#789

Committed 18 Sep 2023 04:32PM UTC coverage: 74.685% (-5.7%) from 80.421%

Build # #789

Build Type

push

circle-ci

Committed by Jiashen Cao

Commit Message

reformat

Run Details

8842 of 11839 relevant lines covered (74.69%)

0.75 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

32.14

/evadb/executor/create_index_executor.py

# coding=utf-8
# Copyright 2018-2023 EvaDB
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from pathlib import Path

import pandas as pd

from evadb.catalog.sql_config import ROW_NUM_COLUMN
from evadb.database import EvaDBDatabase
from evadb.executor.abstract_executor import AbstractExecutor
from evadb.executor.executor_utils import ExecutorError, handle_vector_store_params
from evadb.models.storage.batch import Batch
from evadb.plan_nodes.create_index_plan import CreateIndexPlan
from evadb.storage.storage_engine import StorageEngine
from evadb.third_party.vector_stores.types import FeaturePayload
from evadb.third_party.vector_stores.utils import VectorStoreFactory
from evadb.utils.logging_manager import logger


class CreateIndexExecutor(AbstractExecutor):
    def __init__(self, db: EvaDBDatabase, node: CreateIndexPlan):
        super().__init__(db, node)

    def exec(self, *args, **kwargs):
        if self.catalog().get_index_catalog_entry_by_name(self.node.name):
            msg = f"Index {self.node.name} already exists."
            if self.node.if_not_exists:
                logger.warn(msg)
                return
            else:
                logger.error(msg)
                raise ExecutorError(msg)

        self.index_path = self._get_index_save_path()
        self.index = None
        self._create_index()

        yield Batch(
            pd.DataFrame(
                [f"Index {self.node.name} successfully added to the database."]
            )
        )

    def _get_index_save_path(self) -> Path:
        index_dir = Path(self.config.get_value("storage", "index_dir"))
        if not index_dir.exists():
            index_dir.mkdir(parents=True, exist_ok=True)
        return str(
            index_dir
            / Path("{}_{}.index".format(self.node.vector_store_type, self.node.name))
        )

    def _create_index(self):
        try:
            # Get feature tables.
            feat_catalog_entry = self.node.table_ref.table.table_obj

            # Get feature column.
            feat_col_name = self.node.col_list[0].name
            feat_column = [
                col for col in feat_catalog_entry.columns if col.name == feat_col_name
            ][0]

            # Add features to index.
            # TODO: batch size is hardcoded for now.
            input_dim = -1
            storage_engine = StorageEngine.factory(self.db, feat_catalog_entry)
            for input_batch in storage_engine.read(feat_catalog_entry):
                if self.node.function:
                    # Create index through function expression.
                    # Function(input column) -> 2 dimension feature vector.
                    input_batch.modify_column_alias(feat_catalog_entry.name.lower())
                    feat_batch = self.node.function.evaluate(input_batch)
                    feat_batch.drop_column_alias()
                    input_batch.drop_column_alias()
                    feat = feat_batch.column_as_numpy_array("features")
                else:
                    # Create index on the feature table directly.
                    # Pandas wraps numpy array as an object inside a numpy
                    # array. Use zero index to get the actual numpy array.
                    feat = input_batch.column_as_numpy_array(feat_col_name)

                row_num = input_batch.column_as_numpy_array(ROW_NUM_COLUMN)

                for i in range(len(input_batch)):
                    row_feat = feat[i].reshape(1, -1)
                    if self.index is None:
                        input_dim = row_feat.shape[1]
                        self.index = VectorStoreFactory.init_vector_store(
                            self.node.vector_store_type,
                            self.node.name,
                            **handle_vector_store_params(
                                self.node.vector_store_type, self.index_path
                            ),
                        )
                        self.index.create(input_dim)

                    # Row ID for mapping back to the row.
                    self.index.add([FeaturePayload(row_num[i], row_feat)])

            # Persist index.
            self.index.persist()

            # Save to catalog.
            self.catalog().insert_index_catalog_entry(
                self.node.name,
                self.index_path,
                self.node.vector_store_type,
                feat_column,
                self.node.function.signature() if self.node.function else None,
            )
        except Exception as e:
            # Delete index.
            if self.index:
                self.index.delete()

            # Throw exception back to user.
            raise ExecutorError(str(e))

1	# coding=utf-8
2	# Copyright 2018-2023 EvaDB
3	#
4	# Licensed under the Apache License, Version 2.0 (the "License");
5	# you may not use this file except in compliance with the License.
6	# You may obtain a copy of the License at
7	#
8	# http://www.apache.org/licenses/LICENSE-2.0
9	#
10	# Unless required by applicable law or agreed to in writing, software
11	# distributed under the License is distributed on an "AS IS" BASIS,
12	# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13	# See the License for the specific language governing permissions and
14	# limitations under the License.
15	from pathlib import Path	1✔
16
17	import pandas as pd	1✔
18
19	from evadb.catalog.sql_config import ROW_NUM_COLUMN	1✔
20	from evadb.database import EvaDBDatabase	1✔
21	from evadb.executor.abstract_executor import AbstractExecutor	1✔
22	from evadb.executor.executor_utils import ExecutorError, handle_vector_store_params	1✔
23	from evadb.models.storage.batch import Batch	1✔
24	from evadb.plan_nodes.create_index_plan import CreateIndexPlan	1✔
25	from evadb.storage.storage_engine import StorageEngine	1✔
26	from evadb.third_party.vector_stores.types import FeaturePayload	1✔
27	from evadb.third_party.vector_stores.utils import VectorStoreFactory	1✔
28	from evadb.utils.logging_manager import logger	1✔
29
30
31	class CreateIndexExecutor(AbstractExecutor):	1✔
32	def __init__(self, db: EvaDBDatabase, node: CreateIndexPlan):	1✔
33	super().__init__(db, node)	1✔
34
35	def exec(self, args, *kwargs):	1✔
36	if self.catalog().get_index_catalog_entry_by_name(self.node.name):	×
37	msg = f"Index {self.node.name} already exists."	×
38	if self.node.if_not_exists:	×
39	logger.warn(msg)	×
40	return	×
41	else:
42	logger.error(msg)	×
43	raise ExecutorError(msg)
44
45	self.index_path = self._get_index_save_path()	×
46	self.index = None	×
47	self._create_index()	×
48
49	yield Batch(	×
50	pd.DataFrame(
51	[f"Index {self.node.name} successfully added to the database."]
52	)
53	)
54
55	def _get_index_save_path(self) -> Path:	1✔
56	index_dir = Path(self.config.get_value("storage", "index_dir"))	×
57	if not index_dir.exists():	×
58	index_dir.mkdir(parents=True, exist_ok=True)	×
59	return str(	×
60	index_dir
61	/ Path("{}_{}.index".format(self.node.vector_store_type, self.node.name))
62	)
63
64	def _create_index(self):	1✔
65	try:	×
66	# Get feature tables.
67	feat_catalog_entry = self.node.table_ref.table.table_obj	×
68
69	# Get feature column.
70	feat_col_name = self.node.col_list[0].name	×
71	feat_column = [	×
72	col for col in feat_catalog_entry.columns if col.name == feat_col_name
73	][0]
74
75	# Add features to index.
76	# TODO: batch size is hardcoded for now.
77	input_dim = -1	×
78	storage_engine = StorageEngine.factory(self.db, feat_catalog_entry)	×
79	for input_batch in storage_engine.read(feat_catalog_entry):	×
80	if self.node.function:	×
81	# Create index through function expression.
82	# Function(input column) -> 2 dimension feature vector.
83	input_batch.modify_column_alias(feat_catalog_entry.name.lower())	×
84	feat_batch = self.node.function.evaluate(input_batch)	×
85	feat_batch.drop_column_alias()	×
86	input_batch.drop_column_alias()	×
87	feat = feat_batch.column_as_numpy_array("features")	×
88	else:
89	# Create index on the feature table directly.
90	# Pandas wraps numpy array as an object inside a numpy
91	# array. Use zero index to get the actual numpy array.
92	feat = input_batch.column_as_numpy_array(feat_col_name)	×
93
94	row_num = input_batch.column_as_numpy_array(ROW_NUM_COLUMN)	×
95
96	for i in range(len(input_batch)):	×
97	row_feat = feat[i].reshape(1, -1)	×
98	if self.index is None:	×
99	input_dim = row_feat.shape[1]	×
100	self.index = VectorStoreFactory.init_vector_store(	×
101	self.node.vector_store_type,
102	self.node.name,
103	**handle_vector_store_params(
104	self.node.vector_store_type, self.index_path
105	),
106	)
107	self.index.create(input_dim)	×
108
109	# Row ID for mapping back to the row.
110	self.index.add([FeaturePayload(row_num[i], row_feat)])	×
111
112	# Persist index.
113	self.index.persist()	×
114
115	# Save to catalog.
116	self.catalog().insert_index_catalog_entry(	×
117	self.node.name,
118	self.index_path,
119	self.node.vector_store_type,
120	feat_column,
121	self.node.function.signature() if self.node.function else None,
122	)
123	except Exception as e:
124	# Delete index.
125	if self.index:
126	self.index.delete()
127
128	# Throw exception back to user.
129	raise ExecutorError(str(e))

georgia-tech-db / eva / #789

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous