Transaction Processor Tutorial Python


This tutorial covers the creation of a new Sawtooth transaction family in Python, based on the Sawtooth SDK. We will construct a transaction handler which implements a distributed version of the multi-player game tic- tac-toe.


The SDK contains a fully-implemented version of tic-tac-toe. This tutorial is meant to demonstrate the relevant concepts, rather than to create a complete implementation. See the SDK for full implementations in multiple languages.

A general description of tic-tac-toe, including the rules, can be found on Wikipedia at:

A full implementation of the tic-tac-toe transaction family can be found in



This tutorial assumes that you have gone through Installing and Running Sawtooth and are familiar with the concepts introduced there.

You should be familiar with the concepts introduced in the Installing and Running Sawtooth guide and have a working Sawtooth environment prior to completing this tutorial.

The Transaction Processor

There are two top-level components of a transaction processor: a processor class and a handler class. The SDK provides a general-purpose processor class. The handler class is application-dependent and contains the business logic for a particular family of transactions. Multiple handlers can be connected to an instance of the processor class.

Handlers get called in two ways:

  1. An apply method
  2. Various “metadata” methods

The metadata is used to connect the handler to the processor, and we’ll discuss it at the end of this tutorial. The bulk of the handler, however, is made up of apply and its helper functions, so that’s where we’ll start.

The apply Method

apply gets called with two arguments, transaction and context. The argument transaction is an instance of the class Transaction that is created from the protobuf definition. Also, context is an instance of the class Context from the python SDK.

transaction holds the command that is to be executed (e.g. taking a space or creating a game), while context stores information about the current state of the game (e.g. the board layout and whose turn it is).

The transaction contains payload bytes that are opaque to the validator core, and transaction family specific. When implementing a transaction handler the binary serialization protocol is up to the implementer.

Without yet getting into the details of how this information is encoded, we can start to think about what apply needs to do. apply needs to

  1. unpack the command data from the transaction,
  2. retrieve the game data from the context,
  3. play the game, and
  4. save the updated game data.

Accordingly, a top-down approach to apply might look like this:

def apply(self, transaction, context):
    signer, game_name, action, space = \

    board, state, player1, player2 = \
        self._get_state_data(game_name, context)

    updated_game_data = self._play_xo(
        board, state,
        player1, player2,
        signer, action, space

    self._store_game_data(game_name, updated_game_data, context)

Note that the third step is the only one that actually concerns tic-tac-toe; the other three steps all concern the coordination of data.



Transactions and Batches contains a detailed description of how transactions are structured and used. Please read this document before proceeding, if you have not reviewed it.

So how do we get data out of the transaction? The transaction consists of a header and a payload. The header contains the “signer”, which is used to identify the current player. The payload will contain an encoding of the game name, the action (‘create’ a game, ‘take’ a space), and the space (which will be an empty string if the action isn’t ‘take’). So our _unpack_transaction function will look like this:

def _unpack_transaction(self, transaction):
    header = transaction.header
    signer = header.signer

        game_name, action, space = self._decode_data(transaction.payload)
        raise InvalidTransaction("Invalid payload serialization")

    return signer, game_name, action, space

Before we say how exactly the transaction payload will be decoded, let’s look at _get_state_data. Now, as far as the handler is concerned, it doesn’t matter how the game data is stored. The only thing that matters is that given a game name, the state store is able to give back the correct game data. (In our full XO implementation, the game data is stored in a Merkle-Radix tree.)

def _get_state_data(self, game_name, context):
    game_address = self._make_game_address(game_name)

    state_entries = context.get_state([game_address])

        return self._decode_data(state_entries[0].data)
    except IndexError:
        return None, None, None, None
        raise InternalError("Failed to deserialize game data.")

By convention, we’ll store game data at an address obtained from hashing the game name prepended with some constant:

def _make_game_address(self, game_name):
    prefix = self._namespace_prefix
    game_name_utf8 = game_name.encode('utf-8')
    return prefix + hashlib.sha512(game_name_utf8).hexdigest()[0:64]

Finally, we’ll store the game data. To do this, we simply need to encode the updated state of the game and store it back at the address from which it came.

def _store_game_data(self, game_name, game_data, context):
    game_address = self._make_game_address(game_name)

    encoded_game_data = self._encode_data(game_data)

    addresses = context.set_state(
        {game_address: encoded_game_data}

    if len(addresses) < 1:
        raise InternalError("State Error")

So, how should we encode and decode the data? We have many options in binary encoding schemes; the binary data stored in the validator state is up to the implementer of the handler. In this case, we’ll encode the data as a simple UTF-8 comma-separated value string, but we could use something more sophisticated, BSON.

def _decode_data(self, data):
    return data.decode().split(',')

def _encode_data(self, data):
    return ','.join(data).encode()

Implementing Game Play

Game-play functionality can be implemented in different ways. For our implementation, see the _play_xo function in sawtooth-core/sdk/examples/xo_python/sawtooth_xo/processor/ We choose to represent the board as a string of length 9, with each character in the string representing a space taken by X, a space taken by O, or a free space.

The XoTransactionHandler Class

All that’s left to do is set up the XoTransactionHandler class and its metadata. The metadata is used to register the transaction processor with a validator by sending it information about what kinds of transactions it can handle.

class XoTransactionHandler:
    def __init__(self, namespace_prefix):
        self._namespace_prefix = namespace_prefix

    def family_name(self):
        return 'xo'

    def family_versions(self):
        return ['1.0']

    def encodings(self):
        return ['csv-utf8']

    def namespaces(self):
        return [self._namespace_prefix]

    def apply(self, transaction, context):
        # ...