python - Get data from pandas into a SQL server with PYODBC

32
votes

I am trying to understand how python could pull data from an FTP server into pandas then move this into SQL server. My code here is very rudimentary to say the least and I am looking for any advice or help at all. I have tried to load the data from the FTP server first which works fine.... If I then remove this code and change it to a select from ms sql server it is fine so the connection string works, but the insertion into the SQL server seems to be causing problems.

import pyodbc
import pandas
from ftplib import FTP
from StringIO import StringIO
import csv

ftp = FTP ('ftp.xyz.com','user','pass' )
ftp.set_pasv(True)
r = StringIO()
ftp.retrbinary('filname.csv', r.write)

pandas.read_table (r.getvalue(), delimiter=',')


connStr = ('DRIVER={SQL Server Native Client 10.0};SERVER=localhost;DATABASE=TESTFEED;UID=sa;PWD=pass')
conn = pyodbc.connect(connStr)

cursor = conn.cursor()
cursor.execute("INSERT INTO dbo.tblImport(Startdt, Enddt, x,y,z,)" "VALUES                  (x,x,x,x,x,x,x,x,x,x.x,x)")
cursor.close()
conn.commit()
conn.close()
print"Script has successfully run!"

When I remove the ftp code this runs perfectly, but I do not understand how to make the next jump to get this into Microsoft SQL server, or even if it is possible without saving into a file first.

5
pythonsqlpandaspyodbc
Do you know how to pass parameters to the execute function? If so, all you need to do is iterate over the rows of the DataFrame and, for each one, call execute and pass the row as the values for the SQL parameters. Or, if PyODBC supports executemany, that's even easier—just pass any iterable of rows, which you already have. - abarnert
That being said, is there a reason you're using Pandas instead of csv here? You're not actually doing anything Pandas-y with the data, just iterating over it. In a case like that, csv is a lot simpler, and doesn't require reading the whole thing into memory all at once. - abarnert
thanks for the reply im not really using pandas for any other reason than i read about it and it seemed logical to dump into a dataframe. My basic aim is to get the FTP data into SQL with CSV would this then only be possible by a CVS file after the event? idealy i'd like pull and push into SQL in one go. - andy redmayne
in terms of the data the files are quite small (around 10Kb) so reading into memory isnt an issue - andy redmayne
You don't need to create an actual CSV file; the csv module works just fine with a StringIO, as you're already doing, and I'm pretty sure Pandas does also. Either way, the goal is just to get some iterable of rows and insert each one by looping over the rows and calling execute. - abarnert

5 Answers

59
votes

For the 'write to sql server' part, you can use the convenient to_sql method of pandas (so no need to iterate over the rows and do the insert manually). See the docs on interacting with SQL databases with pandas: http://pandas.pydata.org/pandas-docs/stable/io.html#io-sql

You will need at least pandas 0.14 to have this working, and you also need sqlalchemy installed. An example, assuming df is the DataFrame you got from read_table:

import sqlalchemy
import pyodbc
engine = sqlalchemy.create_engine("mssql+pyodbc://<username>:<password>@<dsnname>")

# write the DataFrame to a table in the sql database
df.to_sql("table_name", engine)

See also the documentation page of to_sql.
More info on how to create the connection engine with sqlalchemy for sql server with pyobdc, you can find here:http://docs.sqlalchemy.org/en/rel_1_1/dialects/mssql.html#dialect-mssql-pyodbc-connect

But if your goal is to just get the csv data into the SQL database, you could also consider doing this directly from SQL. See eg Import CSV file into SQL Server

20
votes

Python3 version using a LocalDB SQL instance:

from sqlalchemy import create_engine
import urllib
import pyodbc
import pandas as pd

df = pd.read_csv("./data.csv")

quoted = urllib.parse.quote_plus("DRIVER={SQL Server Native Client 11.0};SERVER=(localDb)\ProjectsV14;DATABASE=database")
engine = create_engine('mssql+pyodbc:///?odbc_connect={}'.format(quoted))

df.to_sql('TargetTable', schema='dbo', con = engine)

result = engine.execute('SELECT COUNT(*) FROM [dbo].[TargetTable]')
result.fetchall()
5
votes

Yes, the bcp utility seems to be the best solution for most cases.

If you want to stay within Python, the following code should work.

from sqlalchemy import create_engine
import urllib
import pyodbc

quoted = urllib.parse.quote_plus("DRIVER={SQL Server};SERVER=YOUR\ServerName;DATABASE=YOur_Database")
engine = create_engine('mssql+pyodbc:///?odbc_connect={}'.format(quoted))

df.to_sql('Table_Name', schema='dbo', con = engine, chunksize=200, method='multi', index=False, if_exists='replace')

Don't avoid method='multi', because it significantly reduces the task execution time.

Sometimes you may encounter the following error.

ProgrammingError: ('42000', '[42000] [Microsoft][ODBC SQL Server Driver][SQL Server]The incoming request has too many parameters. The server supports a maximum of 2100 parameters. Reduce the number of parameters and resend the request. (8003) (SQLExecDirectW)')

In such a case, determine the number of columns in your dataframe: df.shape[1]. Divide the maximum supported number of parameters by this value and use the result's floor as a chunk size.

3
votes

I found that using bcp utility (https://docs.microsoft.com/en-us/sql/tools/bcp-utility) works best when you have a large dataset. I have 2.7 million rows that inserts at 80K rows/sec. You can store your data frame as csv file (use tabs for separator if your data doesn't have tabs and utf8 encoding). With bcp, I've used format "-c" and it works without issues so far.

0
votes

This worked for me on Python 3.5.2:

import sqlalchemy as sa
import urllib
import pyodbc

conn= urllib.parse.quote_plus('DRIVER={ODBC Driver 17 for SQL Server};SERVER='+server+';DATABASE='+database+';UID='+username+';PWD='+ password)
engine = sa.create_engine('mssql+pyodbc:///?odbc_connect={}'.format(conn))

frame.to_sql("myTable", engine, schema='dbo', if_exists='append', index=False, index_label='myField')