.info(mem_usage)
elif run_type == "locust_search_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
ni_per = collection["ni_per"]
build_index = collection["build_index"]
vector_type = self.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
# if build_index is True:
# index_type = collection["index_type"]
# index_param = collection["index_param"]
# # TODO: debug
# if milvus_instance.exists_collection():
# milvus_instance.drop()
# time.sleep(10)
# other_fields = collection["other_fields"] if "other_fields" in collection else None
# milvus_instance.create_collection(dimension, data_type=vector_type, other_fields=other_fields)
# milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
# res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per)
# milvus_instance.flush()
# logger.debug("Table row counts: %d" % milvus_instance.count())
# if build_index is True:
# logger.debug("Start build index for last file")
# milvus_instance.create_index(index_field_name, index_type, metric_type, index_param=index_param)
real_metric_type = utils.metric_type_trans(metric_type)
### spawn locust requests
task = collection["task"]
connection_type = "single"
connection_num = task["connection_num"]
if connection_num > 1:
connection_type = "multi"
clients_num = task["clients_num"]
hatch_rate = task["hatch_rate"]
during_time = utils.timestr_to_int(task["during_time"])
task_types = task["types"]
# """
# task:
# connection_num: 1
# clients_num: 100
# hatch_rate: 2
# during_time: 5m
# types:
# -
# type: query
# weight: 1
# params:
# top_k: 10
# nq: 1
# # filters:
# # -
# # range:
# # int64:
# # LT: 0
# # GT: 1000000
# search_param:
# nprobe: 16
# """
run_params = {"tasks": {}, "clients_num": clients_num, "spawn_rate": hatch_rate, "during_time": during_time}
for task_type in task_types:
run_params["tasks"].update({task_type["type"]: task_type["weight"] if "weight" in task_type else 1})
#. collect stats
locust_stats = locust_user.locust_executor(self.host, self.port, collection_name, connection_type=connection_type, run_params=run_params)
logger.info(locust_stats)
elif run_type == "search_ids_stability":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
search_params = collection["search_params"]
during_time = collection["during_time"]
ids_length = collection["ids_length"]
ids = collection["ids"]
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
g_top_k = int(collection["top_ks"].split("-")[1])
l_top_k = int(collection["top_ks"].split("-")[0])
g_id = int(ids.split("-")[1])
l_id = int(ids.split("-")[0])
g_id_length = int(ids_length.split("-")[1])
l_id_length = int(ids_length.split("-")[0])
milvus_instance.preload_collection()
start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.debug(start_mem_usage)
start_time = time.time()
while time.time() < start_time + during_time * 60:
search_param = {}
top_k = random.randint(l_top_k, g_top_k)
ids_num = random.randint(l_id_length, g_id_length)
l_ids = random.randint(l_id, g_id-ids_num)
# ids_param = [random.randint(l_id_length, g_id_length) for _ in range(ids_num)]
ids_param = [id for id in range(l_ids, l_ids+ids_num)]
for k, v in search_params.items():
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1]))
logger.debug("Query top-k: %d, ids_num: %d, param: %s" % (top_k, ids_num, json.dumps(search_param)))
result = milvus_instance.query_ids(top_k, ids_param, search_param=search_param)
end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
metrics = {
"during_time": during_time,
"start_mem_usage": start_mem_usage,
"end_mem_usage": end_mem_usage,
"diff_mem": end_mem_usage - start_mem_usage,
}
logger.info(metrics)
elif run_type == "search_performance_concurrents":
data_type, dimension, metric_type = parser.parse_ann_collection_name(collection_name)
hdf5_source_file = collection["source_file"]
use_single_connection = collection["use_single_connection"]
concurrents = collection["concurrents"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
search_params = self.generate_combinations(collection["search_params"])
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
logger.info(milvus_instance.count())
result = milvus_instance.describe_index()
logger.info(result)
milvus_instance.preload_collection()
dataset = utils.get_dataset(hdf5_source_file)
for concurrent_num in concurrents:
top_k = top_ks[0]
for nq in nqs:
mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.info(mem_usage)
query_vectors = self.normalize(metric_type, np.array(dataset["test"][:nq]))
logger.debug(search_params)
for search_param in search_params:
logger.info("Search param: %s" % json.dumps(search_param))
total_time = 0.0
if use_single_connection is True:
connections = [MilvusClient(collection_name=collection_name, host=self.host, port=self.port)]
with concurrent.futures.ThreadPoolExecutor(max_workers=concurrent_num) as executor:
future_results = {executor.submit(
self.do_query_qps, connections[0], query_vectors, top_k, search_param=search_param) : index for index in range(concurrent_num)}
else:
connections = [MilvusClient(collection_name=collection_name, host=self.hos, port=self.port) for i in range(concurrent_num)]
with concurrent.futures.ThreadPoolExecutor(max_workers=concurrent_num) as executor:
future_results = {executor.submit(
self.do_query_qps, connections[index], query_vectors, top_k, search_param=search_param) : index for index in range(concurrent_num)}
for future in concurrent.futures.as_completed(future_results):
total_time = total_time + future.result()
qps_value = total_time / concurrent_num
logger.debug("QPS value: %f, total_time: %f, request_nums: %f" % (qps_value, total_time, concurrent_num))
mem_usage = milvus_instance.get_mem_info()["memory_used"]
logger.info(mem_usage)
elif run_type == "ann_accuracy":
hdf5_source_file = collection["source_file"]
collection_name = collection["collection_name"]
index_types = collection["index_types"]
index_params = collection["index_params"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
search_params = collection["search_params"]
# mapping to search param list
search_params = self.generate_combinations(search_params)
# mapping to index param list
index_params = self.generate_combinations(index_params)
data_type, dimension, metric_type = parser.parse_ann_collection_name(collection_name)
dataset = utils.get_dataset(hdf5_source_file)
true_ids = np.array(dataset["neighbors"])
vector_type = self.get_vector_type_from_metric(metric_type)
vec_field_name = utils.get_default_field_name(vector_type)
real_metric_type = utils.metric_type_trans(metric_type)
# re-create collection
if milvus_instance.exists_collection(collection_name):
milvus_instance.drop()
time.sleep(DELETE_INTERVAL_TIME)
milvus_instance.create_collection(dimension, data_type=vector_type)
insert_vectors = self.normalize(metric_type, np.array(dataset["train"]))
if len(insert_vectors) != dataset["train"].shape[0]:
raise Exception("Row count of insert vectors: %d is not equal to dataset size: %d" % (len(insert_vectors), dataset["train"].shape[0]))
logger.debug("The row count of entities to be inserted: %d" % len(insert_vectors))
# insert batch once
# milvus_instance.insert(insert_vectors)
loops = len(insert_vectors) // INSERT_INTERVAL + 1
for i in range(loops):
start = i*INSERT_INTERVAL
end = min((i+1)*INSERT_INTERVAL, len(insert_vectors))
if start < end:
tmp_vectors = insert_vectors[start:end]
ids = [i for i in range(start, end)]
if not isinstance(tmp_vectors, list):
entities = milvus_instance.generate_entities(tmp_vectors.tolist(), ids)
res_ids = milvus_instance.insert(entities, ids=ids)
else:
entities = milvus_instance.generate_entities(tmp_vectors, ids)
res_ids = milvus_instance.insert(entities, ids=ids)
assert res_ids == ids
milvus_instance.flush()
res_count = milvus_instance.count()
logger.info("Table: %s, row count: %d" % (collection_name, res_count))
if res_count != len(insert_vectors):
raise Exception("Table row count is not equal to insert vectors")
for index_type in index_types:
for index_param in index_params:
logger.debug("Building index with param: %s, metric_type: %s" % (json.dumps(index_param), metric_type))
milvus_instance.create_index(vec_field_name, index_type, metric_type, index_param=index_param)
logger.info("Start preload collection: %s" % collection_name)
milvus_instance.preload_collection()
for search_param in search_params:
for nq in nqs:
query_vectors = self.normalize(metric_type, np.array(dataset["test"][:nq]))
if not isinstance(query_vectors, list):
query_vectors = query_vectors.tolist()
for top_k in top_ks:
logger.debug("Search nq: %d, top-k: %d, search_param: %s, metric_type: %s" % (nq, top_k, json.dumps(search_param), metric_type))
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors,
"metric_type": real_metric_type,
"params": search_param}
}}
result = milvus_instance.query(vector_query)
result_ids = milvus_instance.get_ids(result)
# pdb.set_trace()
acc_value = self.get_recall_value(true_ids[:nq, :top_k].tolist(), result_ids)
logger.info("Query ann_accuracy: %s" % acc_value)
elif run_type == "accuracy":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
search_params = collection["search_params"]
# mapping to search param list
search_params = self.generate_combinations(search_params)
top_ks = collection["top_ks"]
nqs = collection["nqs"]
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
if not milvus_instance.exists_collection():
logger.error("Table name: %s not existed" % collection_name)
return
logger.info(milvus_instance.count())
index_info = milvus_instance.describe_index()
logger.info(index_info)
milvus_instance.preload_collection()
true_ids_all = self.get_groundtruth_ids(collection_size)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
for search_param in search_params:
headers = ["Nq/Top-k"]
res = []
for nq in nqs:
tmp_res = []
for top_k in top_ks:
search_param_group = {
"nq": nq,
"topk": top_k,
"search_param": search_param,
"metric_type": metric_type
}
logger.info("Query params: %s" % json.dumps(search_param_group))
result_ids = self.do_query_ids(milvus_instance, collection_name, vec_field_name, top_k, nq, search_param=search_param)
mem_used = milvus_instance.get_mem_info()["memory_used"]
acc_value = self.get_recall_value(true_ids_all[:nq, :top_k].tolist(), result_ids)
logger.info("Query accuracy: %s" % acc_value)
tmp_res.append(acc_value)
logger.info("Memory usage: %s" % mem_used)
res.append(tmp_res)
headers.extend([str(top_k) for top_k in top_ks])
logger.info("Search param: %s" % json.dumps(search_param))
utils.print_table(headers, nqs, res)
elif run_type == "stability":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(collection_name)
during_time = collection["during_time"]
operations = collection["operations"]
if not milvus_instance.exists_collection():
logger.error(milvus_instance.show_collections())
raise Exception("Table name: %s not existed" % collection_name)
milvus_instance.preload_collection()
start_mem_usage = milvus_instance.get_mem_info()["memory_used"]
start_row_count = milvus_instance.count()
logger.info(start_row_count)
vector_type = self.get_vector_type(data_type)
vec_field_name = utils.get_default_field_name(vector_type)
real_metric_type = utils.metric_type_trans(metric_type)
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(10000)]
if "insert" in operations:
insert_xb = operations["insert"]["xb"]
if "delete" in operations:
delete_xb = operations["delete"]["xb"]
if "query" in operations:
g_top_k = int(operations["query"]["top_ks"].split("-")[1])
l_top_k = int(operations["query"]["top_ks"].split("-")[0])
g_nq = int(operations["query"]["nqs"].split("-")[1])
l_nq = int(operations["query"]["nqs"].split("-")[0])
search_params = operations["query"]["search_params"]
i = 0
start_time = time.time()
while time.time() < start_time + during_time * 60:
i = i + 1
q = self.gen_executors(operations)
for name in q:
try:
if name == "insert":
insert_ids = random.sample(list(range(collection_size)), insert_xb)
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)]
entities = milvus_instance.generate_entities(insert_vectors, insert_ids)
milvus_instance.insert(entities, ids=insert_ids)
elif name == "delete":
delete_ids = random.sample(list(range(collection_size)), delete_xb)
milvus_instance.delete(delete_ids)
elif name == "query":
top_k = random.randint(l_top_k, g_top_k)
nq = random.randint(l_nq, g_nq)
search_param = {}
for k, v in search_params.items():
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1]))
logger.debug("Query nq: %d, top-k: %d, param: %s" % (nq, top_k, json.dumps(search_param)))
vector_query = {"vector": {vec_field_name: {
"topk": top_k,
"query": query_vectors[:nq],
"metric_type": real_metric_type,
"params": search_param}
}}
result = milvus_instance.query(vector_query)
elif name in ["flush", "compact"]:
func = getattr(milvus_instance, name)
func()
logger.debug(milvus_instance.count())
except Exception as e:
logger.error(name)
logger.error(str(e))
raise
logger.debug("Loop time: %d" % i)
end_mem_usage = milvus_instance.get_mem_info()["memory_used"]
end_row_count = milvus_instance.count()
metrics = {
"during_time": during_time,
"start_mem_usage": start_mem_usage,
"end_mem_usage": end_mem_usage,
"diff_mem": end_mem_usage - start_mem_usage,
"row_count_increments": end_row_count - start_row_count
}
logger.info(metrics)
elif run_type == "loop_stability":
# init data
milvus_instance.clean_db()
pull_interval = collection["pull_interval"]
collection_num = collection["collection_num"]
concurrent = collection["concurrent"] if "concurrent" in collection else False
concurrent_num = collection_num
dimension = collection["dimension"] if "dimension" in collection else 128
insert_xb = collection["insert_xb"] if "insert_xb" in collection else 100000
index_types = collection["index_types"] if "index_types" in collection else ['ivf_sq8']
index_param = {"nlist": 256}
collection_names = []
milvus_instances_map = {}
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)]
ids = [i for i in range(insert_xb)]
# initialize and prepare
for i in range(collection_num):
name = utils.get_unique_name(prefix="collection_%d_" % i)
collection_names.append(name)
metric_type = random.choice(["l2", "ip"])
# default float_vector
milvus_instance = MilvusClient(collection_name=name, host=self.host)
milvus_instance.create_collection(dimension, other_fields=None)
index_type = random.choice(index_types)
field_name = utils.get_default_field_name()
milvus_instance.create_index(field_name, index_type, metric_type, index_param=index_param)
logger.info(milvus_instance.describe_index())
insert_vectors = utils.normalize(metric_type, insert_vectors)
entities = milvus_instance.generate_entities(insert_vectors, ids)
res_ids = milvus_instance.insert(entities, ids=ids)
milvus_instance.flush()
milvus_instances_map.update({name: milvus_instance})
logger.info(milvus_instance.describe_index())
# loop time unit: min -> s
pull_interval_seconds = pull_interval * 60
tasks = ["insert_rand", "delete_rand", "query_rand", "flush", "compact"]
i = 1
while True:
logger.info("Loop time: %d" % i)
start_time = time.time()
while time.time() - start_time < pull_interval_seconds:
if concurrent:
threads = []
for name in collection_names:
task_name = random.choice(tasks)
task_run = getattr(milvus_instances_map[name], task_name)
t = threading.Thread(target=task_run, args=())
threads.append(t)
t.start()
for t in threads:
t.join()
# with concurrent.futures.ThreadPoolExecutor(max_workers=concurrent_num) as executor:
# future_results = {executor.submit(getattr(milvus_instances_map[mp[j][0]], mp[j][1])): j for j in range(concurrent_num)}
# for future in concurrent.futures.as_completed(future_results):
# future.result()
else:
tmp_collection_name = random.choice(collection_names)
task_name = random.choice(tasks)
logger.info(tmp_collection_name)
logger.info(task_name)
task_run = getattr(milvus_instances_map[tmp_collection_name], task_name)
task_run()
# new connection
# for name in collection_names:
# milvus_instance = MilvusClient(collection_name=name, host=self.host)
# milvus_instances_map.update({name: milvus_instance})
i = i + 1
elif run_type == "locust_mix_performance":
(data_type, collection_size, dimension, metric_type) = parser.collection_parser(
collection_name)
ni_per = collection["ni_per"]
build_index = collection["build_index"]
vector_type = self.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
# drop exists collection
if milvus_instance.exists_collection():
milvus_instance.drop()
time.sleep(10)
# create collection
other_fields = collection["other_fields"] if "other_fields" in collection else None
milvus_instance.create_collection(dimension, data_type=DataType.FLOAT_VECTOR, collection_name=collection_name, other_fields=other_fields)
logger.info(milvus_instance.get_info())
# insert entities
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(ni_per)]
insert_ids = random.sample(list(range(collection_size)), ni_per)
insert_vectors = utils.normalize(metric_type, insert_vectors)
entities = milvus_instance.generate_entities(insert_vectors, insert_ids, collection_name)
milvus_instance.insert(entities, ids=insert_ids)
# flush
milvus_instance.flush()
logger.info(milvus_instance.get_stats())
logger.debug("Table row counts: %d" % milvus_instance.count())
# create index
if build_index is True:
index_type = collection["index_type"]
index_param = collection["index_param"]
logger.debug("Start build index for last file")
milvus_instance.create_index(index_field_name, index_type, metric_type, index_param)
logger.debug(milvus_instance.describe_index())
# locust
task = collection["tasks"]
task_file = utils.get_unique_name()
task_file_script = task_file + '.py'
task_file_csv = task_file + '_stats.csv'
task_types = task["types"]
connection_type = "single"
connection_num = task["connection_num"]
if connection_num > 1:
connection_type = "multi"
clients_num = task["clients_num"]
hatch_rate = task["hatch_rate"]
during_time = task["during_time"]
def_strs = ""
# define def str
for task_type in task_types:
type = task_type["type"]
weight = task_type["weight"]
if type == "flush":
def_str = """
@task(%d)
def flush(self):
client = get_client(collection_name)
client.flush(collection_name=collection_name)
""" % weight
if type == "compact":
def_str = """
@task(%d)
def compact(self):
client = get_client(collection_name)
client.compact(collection_name)
""" % weight
if type == "query":
def_str = """
@task(%d)
def query(self):
client = get_client(collection_name)
params = %s
X = [[random.random() for i in range(dim)] for i in range(params["nq"])]
vector_query = {"vector": {"%s": {
"topk": params["top_k"],
"query": X,
"metric_type": "%s",
"params": params["search_param"]}}}
client.query(vector_query, filter_query=params["filters"], collection_name=collection_name)
""" % (weight, task_type["params"], index_field_name, utils.metric_type_trans(metric_type))
if type == "insert":
def_str = """
@task(%d)
def insert(self):
client = get_client(collection_name)
params = %s
insert_ids = random.sample(list(range(100000)), params["nb"])
insert_vectors = [[random.random() for _ in range(dim)] for _ in range(params["nb"])]
insert_vectors = utils.normalize("l2", insert_vectors)
entities = generate_entities(insert_vectors, insert_ids)
client.insert(entities,ids=insert_ids, collection_name=collection_name)
""" % (weight, task_type["params"])
if type == "delete":
def_str = """
@task(%d)
def delete(self):
client = get_client(collection_name)
ids = [random.randint(1, 1000000) for i in range(1)]
client.delete(ids, collection_name)
""" % weight
def_strs += def_str
print(def_strs)
# define locust code str
code_str = """
import random
import json
from locust import User, task, between
from locust_task import MilvusTask
from client import MilvusClient
import utils
host = '%s'
port = %s
collection_name = '%s'
dim = %s
connection_type = '%s'
m = MilvusClient(host=host, port=port)
def get_client(collection_name):
if connection_type == 'single':
return MilvusTask(m=m)
elif connection_type == 'multi':
return MilvusTask(connection_type='multi', host=host, port=port, collection_name=collection_name)
def generate_entities(vectors, ids):
return m.generate_entities(vectors, ids, collection_name)
class MixTask(User):
wait_time = between(0.001, 0.002)
%s
""" % (self.host, self.port, collection_name, dimension, connection_type, def_strs)
with open(task_file_script, "w+") as fd:
fd.write(code_str)
locust_cmd = "locust -f %s --headless --csv=%s -u %d -r %d -t %s" % (
task_file_script,
task_file,
clients_num,
hatch_rate,
during_time)
logger.info(locust_cmd)
try:
res = os.system(locust_cmd)
except Exception as e:
logger.error(str(e))
return
# . retrieve and collect test statistics
metric = None
with open(task_file_csv, newline='') as fd:
dr = csv.DictReader(fd)
for row in dr:
if row["Name"] != "Aggregated":
continue
metric = row
logger.info(metric)
else:
raise Exception("Run type not defined")
logger.debug("All test finished")
