Gfrktyl

I work on a small data team where we developed this tool when we began experiencing several job failures on our legacy big data systems. We have several data lakes, each with their own API, and until now had no contingencies for issues centering around query failures. Furthermore, our database sockets had no central repository for accessing our data. This tool brings all query platforms together in a (relatively) seamless experience by way of formulating a socket library and offers redundancy protocols when queries fail. This software is compatible with any SQL-similar interface that has an established Pythonic API, such as Impala, Hive, AWS, and terminal/command-line returns.

The usage is very straightforward:

# Upon instantiation:

# > mquery() imports all sockets and aliases via the factory

# > Default attempts on any socket is 8

# > Default socket is the first socket in the library

# > Default precedence is all sockets of the same type as the default socket

# > Default for fail_on_empty and fail_on_zero are both False



# Functional

a = mquery_lite().query(r"SELECT 'test;")   # Returns the result set only; object instance is destructed on next mquery_lite() call

print(str(a.get_result()))



# Functional with overrides

a = mquery_lite().query(r"SELECT 'test;", query_attempts=2, fail_on_zero=True)   # Returns the result set only, overriding the default query_attempts and fail_on_zero values

print(str(a.get_result()))



# Object Oriented with Overrides

a = mquery_lite(socket_precedence=["pd","data"])   # Precedence is set to both the pandas and all data sockets

a.query_attempts = 4

a.query(r"SELECT 'test';", query_attempts=2, socket_precedence=["pd"])   # Overrides on query submission

# query_attempts returns to 4 and socket_precedence returns to ["pd","data"]

print(str(a.get_result()))



# Object Oriented with Overrides and Handshake

a = mquery_lite(socket_precedence=["all","data"], query_attempts=4)   # Precedence list will be the list of all sockets plus the additional list of all data sockets; there will be duplicates in the precedence list (not problematic)

a.handshake()   # Test each socket in the current precedence list; retain only those that pass. Duplicates will persist.

a.query(r"SELECT 'test';", query_attempts=2)

# query_attempts returns to 4 

print(str(a.get_result()))

Naturally, all proprietary information has been redacted from this post. I've also stripped most of the bells and whistles (e.g. typecasting property decorators) to demonstrate basic functionality and proof of concept in the code below; all pieces of this code can readily be found from open sources on the internet, but I haven't seen them tied together in this manner.

The mquery_lite() object is the primary process of this code and can either be used functionally or instanced as an object. When called, mquery_lite() determines if a query was provided - if so, it will instance itself, perform the query, then return the result pointer from the successful socket. If a query is not passed, mquery_lite() remains instanced and user-modified settings are retained.

Sockets are imported by way of a generator-encapsulated factory. Their aliases are mapped in a separate library for ease of use when calling sockets. Sockets are separated by type, defined in the socket itself (we prefer to group by the expected output of the socket as this ensures consistent output on query failure; e.g. data frame, list of lists, generator, etc.). Sockets retain the query results until a new query is submitted.

The socket and alias libraries are automatically built on instantiation, based on the order in which they are present in the script. Collisions are rectified on a first-come-first-serve basis. The following object variables are created on instantiation:

• 
  query_attempts (default 8) is the number of attempts mquery_lite() will make on a socket before moving to the next socket. An exponential timer (2^n) sets the pause between repeat queries on a socket.


• 
  socket_default (default None) is the socket that will be substituted in the precedence list when an unknown alias is provided. Will default to the first socket in the library if None is detected.


• 
  socket_precedence (default ) is the order in which sockets will be attempted. Will default to all sockets of the same type as the default socket in the library if None is detected.


• 
  fail_on_empty (default False) indicates if a query should raise an exception if it comes back empty (useful for command queries).


• 
  fail_on_zero (default False) indicates if a query should raise an exception if it comes back zero (useful for counts).

Results remain and failures occur at the socket level. Handling of permitted errors (raised from sockets) occurs in the .query() method.

import pandas

import pyodbc

import time



def peek(x):

    try:

        return next(x)

    except StopIteration:

        return None



###############################################################################

### Dynamic Polymorphic Socket Factory ########################################

###############################################################################

class PermittedSocketError(Exception):

    """

    A socket error that should trigger a retry, but not a program termination.

    """

    pass



class socket:

    DSN = "DSN=Your.DSN.Info.Here;"   # Used in pyodbc and pandas sockets

    def handshake(self, query="SELECT 'test';"):

        self.execute(query, fail_on_empty=False, fail_on_zero=False)



# Dynamic socket factory

# https://python-3-patterns-idioms-test.readthedocs.io/en/latest/Factory.html

class factory:

    objects = {}

    def add_factory(id, factory):

        factory.factories.put[id] = factory

    add_factory = staticmethod(add_factory)

    def create_object(id):

        if id not in factory.objects:   # updated for Python 3

            factory.objects[id] = eval(id + ".factory()")

        return factory.objects[id].create()

    create_object = staticmethod(create_object)



###############################################################################

### Socket Library ############################################################

###############################################################################

class generic_socket(socket):

    socket_aliases = ["alias_1", "alias_2"]

    socket_type = "type"



    @property

    def result(self):

        # Any type of return handling can go here (such as a generator to improve post-parsing)

        return self.__data_block



    def execute(self, query, fail_on_empty, fail_on_zero):

        # Set up for query

        self.__data_block = None



        try:

            # Execute query



            # Internal post query handling of error codes should raise exceptions here - useful for non-Pythonic (e.g. command-line) returns

            # Likely not needed if using processes with full Pythonic exception handling

            if /*Permitted Error Behavior*/:

                raise PermittedSocketError("[msg] ")

            else:

                raise



            if fail_on_empty and /*Check if Empty*/:

                raise PermittedSocketError("Empty return detected.")

            if fail_on_zero and /*Check if Zero*/:

                raise PermittedSocketError("Zero return detected.")



# Edit: The if-else statements above were note syntactically valid and should be changed to:

# if fail_on_empty and /*Check if Empty*/:

#     raise PermittedSocketError("Empty return detected.")

# if fail_on_zero and /*Check if Zero*/:

#     raise PermittedSocketError("Zero return detected.")

# if /*Permitted Error Behavior*/:

#     raise PermittedSocketError("[msg] ")

# if /*Non-Permitted Error Behavior*/:

#     raise Exception



        # Exterior post query handling of permitted socket errors - Pythonic exceptions should be caught here

        except PermittedSocketError:

            # Permitted error post-process, such as reinitializing security protocols or invalidating metadata

            # Permitted errors are re-raised and handled within mquery_lite()

            raise



    class factory:

        def create(self):

            return generic_socket()



class pandas_socket(socket):

    socket_aliases = ["pandas","pd"]

    socket_type = "data"



    @property

    def result(self):

        return self.data_block



    def execute(self, query, fail_on_empty, fail_on_zero):

        self.data_block = None

        try:

            connection = pyodbc.connect(self.DSN, autocommit=True)

            self.data_block = pandas.read_sql(query, connection)

            connection.close()

            if fail_on_empty and self.data_block.dropna().empty:

                raise PermittedSocketError("Empty return detected.")

            if fail_on_zero and self.data_block.shape == (1,1) and int(float(self.data_block.iloc[0,0])) == 0:

                raise PermittedSocketError("Zero return detected.")

        except PermittedSocketError:

            raise



    class factory:

        def create(self):

            return pandas_socket()





class pyodbc_socket(socket):

    socket_aliases = ["pyodbc"]

    socket_type = "standard"



    @property

    def result(self):

        return self.data_block



    def execute(self, query, fail_on_empty, fail_on_zero):

        self.data_block = None

        try:

            connection = pyodbc.connect(self.DSN, autocommit=True)

            cursor = connection.cursor()

            cursor.execute(query)

            self.data_block = cursor.fetchall()

            cursor.close()

            connection.close()

            row = peek(iter(self.data_block))

            if fail_on_empty and not row:

                raise PermittedSocketError("Empty return detected.")

            if fail_on_zero and len(row) == 1 and  peek(iter(row)) in (0, "0"):

                raise PermittedSocketError("Zero return detected.")

        except pyodbc.ProgrammingError:

            # Thrown when .fetchall() returns nothing

            self.__data_block = [()]

            raise PermittedSocketError("Empty return detected.")

        except PermittedSocketError:

            raise



    class factory:

        def create(self):

            return pyodbc_socket()



############################################################################### 

### mquery_lite() #############################################################

###############################################################################

class mquery_lite(object):

    def __new__(cls, query=None, query_attempts=8, socket_default=None, socket_precedence=, fail_on_empty=False, fail_on_zero=False):

        # https://howto.lintel.in/python-__new__-magic-method-explained/

        if query is not None:

            mquery_instance = super(mquery_lite, cls).__new__(cls)

            mquery_instance.__init__(query_attempts, socket_default, socket_precedence, fail_on_empty, fail_on_zero)

            mquery_instance.query(query)

            return mquery_instance.get_results()

        else:

            return super(mquery_lite, cls).__new__(cls)



    ### CTOR

    def __init__(self, query_attempts=8, socket_default=None, socket_precedence=, fail_on_empty=False, fail_on_zero=False):



        ### Socket Library

        self.socket_library = {socket.__name__:factory.create_object(socket.__name__) for socket in socket.__subclasses__()}

        self.socket_aliases = ({socket:[socket] for socket in self.socket_library})

        self.socket_aliases.update({alias:[socket] for socket in self.socket_library for alias in self.socket_library[socket].socket_aliases if alias not in self.socket_aliases})

        self.socket_aliases.update({socket_type:[socket for socket in self.socket_library if self.socket_library[socket].socket_type == socket_type] for socket_type in {self.socket_library[unique_socket_type].socket_type for unique_socket_type in self.socket_library}})

        self.socket_aliases.update({"all":[socket for socket in self.socket_library]})



        self.query_attempts:int = query_attempts

        self.socket_default:str = socket_default

        if socket_default is None:

            self.socket_default = next(iter(self.socket_library))

        self.socket_precedence = socket_precedence

        if socket_precedence == :

            self.socket_precedence:list = self.socket_aliases[self.socket_library[self.socket_default].socket_type]

        self.fail_on_empty:bool = fail_on_empty

        self.fail_on_zero:bool = fail_on_empty



    def handshake(self):

        precedence_candidates = 

        for alias in self.socket_precedence:

            for socket in self.socket_aliases[alias]:

                try:

                    self.socket_library[socket].handshake()

                    precedence_candidates.append(socket)

                except PermittedSocketError:

                    continue

        if len(precedence_candidates) != 0:

            self.socket_precedence = precedence_candidates



    def get_results(self):

        return self.result_socket.result



    ### Query Execution

    def query(self, query, query_attempts=None, socket_precedence=, fail_on_empty=None, fail_on_zero=None):

        # Overrides

        if query_attempts is None:

            query_attempts = self.query_attempts

        if socket_precedence==:

            for i in self.socket_precedence:

                for j in self.socket_aliases[i]:

                    if j in self.socket_library:

                        socket_precedence.append(j)

                    else:

                        socket_precedence.append(self.default_socket)

        else:

            candidate_precedence = socket_precedence[:]

            socket_precedence = 

            for i in candidate_precedence:

                for j in self.socket_aliases[i]:

                    if j in self.socket_library:

                        socket_precedence.append(j)

                    else:

                        socket_precedence.append(self.default_socket)

        if fail_on_empty is None: fail_on_empty = self.fail_on_empty

        if fail_on_zero is None: fail_on_empty = self.fail_on_zero



        # Loop through socket precedence list

        for socket in socket_precedence:

            try:

                # Loop through socket attempts on current socket

                for attempt_n in range(query_attempts):

                    try:

                        # Exponential timer; pauses 2^n seconds on the current socket

                        if attempt_n > 0:

                            print("Waiting " + str(2**attempt_n) + " seconds before reattempting...")

                            for k in range(2**attempt_n): time.sleep(1)

                        # Query attempt

                        self.socket_library[socket].execute(query, fail_on_empty, fail_on_zero)

                        self.result_socket = self.socket_library[socket]

                        return



                    except PermittedSocketError:

                        print("mquery() failed on socket "" + str(socket) + "".")

                        if attempt_n+1 == query_attempts:

                            raise

                        pass



            except PermittedSocketError:

                if socket == socket_precedence[-1]:

                    print("mquery() failed after trying all attempts on all sockets.")

                    raise

                print("mquery() failed after all attempts on socket "" + str(socket) + ""; moving to next socket.")

                continue

Questions are mostly along the lines of: We've tried to make this as "Pythonic" as possible - have we missed anything? Are there libraries that already perform this in a more efficient manner?

By PEP8, your classes sockets, factory, etc. should be capitalized. You also need newlines between your class methods. These can all be fixed fairly easily with the use of a stand-alone or IDE-builtin linter.

This comment:

# Dynamic socket factory

# https://python-3-patterns-idioms-test.readthedocs.io/en/latest/Factory.html

should be moved to a docstring on the inside of the class:

class Factory:

    """Dynamic socket factory

    https://python-3-patterns-idioms-test.readthedocs.io/en/latest/Factory.html

    """

So far as I can tell, socket is an abstract base class. You can pull in abc if you want to, but at the very least, you should explicitly show execute as being a "pure virtual" (in C++ parlance) method:

def execute(self, query, fail_on_empty, fail_on_zero):

    raise NotImplementedError()

This code:

        if /*Permitted Error Behavior*/:

            raise PermittedSocketError("[msg] ")

        else:

            raise

can lose the else, because the previous block has already raised.

I'm not sure what's happened here - whether it's a redaction, or what - but it doesn't look syntactically valid:

        if fail_on_empty and /*Check if Empty*/:

            raise PermittedSocketError("Empty return detected.")

        if fail_on_zero and /*Check if Zero*/:

            raise PermittedSocketError("Zero return detected.")

Your except PermittedSocketError: and its accompanying comment Permitted errors are re-raised are a little odd. If the error is permitted, and you're doing nothing but re-raising, why have the try block in the first place?

This:

    else:

        return super(mquery_lite, cls).__new__(cls)

doesn't need an else, for the same reason as that raise I've described above.

The series of list comprehensions seen after ### Socket Library really needs to have most or all of those broken up onto multiple lines, especially when you have multiple for.

self.socket_library is a dictionary, but I'm unclear on your usage. You have this:

                if j in self.socket_library:

                    socket_precedence.append(j)

                else:

                    socket_precedence.append(self.default_socket)

Is your intention to look through the keys of socket_library, ignore the values, and add present keys to socket_precedence? If you want to use its values, this needs to change.