mirror of
https://gitee.com/milvus-io/milvus.git
synced 2024-12-05 05:18:52 +08:00
477 lines
20 KiB
Python
477 lines
20 KiB
Python
import time
|
||
import os
|
||
import numpy as np
|
||
import random
|
||
from sklearn import preprocessing
|
||
from common.common_func import gen_unique_str
|
||
from minio_comm import copy_files_to_minio
|
||
|
||
# TODO: remove hardcode with input configurations
|
||
minio = "minio_address:port" # minio service and port
|
||
bucket_name = "milvus-bulk-load" # bucket name of milvus is using
|
||
|
||
data_source = "/tmp/bulk_load_data"
|
||
|
||
BINARY = "binary"
|
||
FLOAT = "float"
|
||
|
||
|
||
class DataField:
|
||
pk_field = "uid"
|
||
vec_field = "vectors"
|
||
int_field = "int_scalar"
|
||
string_field = "string_scalar"
|
||
bool_field = "bool_scalar"
|
||
float_field = "float_scalar"
|
||
|
||
|
||
class DataErrorType:
|
||
one_entity_wrong_dim = "one_entity_wrong_dim"
|
||
str_on_int_pk = "str_on_int_pk"
|
||
int_on_float_scalar = "int_on_float_scalar"
|
||
float_on_int_pk = "float_on_int_pk"
|
||
typo_on_bool = "typo_on_bool"
|
||
str_on_float_scalar = "str_on_float_scalar"
|
||
str_on_vector_field = "str_on_vector_field"
|
||
|
||
|
||
def gen_file_prefix(row_based=True, auto_id=True, prefix=""):
|
||
if row_based:
|
||
if auto_id:
|
||
return f"{prefix}_row_auto"
|
||
else:
|
||
return f"{prefix}_row_cust"
|
||
else:
|
||
if auto_id:
|
||
return f"{prefix}_col_auto"
|
||
else:
|
||
return f"{prefix}_col_cust"
|
||
|
||
|
||
def entity_suffix(rows):
|
||
if rows // 1000000 > 0:
|
||
suffix = f"{rows // 1000000}m"
|
||
elif rows // 1000 > 0:
|
||
suffix = f"{rows // 1000}k"
|
||
else:
|
||
suffix = f"{rows}"
|
||
return suffix
|
||
|
||
|
||
def gen_float_vectors(nb, dim):
|
||
vectors = [[random.random() for _ in range(dim)] for _ in range(nb)]
|
||
vectors = preprocessing.normalize(vectors, axis=1, norm='l2')
|
||
return vectors.tolist()
|
||
|
||
|
||
def gen_str_invalid_vectors(nb, dim):
|
||
vectors = [[str(gen_unique_str()) for _ in range(dim)] for _ in range(nb)]
|
||
return vectors
|
||
|
||
|
||
def gen_binary_vectors(nb, dim):
|
||
# binary: each int presents 8 dimension
|
||
# so if binary vector dimension is 16，use [x, y], which x and y could be any int between 0 to 255
|
||
vectors = [[random.randint(0, 255) for _ in range(dim)] for _ in range(nb)]
|
||
return vectors
|
||
|
||
|
||
def gen_row_based_json_file(row_file, str_pk, data_fields, float_vect,
|
||
rows, dim, start_uid=0, err_type="", **kwargs):
|
||
|
||
if err_type == DataErrorType.str_on_int_pk:
|
||
str_pk = True
|
||
if err_type in [DataErrorType.one_entity_wrong_dim, DataErrorType.str_on_vector_field]:
|
||
wrong_dim = dim + 8 # add 8 to compatible with binary vectors
|
||
wrong_row = kwargs.get("wrong_position", start_uid)
|
||
|
||
with open(row_file, "w") as f:
|
||
f.write("{")
|
||
f.write("\n")
|
||
f.write('"rows":[')
|
||
f.write("\n")
|
||
for i in range(rows):
|
||
if i > 0:
|
||
f.write(",")
|
||
f.write("\n")
|
||
|
||
# scalar fields
|
||
f.write('{')
|
||
for j in range(len(data_fields)):
|
||
data_field = data_fields[j]
|
||
if data_field == DataField.pk_field:
|
||
if str_pk:
|
||
f.write('"uid":"' + str(gen_unique_str()) + '"')
|
||
else:
|
||
if err_type == DataErrorType.float_on_int_pk:
|
||
f.write('"uid":' + str(i + start_uid + random.random()) + '')
|
||
else:
|
||
f.write('"uid":' + str(i + start_uid) + '')
|
||
if data_field == DataField.int_field:
|
||
if DataField.pk_field in data_fields:
|
||
# if not auto_id, use the same value as pk to check the query results later
|
||
f.write('"int_scalar":' + str(i + start_uid) + '')
|
||
else:
|
||
f.write('"int_scalar":' + str(random.randint(-999999, 9999999)) + '')
|
||
if data_field == DataField.float_field:
|
||
if err_type == DataErrorType.int_on_float_scalar:
|
||
f.write('"float_scalar":' + str(random.randint(-999999, 9999999)) + '')
|
||
elif err_type == DataErrorType.str_on_float_scalar:
|
||
f.write('"float_scalar":"' + str(gen_unique_str()) + '"')
|
||
else:
|
||
f.write('"float_scalar":' + str(random.random()) + '')
|
||
if data_field == DataField.string_field:
|
||
f.write('"string_scalar":"' + str(gen_unique_str()) + '"')
|
||
if data_field == DataField.bool_field:
|
||
if err_type == DataErrorType.typo_on_bool:
|
||
f.write('"bool_scalar":' + str(random.choice(["True", "False", "TRUE", "FALSE", "0", "1"])) + '')
|
||
else:
|
||
f.write('"bool_scalar":' + str(random.choice(["true", "false"])) + '')
|
||
if data_field == DataField.vec_field:
|
||
# vector field
|
||
if err_type == DataErrorType.one_entity_wrong_dim and i == wrong_row:
|
||
vectors = gen_float_vectors(1, wrong_dim) if float_vect else gen_binary_vectors(1, (wrong_dim//8))
|
||
elif err_type == DataErrorType.str_on_vector_field and i == wrong_row:
|
||
vectors = gen_str_invalid_vectors(1, dim) if float_vect else gen_str_invalid_vectors(1, dim//8)
|
||
else:
|
||
vectors = gen_float_vectors(1, dim) if float_vect else gen_binary_vectors(1, (dim//8))
|
||
f.write('"vectors":' + ",".join(str(x).replace("'", '"') for x in vectors) + '')
|
||
# not write common for the last field
|
||
if j != len(data_fields) - 1:
|
||
f.write(',')
|
||
f.write('}')
|
||
f.write("\n")
|
||
f.write("]")
|
||
f.write("\n")
|
||
f.write("}")
|
||
f.write("\n")
|
||
|
||
|
||
def gen_column_base_json_file(col_file, str_pk, data_fields, float_vect,
|
||
rows, dim, start_uid=0, err_type="", **kwargs):
|
||
if err_type == DataErrorType.str_on_int_pk:
|
||
str_pk = True
|
||
|
||
with open(col_file, "w") as f:
|
||
f.write("{")
|
||
f.write("\n")
|
||
if rows > 0:
|
||
# data columns
|
||
for j in range(len(data_fields)):
|
||
data_field = data_fields[j]
|
||
if data_field == DataField.pk_field:
|
||
if str_pk:
|
||
f.write('"uid":["' + ',"'.join(str(gen_unique_str()) + '"' for i in range(rows)) + ']')
|
||
f.write("\n")
|
||
else:
|
||
if err_type == DataErrorType.float_on_int_pk:
|
||
f.write('"uid":[' + ",".join(
|
||
str(i + random.random()) for i in range(start_uid, start_uid + rows)) + "]")
|
||
else:
|
||
f.write('"uid":[' + ",".join(str(i) for i in range(start_uid, start_uid + rows)) + "]")
|
||
f.write("\n")
|
||
if data_field == DataField.int_field:
|
||
if DataField.pk_field in data_fields:
|
||
# if not auto_id, use the same value as pk to check the query results later
|
||
f.write('"int_scalar":[' + ",".join(str(i) for i in range(start_uid, start_uid + rows)) + "]")
|
||
else:
|
||
f.write('"int_scalar":[' + ",".join(str(
|
||
random.randint(-999999, 9999999)) for i in range(rows)) + "]")
|
||
f.write("\n")
|
||
if data_field == DataField.float_field:
|
||
if err_type == DataErrorType.int_on_float_scalar:
|
||
f.write('"float_scalar":[' + ",".join(
|
||
str(random.randint(-999999, 9999999)) for i in range(rows)) + "]")
|
||
elif err_type == DataErrorType.str_on_float_scalar:
|
||
f.write('"float_scalar":["' + ',"'.join(str(
|
||
gen_unique_str()) + '"' for i in range(rows)) + ']')
|
||
else:
|
||
f.write('"float_scalar":[' + ",".join(
|
||
str(random.random()) for i in range(rows)) + "]")
|
||
f.write("\n")
|
||
if data_field == DataField.string_field:
|
||
f.write('"string_scalar":["' + ',"'.join(str(
|
||
gen_unique_str()) + '"' for i in range(rows)) + ']')
|
||
f.write("\n")
|
||
if data_field == DataField.bool_field:
|
||
if err_type == DataErrorType.typo_on_bool:
|
||
f.write('"bool_scalar":[' + ",".join(
|
||
str(random.choice(["True", "False", "TRUE", "FALSE", "1", "0"])) for i in range(rows)) + "]")
|
||
else:
|
||
f.write('"bool_scalar":[' + ",".join(
|
||
str(random.choice(["true", "false"])) for i in range(rows)) + "]")
|
||
f.write("\n")
|
||
if data_field == DataField.vec_field:
|
||
# vector columns
|
||
if err_type == DataErrorType.one_entity_wrong_dim:
|
||
wrong_dim = dim + 8 # add 8 to compatible with binary vectors
|
||
wrong_row = kwargs.get("wrong_position", 0)
|
||
if wrong_row <= 0:
|
||
vectors1 = []
|
||
else:
|
||
vectors1 = gen_float_vectors(wrong_row, dim) if float_vect else \
|
||
gen_binary_vectors(wrong_row, (dim//8))
|
||
if wrong_row >= rows -1:
|
||
vectors2 = []
|
||
else:
|
||
vectors2 = gen_float_vectors(rows-wrong_row-1, dim) if float_vect else\
|
||
gen_binary_vectors(rows-wrong_row-1, (dim//8))
|
||
vectors_wrong_dim = gen_float_vectors(1, wrong_dim) if float_vect else \
|
||
gen_binary_vectors(1, (wrong_dim//8))
|
||
vectors = vectors1 + vectors_wrong_dim + vectors2
|
||
elif err_type == DataErrorType.str_on_vector_field:
|
||
wrong_row = kwargs.get("wrong_position", 0)
|
||
if wrong_row <= 0:
|
||
vectors1 = []
|
||
else:
|
||
vectors1 = gen_float_vectors(wrong_row, dim) if float_vect else \
|
||
gen_binary_vectors(wrong_row, (dim//8))
|
||
if wrong_row >= rows - 1:
|
||
vectors2 = []
|
||
else:
|
||
vectors2 = gen_float_vectors(rows - wrong_row - 1, dim) if float_vect else \
|
||
gen_binary_vectors(rows - wrong_row - 1, (dim // 8))
|
||
invalid_str_vectors = gen_str_invalid_vectors(1, dim) if float_vect else \
|
||
gen_str_invalid_vectors(1, (dim // 8))
|
||
vectors = vectors1 + invalid_str_vectors + vectors2
|
||
else:
|
||
vectors = gen_float_vectors(rows, dim) if float_vect else gen_binary_vectors(rows, (dim//8))
|
||
f.write('"vectors":[' + ",".join(str(x).replace("'", '"') for x in vectors) + "]")
|
||
f.write("\n")
|
||
if j != len(data_fields) - 1:
|
||
f.write(",")
|
||
f.write("}")
|
||
f.write("\n")
|
||
|
||
|
||
def gen_vectors_in_numpy_file(dir, float_vector, rows, dim, force=False):
|
||
file_name = f"{DataField.vec_field}.npy"
|
||
file = f'{dir}/{file_name}'
|
||
|
||
if not os.path.exists(file) or force:
|
||
# vector columns
|
||
vectors = []
|
||
if rows > 0:
|
||
if float_vector:
|
||
vectors = gen_float_vectors(rows, dim)
|
||
else:
|
||
vectors = gen_binary_vectors(rows, (dim // 8))
|
||
arr = np.array(vectors)
|
||
np.save(file, arr)
|
||
return file_name
|
||
|
||
|
||
def gen_int_or_float_in_numpy_file(dir, data_field, rows, start=0, force=False):
|
||
file_name = f"{data_field}.npy"
|
||
file = f"{dir}/{file_name}"
|
||
if not os.path.exists(file) or force:
|
||
# non vector columns
|
||
data = []
|
||
if rows > 0:
|
||
if data_field == DataField.float_field:
|
||
data = [random.random() for _ in range(rows)]
|
||
elif data_field == DataField.pk_field:
|
||
data = [i for i in range(start, start + rows)]
|
||
elif data_field == DataField.int_field:
|
||
data = [random.randint(-999999, 9999999) for _ in range(rows)]
|
||
arr = np.array(data)
|
||
np.save(file, arr)
|
||
return file_name
|
||
|
||
|
||
def gen_file_name(row_based, rows, dim, auto_id, str_pk,
|
||
float_vector, data_fields, file_num, file_type, err_type):
|
||
row_suffix = entity_suffix(rows)
|
||
field_suffix = ""
|
||
if len(data_fields) > 3:
|
||
field_suffix = "multi_scalars_"
|
||
else:
|
||
for data_field in data_fields:
|
||
if data_field != DataField.vec_field:
|
||
field_suffix += f"{data_field}_"
|
||
|
||
vt = ""
|
||
if DataField.vec_field in data_fields:
|
||
vt = "float_vectors_" if float_vector else "binary_vectors_"
|
||
|
||
pk = ""
|
||
if str_pk:
|
||
pk = "str_pk_"
|
||
prefix = gen_file_prefix(row_based=row_based, auto_id=auto_id, prefix=err_type)
|
||
|
||
file_name = f"{prefix}_{pk}{vt}{field_suffix}{dim}d_{row_suffix}_{file_num}{file_type}"
|
||
return file_name
|
||
|
||
|
||
def gen_subfolder(root, dim, rows, file_num):
|
||
suffix = entity_suffix(rows)
|
||
subfolder = f"{dim}d_{suffix}_{file_num}"
|
||
path = f"{root}/{subfolder}"
|
||
if not os.path.isdir(path):
|
||
os.mkdir(path)
|
||
return subfolder
|
||
|
||
|
||
def gen_json_files(row_based, rows, dim, auto_id, str_pk,
|
||
float_vector, data_fields, file_nums, multi_folder,
|
||
file_type, err_type, force, **kwargs):
|
||
# gen json files
|
||
files = []
|
||
start_uid = 0
|
||
# make sure pk field exists when not auto_id
|
||
if not auto_id and DataField.pk_field not in data_fields:
|
||
data_fields.append(DataField.pk_field)
|
||
for i in range(file_nums):
|
||
file_name = gen_file_name(row_based=row_based, rows=rows, dim=dim,
|
||
auto_id=auto_id, str_pk=str_pk, float_vector=float_vector,
|
||
data_fields=data_fields, file_num=i, file_type=file_type, err_type=err_type)
|
||
file = f"{data_source}/{file_name}"
|
||
if multi_folder:
|
||
subfolder = gen_subfolder(root=data_source, dim=dim, rows=rows, file_num=i)
|
||
file = f"{data_source}/{subfolder}/{file_name}"
|
||
if not os.path.exists(file) or force:
|
||
if row_based:
|
||
gen_row_based_json_file(row_file=file, str_pk=str_pk, float_vect=float_vector,
|
||
data_fields=data_fields, rows=rows, dim=dim,
|
||
start_uid=start_uid, err_type=err_type, **kwargs)
|
||
else:
|
||
gen_column_base_json_file(col_file=file, str_pk=str_pk, float_vect=float_vector,
|
||
data_fields=data_fields, rows=rows, dim=dim,
|
||
start_uid=start_uid, err_type=err_type, **kwargs)
|
||
start_uid += rows
|
||
if multi_folder:
|
||
files.append(f"{subfolder}/{file_name}")
|
||
else:
|
||
files.append(file_name)
|
||
return files
|
||
|
||
|
||
def gen_npy_files(float_vector, rows, dim, data_fields, file_nums=1, force=False):
|
||
# gen numpy files
|
||
files = []
|
||
start_uid = 0
|
||
if file_nums == 1:
|
||
# gen the numpy file without subfolders if only one set of files
|
||
for data_field in data_fields:
|
||
if data_field == DataField.vec_field:
|
||
file_name = gen_vectors_in_numpy_file(dir=data_source, float_vector=float_vector,
|
||
rows=rows, dim=dim, force=force)
|
||
else:
|
||
file_name = gen_int_or_float_in_numpy_file(dir=data_source, data_field=data_field,
|
||
rows=rows, force=force)
|
||
files.append(file_name)
|
||
else:
|
||
for i in range(file_nums):
|
||
subfolder = gen_subfolder(root=data_source, dim=dim, rows=rows, file_num=i)
|
||
dir = f"{data_source}/{subfolder}"
|
||
for data_field in data_fields:
|
||
if data_field == DataField.vec_field:
|
||
file_name = gen_vectors_in_numpy_file(dir=dir, float_vector=float_vector, rows=rows, dim=dim, force=force)
|
||
else:
|
||
file_name = gen_int_or_float_in_numpy_file(dir=dir, data_field=data_field, rows=rows, start=start_uid, force=force)
|
||
files.append(f"{subfolder}/{file_name}")
|
||
start_uid += rows
|
||
return files
|
||
|
||
|
||
def prepare_bulk_load_json_files(row_based=True, rows=100, dim=128,
|
||
auto_id=True, str_pk=False, float_vector=True,
|
||
data_fields=[], file_nums=1, multi_folder=False,
|
||
file_type=".json", err_type="", force=False, **kwargs):
|
||
"""
|
||
Generate files based on the params in json format and copy them to minio
|
||
|
||
:param row_based: indicate the file(s) to be generated is row based or not
|
||
:type row_based: boolean
|
||
|
||
:param rows: the number entities to be generated in the file(s)
|
||
:type rows: int
|
||
|
||
:param dim: dim of vector data
|
||
:type dim: int
|
||
|
||
:param auto_id: generate primary key data or not
|
||
:type auto_id: boolean
|
||
|
||
:param str_pk: generate string or int as primary key
|
||
:type str_pk: boolean
|
||
|
||
:param: float_vector: generate float vectors or binary vectors
|
||
:type float_vector: boolean
|
||
|
||
:param: data_fields: data fields to be generated in the file(s):
|
||
It supports one or all of [pk, vectors, int, float, string, boolean]
|
||
Note: it automatically adds pk field if auto_id=False
|
||
:type data_fields: list
|
||
|
||
:param file_nums: file numbers to be generated
|
||
:type file_nums: int
|
||
|
||
:param multi_folder: generate the files in bucket root folder or new subfolders
|
||
:type multi_folder: boolean
|
||
|
||
:param file_type: specify the file suffix to be generate
|
||
:type file_type: str
|
||
|
||
:param err_type: inject some errors in the file(s).
|
||
All errors should be predefined in DataErrorType
|
||
:type err_type: str
|
||
|
||
:param force: re-generate the file(s) regardless existing or not
|
||
:type force: boolean
|
||
|
||
:param **kwargs
|
||
* *wrong_position* (``int``) --
|
||
indicate the error entity in the file if DataErrorType.one_entity_wrong_dim
|
||
|
||
:return list
|
||
file names list
|
||
"""
|
||
files = gen_json_files(row_based=row_based, rows=rows, dim=dim,
|
||
auto_id=auto_id, str_pk=str_pk, float_vector=float_vector,
|
||
data_fields=data_fields, file_nums=file_nums, multi_folder=multi_folder,
|
||
file_type=file_type, err_type=err_type, force=force, **kwargs)
|
||
|
||
copy_files_to_minio(host=minio, r_source=data_source, files=files, bucket_name=bucket_name, force=force)
|
||
return files
|
||
|
||
|
||
def prepare_bulk_load_numpy_files(rows, dim, data_fields=[DataField.vec_field],
|
||
float_vector=True, file_nums=1, force=False):
|
||
"""
|
||
Generate column based files based on params in numpy format and copy them to the minio
|
||
Note: each field in data_fields would be generated one numpy file.
|
||
|
||
:param rows: the number entities to be generated in the file(s)
|
||
:type rows: int
|
||
|
||
:param dim: dim of vector data
|
||
:type dim: int
|
||
|
||
:param: float_vector: generate float vectors or binary vectors
|
||
:type float_vector: boolean
|
||
|
||
:param: data_fields: data fields to be generated in the file(s):
|
||
it support one or all of [int_pk, vectors, int, float]
|
||
Note: it does not automatically adds pk field
|
||
:type data_fields: list
|
||
|
||
:param file_nums: file numbers to be generated
|
||
The file(s) would be geneated in data_source folder if file_nums = 1
|
||
The file(s) would be generated in different subfolers if file_nums > 1
|
||
:type file_nums: int
|
||
|
||
:param force: re-generate the file(s) regardless existing or not
|
||
:type force: boolean
|
||
|
||
Return: List
|
||
File name list or file name with subfolder list
|
||
"""
|
||
files = gen_npy_files(rows=rows, dim=dim, float_vector=float_vector,
|
||
data_fields=data_fields,
|
||
file_nums=file_nums, force=force)
|
||
|
||
copy_files_to_minio(host=minio, r_source=data_source, files=files, bucket_name=bucket_name, force=force)
|
||
return files
|
||
|