For one batch process I query my database ~30_000x. Every query returns between 0 and 20_000 rows. Profiling the code shows that the most time is spend getting this data out.
I tested 2 different methods with similar results. Assuming that the database schema is not the bottleneck, what can I do to speed up the data retrieval (besides going parallel)?
I was thinking on using another library/wrapper for sqlite
getting rid of pandas, but a lot of computation is done later on with the df, dask eventual faster?
wrapping it in cpython/julia?
db = sqlite.connect('D:\data.db')
cur = db.cursor()
# 1.just pandas, 231 ms
t = pd.read_sql("SELECT * FROM daily d WHERE d.id == 1 ", db)
# 2. db-api & pandas, 242ms
# query is actual half the time of 1., but creating that df cost time
cur.execute("SELECT * FROM daily d WHERE d.id ==1")
rows = cur.fetchall()
t = pd.DataFrame(rows, columns=col)
The question is very general given that we know little about the processing you do with Pandas.
The one suggestion though, would be to move as much processing as possible (particularly limiting the size of the dataframe) to SQL for SQLite to process.
So, if there is any filtering done in Pandas, I would strive to move it to SQL even if the condition for SELECT or GROUP BY is somewhat cumbersome. There is a cost of copying the data to Python realm and pandas is eating up memory and time.
I would like to send a large pandas.DataFrame to a remote server running MS SQL. The way I do it now is by converting a data_frame object to a list of tuples and then send it away with pyODBC's executemany() function. It goes something like this:
import pyodbc as pdb
list_of_tuples = convert_df(data_frame)
connection = pdb.connect(cnxn_str)
cursor = connection.cursor()
cursor.fast_executemany = True
cursor.executemany(sql_statement, list_of_tuples)
connection.commit()
cursor.close()
connection.close()
I then started to wonder if things can be sped up (or at least more readable) by using data_frame.to_sql() method. I have came up with the following solution:
import sqlalchemy as sa
engine = sa.create_engine("mssql+pyodbc:///?odbc_connect=%s" % cnxn_str)
data_frame.to_sql(table_name, engine, index=False)
Now the code is more readable, but the upload is at least 150 times slower...
Is there a way to flip the fast_executemany when using SQLAlchemy?
I am using pandas-0.20.3, pyODBC-4.0.21 and sqlalchemy-1.1.13.
EDIT (2019-03-08): Gord Thompson commented below with good news from the update logs of sqlalchemy: Since SQLAlchemy 1.3.0, released 2019-03-04, sqlalchemy now supports engine = create_engine(sqlalchemy_url, fast_executemany=True) for the mssql+pyodbc dialect. I.e., it is no longer necessary to define a function and use #event.listens_for(engine, 'before_cursor_execute') Meaning the below function can be removed and only the flag needs to be set in the create_engine statement - and still retaining the speed-up.
Original Post:
Just made an account to post this. I wanted to comment beneath the above thread as it's a followup on the already provided answer. The solution above worked for me with the Version 17 SQL driver on a Microsft SQL storage writing from a Ubuntu based install.
The complete code I used to speed things up significantly (talking >100x speed-up) is below. This is a turn-key snippet provided that you alter the connection string with your relevant details. To the poster above, thank you very much for the solution as I was looking quite some time for this already.
import pandas as pd
import numpy as np
import time
from sqlalchemy import create_engine, event
from urllib.parse import quote_plus
conn = "DRIVER={ODBC Driver 17 for SQL Server};SERVER=IP_ADDRESS;DATABASE=DataLake;UID=USER;PWD=PASS"
quoted = quote_plus(conn)
new_con = 'mssql+pyodbc:///?odbc_connect={}'.format(quoted)
engine = create_engine(new_con)
#event.listens_for(engine, 'before_cursor_execute')
def receive_before_cursor_execute(conn, cursor, statement, params, context, executemany):
print("FUNC call")
if executemany:
cursor.fast_executemany = True
table_name = 'fast_executemany_test'
df = pd.DataFrame(np.random.random((10**4, 100)))
s = time.time()
df.to_sql(table_name, engine, if_exists = 'replace', chunksize = None)
print(time.time() - s)
Based on the comments below I wanted to take some time to explain some limitations about the pandas to_sql implementation and the way the query is handled. There are 2 things that might cause the MemoryError being raised afaik:
1) Assuming you're writing to a remote SQL storage. When you try to write a large pandas DataFrame with the to_sql method it converts the entire dataframe into a list of values. This transformation takes up way more RAM than the original DataFrame does (on top of it, as the old DataFrame still remains present in RAM). This list is provided to the final executemany call for your ODBC connector. I think the ODBC connector has some troubles handling such large queries. A way to solve this is to provide the to_sql method a chunksize argument (10**5 seems to be around optimal giving about 600 mbit/s (!) write speeds on a 2 CPU 7GB ram MSSQL Storage application from Azure - can't recommend Azure btw). So the first limitation, being the query size, can be circumvented by providing a chunksize argument. However, this won't enable you to write a dataframe the size of 10**7 or larger, (at least not on the VM I am working with which has ~55GB RAM), being issue nr 2.
This can be circumvented by breaking up the DataFrame with np.split (being 10**6 size DataFrame chunks) These can be written away iteratively. I will try to make a pull request when I have a solution ready for the to_sql method in the core of pandas itself so you won't have to do this pre-breaking up every time. Anyhow I ended up writing a function similar (not turn-key) to the following:
import pandas as pd
import numpy as np
def write_df_to_sql(df, **kwargs):
chunks = np.split(df, df.shape()[0] / 10**6)
for chunk in chunks:
chunk.to_sql(**kwargs)
return True
A more complete example of the above snippet can be viewed here: https://gitlab.com/timelord/timelord/blob/master/timelord/utils/connector.py
It's a class I wrote that incorporates the patch and eases some of the necessary overhead that comes with setting up connections with SQL. Still have to write some documentation. Also I was planning on contributing the patch to pandas itself but haven't found a nice way yet on how to do so.
I hope this helps.
After contacting the developers of SQLAlchemy, a way to solve this problem has emerged. Many thanks to them for the great work!
One has to use a cursor execution event and check if the executemany flag has been raised. If that is indeed the case, switch the fast_executemany option on. For example:
from sqlalchemy import event
#event.listens_for(engine, 'before_cursor_execute')
def receive_before_cursor_execute(conn, cursor, statement, params, context, executemany):
if executemany:
cursor.fast_executemany = True
More information on execution events can be found here.
UPDATE: Support for fast_executemany of pyodbc was added in SQLAlchemy 1.3.0, so this hack is not longer necessary.
I ran into the same problem but using PostgreSQL. They now just release pandas version 0.24.0 and there is a new parameter in the to_sql function called method which solved my problem.
from sqlalchemy import create_engine
engine = create_engine(your_options)
data_frame.to_sql(table_name, engine, method="multi")
Upload speed is 100x faster for me.
I also recommend setting the chunksize parameter if you are going to send lots of data.
I just wanted to post this full example as an additional, high-performance option for those who can use the new turbodbc library: http://turbodbc.readthedocs.io/en/latest/
There clearly are many options in flux between pandas .to_sql(), triggering fast_executemany through sqlalchemy, using pyodbc directly with tuples/lists/etc., or even trying BULK UPLOAD with flat files.
Hopefully, the following might make life a bit more pleasant as functionality evolves in the current pandas project or includes something like turbodbc integration in the future.
import pandas as pd
import numpy as np
from turbodbc import connect, make_options
from io import StringIO
test_data = '''id,transaction_dt,units,measures
1,2018-01-01,4,30.5
1,2018-01-03,4,26.3
2,2018-01-01,3,12.7
2,2018-01-03,3,8.8'''
df_test = pd.read_csv(StringIO(test_data), sep=',')
df_test['transaction_dt'] = pd.to_datetime(df_test['transaction_dt'])
options = make_options(parameter_sets_to_buffer=1000)
conn = connect(driver='{SQL Server}', server='server_nm', database='db_nm', turbodbc_options=options)
test_query = '''DROP TABLE IF EXISTS [db_name].[schema].[test]
CREATE TABLE [db_name].[schema].[test]
(
id int NULL,
transaction_dt datetime NULL,
units int NULL,
measures float NULL
)
INSERT INTO [db_name].[schema].[test] (id,transaction_dt,units,measures)
VALUES (?,?,?,?) '''
cursor.executemanycolumns(test_query, [df_test['id'].values, df_test['transaction_dt'].values, df_test['units'].values, df_test['measures'].values]
turbodbc should be VERY fast in many use cases (particularly with numpy arrays). Please observe how straightforward it is to pass the underlying numpy arrays from the dataframe columns as parameters to the query directly. I also believe this helps prevent the creation of intermediate objects that spike memory consumption excessively. Hope this is helpful!
It seems that Pandas 0.23.0 and 0.24.0 use multi values inserts with PyODBC, which prevents fast executemany from helping – a single INSERT ... VALUES ... statement is emitted per chunk. The multi values insert chunks are an improvement over the old slow executemany default, but at least in simple tests the fast executemany method still prevails, not to mention no need for manual chunksize calculations, as is required with multi values inserts. Forcing the old behaviour can be done by monkeypatching, if no configuration option is provided in the future:
import pandas.io.sql
def insert_statement(self, data, conn):
return self.table.insert(), data
pandas.io.sql.SQLTable.insert_statement = insert_statement
The future is here and at least in the master branch the insert method can be controlled using the keyword argument method= of to_sql(). It defaults to None, which forces the executemany method. Passing method='multi' results in using the multi values insert. It can even be used to implement DBMS specific approaches, such as Postgresql COPY.
As pointed out by #Pylander
Turbodbc is the best choice for data ingestion, by far!
I got so excited about it that I wrote a 'blog' on it on my github and medium:
please check https://medium.com/#erickfis/etl-process-with-turbodbc-1d19ed71510e
for a working example and comparison with pandas.to_sql
Long story short,
with turbodbc
I've got 10000 lines (77 columns) in 3 seconds
with pandas.to_sql
I've got the same 10000 lines (77 columns) in 198 seconds...
And here is what I'm doing in full detail
The imports:
import sqlalchemy
import pandas as pd
import numpy as np
import turbodbc
import time
Load and treat some data - Substitute my sample.pkl for yours:
df = pd.read_pickle('sample.pkl')
df.columns = df.columns.str.strip() # remove white spaces around column names
df = df.applymap(str.strip) # remove white spaces around values
df = df.replace('', np.nan) # map nans, to drop NAs rows and columns later
df = df.dropna(how='all', axis=0) # remove rows containing only NAs
df = df.dropna(how='all', axis=1) # remove columns containing only NAs
df = df.replace(np.nan, 'NA') # turbodbc hates null values...
Create the table using sqlAlchemy
Unfortunately, turbodbc requires a lot of overhead with a lot of sql manual labor, for creating the tables and for inserting data on it.
Fortunately, Python is pure joy and we can automate this process of writing sql code.
The first step is creating the table which will receive our data. However, creating the table manually writing sql code can be problematic if your table has more than a few columns. In my case, very often the tables have 240 columns!
This is where sqlAlchemy and pandas still can help us: pandas is bad for writing a large number of rows (10000 in this example), but what about just 6 rows, the head of the table? This way, we automate the process of creating the tables.
Create sqlAlchemy connection:
mydb = 'someDB'
def make_con(db):
"""Connect to a specified db."""
database_connection = sqlalchemy.create_engine(
'mssql+pymssql://{0}:{1}#{2}/{3}'.format(
myuser, mypassword,
myhost, db
)
)
return database_connection
pd_connection = make_con(mydb)
Create table on SQL Server
Using pandas + sqlAlchemy, but just for preparing room for turbodbc as previously mentioned. Please note that df.head() here: we are using pandas + sqlAlchemy for inserting only 6 rows of our data. This will run pretty fast and is being done to automate the table creation.
table = 'testing'
df.head().to_sql(table, con=pd_connection, index=False)
Now that the table is already in place, let’s get serious here.
Turbodbc connection:
def turbo_conn(mydb):
"""Connect to a specified db - turbo."""
database_connection = turbodbc.connect(
driver='ODBC Driver 17 for SQL Server',
server=myhost,
database=mydb,
uid=myuser,
pwd=mypassword
)
return database_connection
Preparing sql comands and data for turbodbc. Let’s automate this code creation being creative:
def turbo_write(mydb, df, table):
"""Use turbodbc to insert data into sql."""
start = time.time()
# preparing columns
colunas = '('
colunas += ', '.join(df.columns)
colunas += ')'
# preparing value place holders
val_place_holder = ['?' for col in df.columns]
sql_val = '('
sql_val += ', '.join(val_place_holder)
sql_val += ')'
# writing sql query for turbodbc
sql = f"""
INSERT INTO {mydb}.dbo.{table} {colunas}
VALUES {sql_val}
"""
# writing array of values for turbodbc
valores_df = [df[col].values for col in df.columns]
# cleans the previous head insert
with connection.cursor() as cursor:
cursor.execute(f"delete from {mydb}.dbo.{table}")
connection.commit()
# inserts data, for real
with connection.cursor() as cursor:
try:
cursor.executemanycolumns(sql, valores_df)
connection.commit()
except Exception:
connection.rollback()
print('something went wrong')
stop = time.time() - start
return print(f'finished in {stop} seconds')
Writing data using turbodbc - I’ve got 10000 lines (77 columns) in 3 seconds:
turbo_write(mydb, df.sample(10000), table)
Pandas method comparison - I’ve got the same 10000 lines (77 columns) in 198 seconds…
table = 'pd_testing'
def pandas_comparisson(df, table):
"""Load data using pandas."""
start = time.time()
df.to_sql(table, con=pd_connection, index=False)
stop = time.time() - start
return print(f'finished in {stop} seconds')
pandas_comparisson(df.sample(10000), table)
Environment and conditions
Python 3.6.7 :: Anaconda, Inc.
TURBODBC version ‘3.0.0’
sqlAlchemy version ‘1.2.12’
pandas version ‘0.23.4’
Microsoft SQL Server 2014
user with bulk operations privileges
Please check https://erickfis.github.io/loose-code/ for updates in this code!
SQL Server INSERT performance: pyodbc vs. turbodbc
When using to_sql to upload a pandas DataFrame to SQL Server, turbodbc will definitely be faster than pyodbc without fast_executemany. However, with fast_executemany enabled for pyodbc, both approaches yield essentially the same performance.
Test environments:
[venv1_pyodbc]
pyodbc 2.0.25
[venv2_turbodbc]
turbodbc 3.0.0
sqlalchemy-turbodbc 0.1.0
[common to both]
Python 3.6.4 64-bit on Windows
SQLAlchemy 1.3.0b1
pandas 0.23.4
numpy 1.15.4
Test code:
# for pyodbc
engine = create_engine('mssql+pyodbc://sa:whatever#SQL_panorama', fast_executemany=True)
# for turbodbc
# engine = create_engine('mssql+turbodbc://sa:whatever#SQL_panorama')
# test data
num_rows = 10000
num_cols = 100
df = pd.DataFrame(
[[f'row{x:04}col{y:03}' for y in range(num_cols)] for x in range(num_rows)],
columns=[f'col{y:03}' for y in range(num_cols)]
)
t0 = time.time()
df.to_sql("sqlalchemy_test", engine, if_exists='replace', index=None)
print(f"pandas wrote {num_rows} rows in {(time.time() - t0):0.1f} seconds")
Tests were run twelve (12) times for each environment, discarding the single best and worst times for each. Results (in seconds):
rank pyodbc turbodbc
---- ------ --------
1 22.8 27.5
2 23.4 28.1
3 24.6 28.2
4 25.2 28.5
5 25.7 29.3
6 26.9 29.9
7 27.0 31.4
8 30.1 32.1
9 33.6 32.5
10 39.8 32.9
---- ------ --------
average 27.9 30.0
Just wanted to add to the #J.K.'s answer.
If you are using this approach:
#event.listens_for(engine, 'before_cursor_execute')
def receive_before_cursor_execute(conn, cursor, statement, params, context, executemany):
if executemany:
cursor.fast_executemany = True
And you are getting this error:
"sqlalchemy.exc.DBAPIError: (pyodbc.Error) ('HY010', '[HY010]
[Microsoft][SQL Server Native Client 11.0]Function sequence error (0)
(SQLParamData)') [SQL: 'INSERT INTO ... (...) VALUES (?, ?)']
[parameters: ((..., ...), (..., ...)] (Background on this error at:
http://sqlalche.me/e/dbapi)"
Encode your string values like this: 'yourStringValue'.encode('ascii')
This will solve your problem.
I just modify engine line which helps me to speedup the insertion 100 times.
Old Code -
import json
import maya
import time
import pandas
import pyodbc
import pandas as pd
from sqlalchemy import create_engine
retry_count = 0
retry_flag = True
hostInfoDf = pandas.read_excel('test.xlsx', sheet_name='test')
print("Read Ok")
engine = create_engine("mssql+pyodbc://server_name/db_name?trusted_connection=yes&driver=ODBC+Driver+17+for+SQL+Server")
while retry_flag and retry_count < 5:
try:
df.to_sql("table_name",con=engine,if_exists="replace",index=False,chunksize=5000,schema="dbo")
retry_flag = False
except:
retry_count = retry_count + 1
time.sleep(30)
Modified engine line -
From -
engine = create_engine("mssql+pyodbc://server_name/db_name?trusted_connection=yes&driver=ODBC+Driver+17+for+SQL+Server")
to -
engine = create_engine("mssql+pyodbc://server_name/db_name?trusted_connection=yes&driver=ODBC+Driver+17+for+SQL+Server", fast_executemany=True)
ask me any Query related python to SQL connectivity, I will be happy to help you.
I am populating a graph from an SQLite3 database into neo4j, using py2neo with Python 3.2 on Ubuntu linux. Although speed is not of the utmost concern, the graph has only gotten 40K rows (one relation for each sql-row) in about 3hours, out of a total of 5 million rows.
Here is the main loop:
from py2neo import neo4j as neo
import sqlite3 as sql
#select all 5M rows from sql-database
sql_str = """select * from bigram_with_number"""
#loop through each row
for (freq, first, firstfreq, second, secondfreq) in sql_cursor.execute(sql_str):
# create the Cypher query string using cypher 2.0 with merge
# so that nodes are created only if needed
query = neo.CypherQuery(neo4j_db,"""
CYPHER 2.0
merge (n:word {form: {firstvar}, freq: {freqfirst}})
merge(m:word {form: {secondvar}, freq: {freqsecond}})
create unique (n)-[:bigram {freq: {freqbigram}}]->(m) return n, m""")
#execute the string with parameters from sql-query
result = query.execute(freqbigram = freq, firstvar = first, freqfirst=firstfreq, secondvar=second, freqsecond=secondfreq)
Although the database populates nicely, it will take weeks before it is finished.
I suspect it is possible to do this faster.
For bulk loading, you're probably better off bypassing the REST interface and using something lower level such as Michael Hunger's load tools: https://github.com/jexp/neo4j-shell-tools. Even at optimal performance, the REST interface is unlikely to ever achieve the speeds you're looking for.
As an aside, please note that I don't officially support Python 3.2 although I do support 3.3.