Persistence and Serialization
Overview
Persistence and Serialization are closely related.
Serialization means taking a potentially complex data structure and converting it into a single string of bytes.
Persistance is storing data in a way that it will persist beyond the run-time of your program.
They are closely related, because most forms of persistent storage – simple text files, databases, etc. – require that it be turned into a simple string of bytes first. After all, at the end of the day, everything done with computers is ultimately a serial string of bytes.
Serialization is also very useful for transmitting information between systems – over a network, etc.
Serialization
This module is less about concepts and more about learning to use a given module. So less talk, more coding.
This material is focused on methods available in the Python standard library.
There are third party packages with more options as well.
Persistence
Persistence is saving your python data structure(s) to disk – so they will persist once the python process is finished.
Any serial form can provide persistence – by dumping/loading it to/from a file, for example – but not all persistence mechanisms are serial.
Python Specific Formats
These are formats specific to python. They are convenient to use, but not useful for interchange with other systems.
Python Literals
You can put plain old python literals in your file. It gives you a nice, human-editable form for config files, etc.
But, do NOT use for untrusted sources!!!
It’s good for basic python types and it can work for your own classes, too – if you write a good __repr__ implementations.
In theory, repr() always gives a form that can be re-constructed. Often the str() form works too. But theory and practice do not always align.
The pprint (pretty print) module can make it easier to read.
Python Literal Example
# a list of dicts
data = [{'this':5, 'that':4}, {'spam':7, 'eggs':3.4}]
In [51]: s = repr(data) # save a string version:
In [52]: data2 = eval(s) # re-construct with eval:
In [53]: data2 == data # they are equal
Out[53]: True
In [54]: data is data2 # but not the same object
Out[54]: False
You can save the string to a file and even use import.
In fact, using a python file and importing it is a great way to handle configuration for your app – very powerful and flexible.
NOTE: eval() is DANGEROUS.
It’s not so bad if you know where your data is coming from, but eval() will run any code it gets, even:
import sys
sys.system('cd /; rm -rf *')
You really don’t want that run on your machine!
The alternative:
ast.literal_evalis safer than eval:
It will only evaluate literals.
Pretty Print
In [68]: data = [{'this': 5, 'that': 4}, {'eggs': 3.4, 'spam': 7},
{'foo': 86, 'bar': 4.5}, {'fun': 43, 'baz': 6.5}]
In [69]: import pprint
In [71]: repr(data)
Out[71]: "[{'this': 5, 'that': 4}, {'eggs': 3.4, 'spam': 7}, {'foo': 86, 'bar': 4.5}, {'fun': 43, 'baz': 6.5}]"
In [72]: s = pprint.pformat(data)
In [73]: print(s)
[{'that': 4, 'this': 5},
{'eggs': 3.4, 'spam': 7},
{'bar': 4.5, 'foo': 86},
{'baz': 6.5, 'fun': 43}]
This is a nice option if you want the saved form to be human readable and editable.
Pickle
Pickle is a custom binary format for python objects.
You can essentially dump any python object to disk (or string, or socket, or…
In [87]: import pickle
In [83]: data
Out[83]:
[{'that': 4, 'this': 5},
{'eggs': 3.4, 'spam': 7},
{'bar': 4.5, 'foo': 86},
{'baz': 6.5, 'fun': 43}]
In [84]: pickle.dump(data, open('data.pkl', 'wb'))
In [85]: data2 = pickle.load(open('data.pkl', 'rb'))
In [86]: data2 == data
Out[86]: True
https://docs.python.org/3/library/pickle.html
NOTE: The pickle module is NOT SECURE against erroneous or maliciously constructed data. Never unpickle data received from an untrusted or unauthenticated source.
pickle is cool because it can serialize virtually ANY object – including your self-defined classes.
But to do this, it must run essentially arbitrary code – so not safe.
Do not use it for receiving data from an external source.
But you probably won’t want to do that anyway – pickle is python-specific and thus not very useful for data interchange.
Shelve
A “shelf” is a persistent, dictionary-like object. It’s also a place you can put a jar of pickles.
The values – not the keys! – can be essentially arbitrary Python objects, anything picklable.
NOTE: It will not reflect changes in mutable objects without re-writing them to the database, or use writeback=True.
If your data is less than hundreds of MB – just use a dict and pickle it.
shelve presents a dict interface:
import shelve
d = shelve.open(filename)
d[key] = data # store data at key
data = d[key] # retrieve a COPY of data at key
del d[key] # delete data stored at key
flag = d.has_key(key) # true if the key exists
d.close() # close it
It uses pickle under the hood, so the same security issues.
LAB
Here are two datasets embedded in Python:
And:
They can be loaded with:
from add_book_data import AddressBook
They have address book data – one with a nested dict, one “flat”. Use the nested version for this exercise.
Write a module that saves the data as python literals in a file
and reads it back in
Write a module that saves the data as a pickle in a file
and reads it back in
Write a module that saves the data in a shelve
and accesses it one by one.
Write some tests to make sure its working!
Interchange Formats
These are formats suitable for interchanging data with other systems – written in arbitrary other languages.
In other words: standard formats.
INI
INI files, also known as the old Windows config files.
[Section1]
int = 15
bool = true
float = 3.1415
[Section2]
int = 32
...
Good for configuration data, etc.
ConfigParser
The configparser module provides tools for working with INI files.
Writing ini Files
import configparser
config = configparser.ConfigParser()
config.add_section('Section1')
config.set('Section1', 'an_integer', '15')
config.set('Section1', 'a_boolean', 'true')
config.set('Section1', 'a_float', '3.1415')
# Writing our configuration file to 'example.cfg'
config.write(open('example.cfg', 'w'))
Note: all keys and values are strings.
Reading ini Files
>>> config = configparser.ConfigParser()
>>> config.read('example.cfg')
>>> config.sections()
['Section1']
>>> config.get('Section1', 'a_float')
'3.1415'
>>> config.items('Section1')
[('an_integer', '15'), ('a_boolean', 'true'), ('a_float', '3.1415')]
CSV
CSV (Comma Separated Values) format is the most common import and export format for spreadsheets and databases.
There’s no real standard – the Python csv package more or less follows MS Excel “standard” with other “dialects” available.
It’s possible to use delimiters other than commas like tabs or pipes.
It’s most useful for simple tabular data.
Reading CSV Files
This uses: eggs.csv
In [14]: import csv
In [17]: spam_reader = csv.reader(open('eggs.csv'),
skipinitialspace=True)
In [19]: for row in spam_reader:
....: print(row)
['Spam', ' Spam', ' Spam', ' Spam', ' Spam', ' Baked Beans']
['Spam', ' Lovely Spam', ' Wonderful Spam']
The csv module takes care of string quoting, etc. for you.
This is a pretty big deal – that can be a real pain!
NOTE: skipinitialspace is False by default, which can mess up interpreting quotes correctly.
Writing CSV Files
>>> import csv
>>> with open('eggs2.csv', 'w') as outfile:
>>> spam_writer = csv.writer(outfile,
quoting=csv.QUOTE_MINIMAL)
>>> spam_writer.writerow(['Spam'] * 5 + ['Baked Beans'])
>>> spam_writer.writerow(['Spam', 'Lovely Spam', 'Wonderful Spam'])
>>> spam_writer.writerow(['Spam', 'Spam, Wonderful spam..', 'Very-Wonderful Spam'])
The csv module takes care of string quoting, etc. for you.
You can set the quoting attribute on the dialect object to control that.
JSON
JSON (JavaScript Object Notation) is a subset of JavaScript syntax used as a lightweight data interchange format.
LOTS of systems can read JSON – notably browsers.
The Python module has an interface similar to pickle.
It can handle the standard Python data types but it does have issues with anything beyond the basic Python data types.
It’s commonly used for configuration files, etc.
Python json Module
In [93]: import json
In [94]: s = json.dumps(data)
Out[95]: '[{"this": 5, "that": 4}, {"eggs": 3.4, "spam": 7},
{"foo": 86, "bar": 4.5}, {"fun": 43, "baz": 6.5}]'
In [96]: data2 = json.loads(s)
Out[97]:
[{u'that': 4, u'this': 5},
{u'eggs': 3.4, u'spam': 7},
...
In [98]: data2 == data
Out[98]: True # they are the same
See also json.dump() and json.load() for working with files directly.
NOTE: JSON is less “rich” than python – no tuples, no distinction between integers and floats, and no comments! Keys can only be strings.
LAB
Use the same addressbook data:
# load with:
from add_book_data import AddressBook
Write a module that saves the data as an INI file
and reads it back in
Write a module that saves the data as a CSV file
and reads it back in
You’ll need the “flat” version for this lab.
Write a module that saves the data in JSON
and reads it back in
XML
XML is a standardized version of SGML, designed for use as a data storage / interchange format.
NOTE: HTML is also SGML, and modern versions conform to the XML standard.
XML in the python std lib
xml.dom
xml.sax
xml.parsers.expat
xml.etree
elementtree
elementtree is the simplest tool. It maps pretty directly to XML.
The Element type is a flexible container object, designed to store hierarchical data structures in memory.
Essentially an in-memory XML that can be read from/written to XML
An ElementTree is an entire XML doc
An Element is a node in that tree.
https://docs.python.org/3/library/xml.etree.elementtree.html
Write a module that saves the data in XML
and reads it back in
this gets ugly!
Databases
A database is a system for storing and retrieving data – usually in a filesystem.
We usually think RDBMS and SQL – but there are simpler systems.
dbm
dbm is a generic interface to variants of the DBM database that is suitable for storing data that fits well into a python dict with strings as both keys and values.
Note: dbm will use the dbm system that works on your system. This may be different on different system. So the db files may NOT be compatible! whichdb will try to figure it out, but it’s not guaranteed.
https://docs.python.org/3/library/dbm.html
NOTE: dbm is getting pretty old fashioned – e.g. it doesn’t handle Unicode. It’s here for completeness, but there are probably better options.
The dbm Module
Writing data:
# creating a dbm file
import dbm
dbm.open(filename, 'n')
Flag options are:
‘r’ – Open existing database for reading only (default)
‘w’ – Open existing database for reading and writing
‘c’ – Open database for reading and writing, creating it if it doesn’t exist
‘n’ – Always create a new, empty database, open for reading and writing
Caution – these are different than the regular file open modes!
dbm provides a dict-like interface:
import dbm
db = dbm.open("dbm", "c")
db["first"] = "bruce"
db["second"] = "micheal"
db["third"] = "fred"
db["second"] = "john" # overwrite
db.close()
# read it:
db = dbm.open("dbm", "r")
for key in db.keys():
print(key, db[key])
This is a lot like shelve, though theoretically compatible with other systems.
sqlite
SQLite is a C library providing a lightweight disk-based single-file database. It provides a non-standard variant of the SQL query language.
It is very broadly used as as an embedded databases for storing application-specific data etc.
Python sqlite Module
The sqlite3 Python module wraps the C library and provides standard DB-API interface. It allows and requires SQL queries.
It can provide high performance, flexible, portable storage for your app.
Example
import sqlite3
# open a connection to a db file:
conn = sqlite3.connect('example.db')
# or build one in-memory
conn = sqlite3.connect(':memory:')
# create a cursor
c = conn.cursor()
Execute SQL With the Cursor
# Create table
c.execute("CREATE TABLE stocks (date text, trans text, symbol text, qty real, price real)")
# Insert a row of data
c.execute("INSERT INTO stocks VALUES ('2006-01-05','BUY','RHAT',100,35.14)")
# Save (commit) the changes
conn.commit()
# Close the cursor if we are done with it
c.close()
SELECT creates a cursor that can be iterated:
>>> for row in c.execute('SELECT * FROM stocks ORDER BY price'):
print row
('2006-01-05', 'BUY', 'RHAT', 100, 35.14)
('2006-03-28', 'BUY', 'IBM', 1000, 45.0)
...
Or you can get the rows one by one or in a list:
c.fetchone()
c.fetchall()
Good idea to use the DB-API’s parameter substitution to avoid SQL injection security bugs:
t = (symbol,)
c.execute('SELECT * FROM stocks WHERE symbol=?', t)
print c.fetchone()
# Larger example that inserts many records at a time
purchases = [('2006-03-28', 'BUY', 'IBM', 1000, 45.00),
('2006-04-05', 'BUY', 'MSFT', 1000, 72.00),
('2006-04-06', 'SELL', 'IBM', 500, 53.00),
]
c.executemany('INSERT INTO stocks VALUES (?,?,?,?,?)', purchases)
DB-API
The DB-API spec – PEP 249 is a specification for interaction between Python and Relational Databases.
It has support for a large number of third-party Database drivers:
MySQL
PostgreSQL
Oracle
MSSQL
LAB Extras
A few more things you could do:
Use pickle to save/reload a custom class of yours. Perhaps the Circle class from the first quarter?
Try writing a json writer for a non-standard data type. A custom class, or a more complex built-in?
Other Options
There are a lot of other possibilities outside the standard lib.
Object-Relation Mappers
There are systems for mapping Python objects to tables.
These save you writing that glue code and SQL.
These usually deal with mapping to variety of back-ends. For example, it’s a common pattern to test locally with SQLite and then deploy to PostgreSQL or MySQL.
SQLAlchemy
Django ORM
Object Databases
We will be talking more about this in another class: No SQL Databases
These directly store and retrieve Python Objects.
They’re kind of like shelve, but more flexible, and they give extra features like full text searching, etc.
NoSQL
Document-Oriented Storage
MongoDB (BSON interface, JSON documents)
CouchDB (Apache):
JSON documents
Javascript querying (MapReduce)
HTTP API
LAB
Load data with:
from add_book_data import AddressBook
Write a module that saves the data in a dbm database and then reads it back in.
Write a module that saves the data in an SQLite database and then reads it back in.
Optional:
Do the same with an ORM of your choice.