cx_Oracle and Exception Handling - Good practices?

victorhooi picture victorhooi · Sep 19, 2011 · Viewed 57.5k times · Source

I'm trying to use cx_Oracle to connect to an Oracle instance and execute some DDL statements:

db = None
try:
    db = cx_Oracle.connect('username', 'password', 'hostname:port/SERVICENAME')
#print(db.version)
except cx_Oracle.DatabaseError as e:
    error, = e.args
    if error.code == 1017:
        print('Please check your credentials.')
        # sys.exit()?
    else:
        print('Database connection error: %s'.format(e))
cursor = db.cursor()
try:
    cursor.execute(ddl_statements)
except cx_Oracle.DatabaseError as e:
    error, = e.args
    if error.code == 955:
        print('Table already exists')
    if error.code == 1031:
        print("Insufficient privileges - are you sure you're using the owner account?")
    print(error.code)
    print(error.message)
    print(error.context)
cursor.close()
db.commit()
db.close()

However, I'm not quite sure what's the best design for exception handling here.

Firstly, I create the db object inside a try block, to catch any connection errors.

However, if it can't connect, then db won't exist further down - which is why I set db = None above. However, is that good practice?

Ideally, I need to catch errors with connecting, then errors with running the DDL statements, and so on.

Is nesting exceptions a good idea? Or is there a better way of dealing with dependent/cascaded exceptions like this?

Also, there are some parts (e.g. connection failures) where I'd like the script to just terminate - hence the commented out sys.exit() call. However, I've heard that using exception handling for flow control like this is bad practice. Thoughts?

Answer

Ben picture Ben · Mar 24, 2012

However, if it can't connect, then db won't exist further down - which is why I set db = None above. However, is that good practice?

No, setting db = None is not best practice. There are two possibilities, either connecting to the database will work or it won't.

  • Connecting to the database doesn't work:

    As the raised exception has been caught and not re-raised you continue until you reach cursor = db.Cursor().

    db == None, so, an exception that resembles TypeError: 'NoneType' object has no attribute 'Cursor' will be raised. As the exception generated when the database connection failed has already been caught, the reason for the failure is disguised.

    Personally, I'd always raise a connection exception unless you're going to try again shortly. How you catch it is up to you; if the error persists I e-mail to say "go and check the database".

  • Connecting to the database does work:

    The variable db is assigned in your try:... except block. If the connect method does work then db is replaced with the connection object.

Either way the initial value of db is never used.

However, I've heard that using exception handling for flow control like this is bad practice.

Unlike other languages Python does use exception handling for flow control. At the end of my answer I've linked to several questions on Stack Overflow and Programmers that ask a similar question. In every example you'll see the words "but in Python".

That's not to say that you should go overboard but Python generally uses the mantra EAFP, "It's easier to ask forgiveness than permission." The top three voted examples in How do I check if a variable exists? are good examples of how you can both use flow control or not.

Is nesting exceptions a good idea? Or is there a better way of dealing with dependent/cascaded exceptions like this?

There's nothing wrong with nested exceptions, once again as long as you do it sanely. Consider your code. You could remove all exceptions and wrap the entire thing in a try:... except block. If an exception is raised then you know what it was, but it's a little harder to track down exactly what went wrong.

What then happens if you want to say e-mail yourself on the failure of cursor.execute? You should have an exception around cursor.execute in order to perform this one task. You then re-raise the exception so it's caught in your outer try:.... Not re-raising would result in your code continuing as if nothing had happened and whatever logic you had put in your outer try:... to deal with an exception would be ignored.

Ultimately all exceptions are inherited from BaseException.

Also, there are some parts (e.g. connection failures) where I'd like the script to just terminate - hence the commented out sys.exit() call.

I've added a simple class and how to call it, which is roughly how I would do what you're trying to do. If this is going to be run in the background then the printing of the errors isn't worthwhile - people won't be sitting there manually looking out for errors. They should be logged in whatever your standard way is and the appropriate people notified. I've removed the printing for this reason and replaced with a reminder to log.

As I've split the class out into multiple functions when the connect method fails and an exception is raised the execute call will not be run and the script will finish, after attempting to disconnect.

import cx_Oracle

class Oracle(object):

    def connect(self, username, password, hostname, port, servicename):
        """ Connect to the database. """

        try:
            self.db = cx_Oracle.connect(username, password
                                , hostname + ':' + port + '/' + servicename)
        except cx_Oracle.DatabaseError as e:
            # Log error as appropriate
            raise

        # If the database connection succeeded create the cursor
        # we-re going to use.
        self.cursor = self.db.cursor()

    def disconnect(self):
        """
        Disconnect from the database. If this fails, for instance
        if the connection instance doesn't exist, ignore the exception.
        """

        try:
            self.cursor.close()
            self.db.close()
        except cx_Oracle.DatabaseError:
            pass

    def execute(self, sql, bindvars=None, commit=False):
        """
        Execute whatever SQL statements are passed to the method;
        commit if specified. Do not specify fetchall() in here as
        the SQL statement may not be a select.
        bindvars is a dictionary of variables you pass to execute.
        """

        try:
            self.cursor.execute(sql, bindvars)
        except cx_Oracle.DatabaseError as e:
            # Log error as appropriate
            raise

        # Only commit if it-s necessary.
        if commit:
            self.db.commit()

Then call it:

if __name__ == "__main__":

    oracle = Oracle.connect('username', 'password', 'hostname'
                           , 'port', 'servicename')

    try:
        # No commit as you don-t need to commit DDL.
        oracle.execute('ddl_statements')

    # Ensure that we always disconnect from the database to avoid
    # ORA-00018: Maximum number of sessions exceeded. 
    finally:
        oracle.disconnect()

Further reading:

cx_Oracle documentation

Why not use exceptions as regular flow of control?
Is python exception handling more efficient than PHP and/or other languages?
Arguments for or against using try catch as logical operators