Extensibility guide#

All of Boto3’s resource and client classes are generated at runtime. This means that you cannot directly inherit and then extend the functionality of these classes because they do not exist until the program actually starts running.

However it is still possible to extend the functionality of classes through Boto3’s event system.

An introduction to the event system#

Boto3’s event system allows users to register a function to a specific event. Then once the running program reaches a line that emits that specific event, Boto3 will call every function registered to the event in the order in which they were registered. When Boto3 calls each of these registered functions, it will call each of them with a specific set of keyword arguments that are associated with that event. Then once the registered function is called, the function may modify the keyword arguments passed to that function or return a value. Here is an example of how the event system works:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function
def add_my_bucket(params, **kwargs):
    # Add the name of the bucket you want to default to.
    if 'Bucket' not in params:
        params['Bucket'] = 'mybucket'

# Register the function to an event
event_system.register('provide-client-params.s3.ListObjectsV2', add_my_bucket)

response = s3.list_objects_v2()

In this example, the handler add_my_bucket is registered such that the handler will inject the value 'mybucket' for the Bucket parameter whenever the list_objects_v2 client call is made without the Bucket parameter. Note that if the same list_objects_v2 call is made without the Bucket parameter and the registered handler, it will result in a validation error.

Here are the takeaways from this example:

  • All clients have their own event system that you can use to fire events and register functions. You can access the event system through the meta.events attribute on the client.

  • All functions registered to the event system must have **kwargs in the function signature. This is because emitting an event can have any number of keyword arguments emitted alongside it, and so if your function is called without **kwargs, its signature will have to match every keyword argument emitted by the event. This also allows for more keyword arguments to be added to the emitted event in the future without breaking existing handlers.

  • To register a function to an event, call the register method on the event system with the name of the event you want to register the function to and the function handle. Note that if you register the event after the event is emitted, the function will not be called unless the event is emitted again. In the example, the add_my_bucket handler was registered to the 'provide-client-params.s3.ListObjectsV2' event, which is an event that can be used to inject and modify parameters passed in by the client method. To read more about the event refer to provide-client-params

A hierarchical structure#

The event system also provides a hierarchy for registering events such that you can register a function to a set of events depending on the event name hierarchy.

An event name can have its own hierarchy by specifying . in its name. For example, take the event name 'general.specific.more_specific'. When this event is emitted, the registered functions will be called in the order from most specific to least specific registration. So in this example, the functions will be called in the following order:

  1. Functions registered to 'general.specific.more_specific'

  2. Functions registered to 'general.specific'

  3. Functions registered to 'general'

Here is a deeper example of how the event system works with respect to its hierarchical structure:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

def add_my_general_bucket(params, **kwargs):
    if 'Bucket' not in params:
        params['Bucket'] = 'mybucket'

def add_my_specific_bucket(params, **kwargs):
    if 'Bucket' not in params:
        params['Bucket'] = 'myspecificbucket'

event_system.register('provide-client-params.s3', add_my_general_bucket)
event_system.register('provide-client-params.s3.ListObjectsV2', add_my_specific_bucket)

list_obj_response = s3.list_objects_v2()
put_obj_response = s3.put_object(Key='mykey', Body=b'my body')

In this example, the list_objects_v2 method call will use the 'myspecificbucket' for the bucket instead of 'mybucket' because the add_my_specific_bucket method was registered to the 'provide-client-params.s3.ListObjectsV2' event which is more specific than the 'provide-client-params.s3' event. Thus, the add_my_specific_bucket function is called before the add_my_general_bucket function is called when the event is emitted.

However for the put_object call, the bucket used is 'mybucket'. This is because the event emitted for the put_object client call is 'provide-client-params.s3.PutObject' and the add_my_general_bucket method is called via its registration to 'provide-client-params.s3'. The 'provide-client-params.s3.ListObjectsV2' event is never emitted so the registered add_my_specific_bucket function is never called.

Wildcard matching#

Another aspect of Boto3’s event system is that it has the capability to do wildcard matching using the '*' notation. Here is an example of using wildcards in the event system:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

def add_my_wildcard_bucket(params, **kwargs):
    if 'Bucket' not in params:
        params['Bucket'] = 'mybucket'

event_system.register('provide-client-params.s3.*', add_my_wildcard_bucket)
response = s3.list_objects_v2()

The '*' allows you to register to a group of events without having to know the actual name of the event. This is useful when you have to apply the same handler in multiple places. Also note that if the wildcard is used, it must be isolated. It does not handle globbing with additional characters. So in the previous example, if the my_wildcard_function was registered to 'provide-client-params.s3.*objects', the handler would not be called because it will consider 'provide-client-params.s3.*objects' to be a specific event.

The wildcard also respects the hierarchical structure of the event system. If another handler was registered to the 'provide-client-params.s3' event, the add_my_wildcard_bucket would be called first because it is registered to 'provide-client-params.s3.*' which is more specific than the event 'provide-client.s3'.

Isolation of event systems#

The event system in Boto3 has the notion of isolation: all clients maintain their own set of registered handlers. For example if a handler is registered to one client’s event system, it will not be registered to another client’s event system:

import boto3

client1 = boto3.client('s3')
client2 = boto3.client('s3')

def add_my_bucket(params, **kwargs):
    if 'Bucket' not in params:
        params['Bucket'] = 'mybucket'

def add_my_other_bucket(params, **kwargs):
    if 'Bucket' not in params:
        params['Bucket'] = 'myotherbucket'

client1.meta.events.register(
    'provide-client-params.s3.ListObjectsV2', add_my_bucket)
client2.meta.events.register(
    'provide-client-params.s3.ListObjectsV2', add_my_other_bucket)

client1_response = client1.list_objects_v2()
client2_response = client2.list_objects_v2()

Thanks to the isolation of clients’ event systems, client1 will inject 'mybucket' for its list_objects_v2 method call while client2 will inject 'myotherbucket' for its list_objects_v2 method call because add_my_bucket was registered to client1 while add_my_other_bucket was registered to client2.

Boto3 specific events#

Boto3 emits a set of events that users can register to customize clients or resources and modify the behavior of method calls.

Here is a table of events that users of Boto3 can register handlers to. More information about each event can be found in the corresponding sections below:

Note

Events with a * in their order number are conditionally emitted while all others are always emitted. An explanation of all 3 conditional events is provided below.

2 * - creating-resource-class is emitted ONLY when using a service resource.

8 * - after-call is emitted when a successful API response is received.

9 * - after-call-error is emitted when an unsuccessful API response is received.

Event Name

Order

Emit Location

creating-client-class

1

Location

creating-resource-class

2 *

Location

provide-client-params

3

Location

before-call

4

Location

request-created

5

Location

before-send

6

Location

needs-retry

7

Location

after-call

8 *

Location

after-call-error

9 *

Location

Note

If any of the following keywords are included in an event’s full name, you’ll need to replace it with the corresponding value:

  • service-name - The value used to instantiate a client as in boto3.client('service-name').

  • operation-name - The underlying API operation name of the corresponding client method. To access the operation API name, retrieve the value from the client.meta.method_to_api_mapping dictionary using the name of the desired client method as the key.

  • resource-name - The name of the resource class such as ServiceResource.

creating-client-class#

Full Event Name:

'creating-client-class.service-name'

Description:

This event is emitted upon creation of the client class for a service. The client class for a service is not created until the first instantiation of the client class. Use this event for adding methods to the client class or adding classes for the client class to inherit from.

Keyword Arguments Emitted:
type class_attributes:

dict

param class_attributes:

A dictionary where the keys are the names of the attributes of the class and the values are the actual attributes of the class.

type base_classes:

list

param base_classes:

A list of classes that the client class will inherit from where the order of inheritance is the same as the order of the list.

Expected Return Value:

None

Example:

Here is an example of how to add a method to the client class:

from boto3.session import Session

def custom_method(self):
    print('This is my custom method')

def add_custom_method(class_attributes, **kwargs):
    class_attributes['my_method'] = custom_method

session = Session()
session.events.register('creating-client-class.s3', add_custom_method)

client = session.client('s3')
client.my_method()

This should output:

This is my custom method

Here is an example of how to add a new class for the client class to inherit from:

from boto3.session import Session

class MyClass(object):
    def __init__(self, *args, **kwargs):
        super(MyClass, self).__init__(*args, **kwargs)
        print('Client instantiated!')

def add_custom_class(base_classes, **kwargs):
    base_classes.insert(0, MyClass)

session = Session()
session.events.register('creating-client-class.s3', add_custom_class)

client = session.client('s3')

This should output:

Client instantiated!

creating-resource-class#

Full Event Name:

'creating-resource-class.service-name.resource-name'

Description:

This event is emitted upon creation of the resource class. The resource class is not created until the first instantiation of the resource class. Use this event for adding methods to the resource class or adding classes for the resource class to inherit from.

Keyword Arguments Emitted:
type class_attributes:

dict

param class_attributes:

A dictionary where the keys are the names of the attributes of the class and the values are the actual attributes of the class.

type base_classes:

list

param base_classes:

A list of classes that the resource class will inherit from where the order of inheritance is the same as the order of the list.

Expected Return Value:

None

Example:

Here is an example of how to add a method to a resource class:

from boto3.session import Session

def custom_method(self):
    print('This is my custom method')

def add_custom_method(class_attributes, **kwargs):
    class_attributes['my_method'] = custom_method

session = Session()
session.events.register('creating-resource-class.s3.ServiceResource',
                        add_custom_method)

resource = session.resource('s3')
resource.my_method()

This should output:

This is my custom method

Here is an example of how to add a new class for a resource class to inherit from:

from boto3.session import Session

class MyClass(object):
    def __init__(self, *args, **kwargs):
        super(MyClass, self).__init__(*args, **kwargs)
        print('Resource instantiated!')

def add_custom_class(base_classes, **kwargs):
    base_classes.insert(0, MyClass)

session = Session()
session.events.register('creating-resource-class.s3.ServiceResource',
                        add_custom_class)

resource = session.resource('s3')

This should output:

Resource instantiated!

provide-client-params#

Full Event Name:

'provide-client-params.service-name.operation-name'

Description:

This event is emitted before operation parameters are validated and built into the HTTP request that will be sent over the wire. Use this event to inject or modify parameters.

Keyword Arguments Emitted:
type params:

dict

param params:

A dictionary containing key value pairs consisting of the parameters passed through to the client method.

type model:

botocore.model.OperationModel

param model:

A model representing the underlying API operation of the client method.

Expected Return Value:

None or return a dict containing parameters to use when making the request.

Example:

Here is an example of how to inject a parameter using the event:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function
def add_my_bucket(params, **kwargs):
    # Add the name of the bucket you want to default to.
    if 'Bucket' not in params:
        params['Bucket'] = 'mybucket'

# Register the function to an event
event_system.register('provide-client-params.s3.ListObjectsV2', add_my_bucket)

response = s3.list_objects_v2()

before-call#

Full Event Name:

'before-call.service-name.operation-name'

Description:

This event is emitted just before creating and sending the HTTP request. Use this event for modifying various HTTP request components prior to the request being created. A response tuple may optionally be returned to trigger a short-circuit and prevent the request from being made. This is useful for testing and is how the botocore stubber mocks responses.

Keyword Arguments Emitted:
type model:

botocore.model.OperationModel

param model:

A model representing the underlying API operation of the client method.

type params:

dict

param params:

A dictionary containing key value pairs for various components of an HTTP request such as url_path, host_prefix, query_string, headers, body, and method.

type request_signer:

botocore.signers.RequestSigner

param request_signer:

An object to sign requests before they are sent over the wire using one of the authentication mechanisms defined in auth.py.

Expected Return Value:

None or a tuple that includes both the botocore.awsrequest.AWSResponse and a dict that represents the parsed response described by the model.

Example:

Here is an example of how to add a custom header before making an API call:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function that adds a custom header and prints all headers.
def add_custom_header_before_call(model, params, request_signer, **kwargs):
    params['headers']['my-custom-header'] = 'header-info'
    headers = params['headers']
    print(f'param headers: {headers}')

#  Register the function to an event.
event_system.register('before-call.s3.ListBuckets', add_custom_header_before_call)

s3.list_buckets()

This should output:

param headers: { ... , 'my-custom-header': 'header-info'}

request-created#

Full Event Name:

'request-created.service-name.operation-name'

Description:

This event is emitted just after the request is created and triggers request signing.

Keyword Arguments Emitted:
type request:

botocore.awsrequest.AWSRequest

param request:

An AWSRequest object which represents the request that was created given some params and an operation model.

type operation_name:

str

param operation_name:

The name of the service operation model i.e. ListObjectsV2.

Expected Return Value:

None

Example:

Here is an example of how to inspect the request once it’s created:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function that prints the request information.
def inspect_request_created(request, operation_name, **kwargs):
    print('Request Info:')
    print(f'method: {request.method}')
    print(f'url: {request.url}')
    print(f'data: {request.data}')
    print(f'params: {request.params}')
    print(f'auth_path: {request.auth_path}')
    print(f'stream_output: {request.stream_output}')
    print(f'headers: {request.headers}')
    print(f'operation_name: {operation_name}')

# Register the function to an event
event_system.register('request-created.s3.ListObjectsV2', inspect_request_created)

response = s3.list_objects_v2(Bucket='my-bucket')

This should output:

Request Info:
method: GET
url: https://my-bucket.s3 ...
data: ...
params: { ... }
auth_path: ...
stream_output: ...
headers: ...
operation_name: ListObjectsV2

before-send#

Full Event Name:

'before-send.service-name.operation-name'

Description:

This event is emitted when the operation has been fully serialized, signed, and is ready to be sent over the wire. This event allows the finalized request to be inspected and allows a response to be returned that fulfills the request. If no response is returned botocore will fulfill the request as normal.

Keyword Arguments Emitted:
type request:

botocore.awsrequest.AWSPreparedRequest

param request:

A data class representing a finalized request to be sent over the wire.

Expected Return Value:

None or an instance of botocore.awsrequest.AWSResponse.

Example:

Here is an example of how to register a function that allows you to inspect the prepared request before it’s sent:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function that inspects the prepared request.
def inspect_request_before_send(request, **kwargs):
    print(f'request: {request}')

# Register the function to an event
event_system.register('before-send.s3.ListBuckets', inspect_request_before_send)

s3.list_buckets()

This should output:

request: <AWSPreparedRequest ... >

needs-retry#

Full Event Name:

'needs-retry.service-name.operation-name'

Description:

This event is emitted before checking if the most recent request needs to be retried. Use this event to define custom retry behavior when the configurable retry modes are not sufficient.

Keyword Arguments Emitted:
type response:

tuple

param response:

A tuple that includes both the botocore.awsrequest.AWSResponse and a dict that represents the parsed response described by the model.

type endpoint:

botocore.endpoint.Endpoint

param endpoint:

Represents an endpoint for a particular service.

type operation:

botocore.model.OperationModel

param operation:

A model representing the underlying API operation of the client method.

type attempts:

int

param attempts:

A number representing the amount of retries that have been attempted.

type caught_exception:

Exception | None

param caught_exception:

The exception raised after making an api call. If there was no exception, this will be None.

type request_dict:

dict

param request_dict:

A dictionary containing key value pairs for various components of an HTTP request such as url_path, host_prefix, query_string, headers, body, and method.

Expected Return Value:

Return None if no retry is needed, or return an int representing the retry delay in seconds.

Example:

Here is an example of how to add custom retry behavior:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a handler that determines retry behavior.
def needs_retry_handler(**kwargs):
    # Implement custom retry logic
    if some_condition:
        return None
    else:
        return some_delay

# Register the function to an event
event_system.register('needs-retry', needs_retry_handler)

s3.list_buckets()

after-call#

Full Event Name:

'after-call.service-name.operation-name'

Description:

This event is emitted just after the service client makes an API call. This event allows developers to postprocess or inspect the API response according to the specific requirements of their application if needed.

Keyword Arguments Emitted:
type http_response:

botocore.awsrequest.AWSResponse

param http_response:

A data class representing an HTTP response received from the server.

type parsed:

dict

param params:

A parsed version of the AWSResponse in the form of a python dictionary.

type model:

botocore.model.OperationModel

param model:

A model representing the underlying API operation of the client method.

Expected Return Value:

None

Example:

Here is an example that inspects args emitted from the after-call event:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Create a function that prints the after-call event args.
def print_after_call_args(http_response, parsed, model, **kwargs):
    print(f'http_response: {http_response}')
    print(f'parsed: {parsed}')
    print(f'model: {model.name}')

# Register the function to an event
event_system.register('after-call.s3.ListObjectsV2', print_after_call_args)

s3.list_objects_v2(Bucket='my-bucket')

This should output:

http_response: <botocore.awsrequest.AWSResponse object at ...>
parsed: { ... }
model: ListObjectsV2

after-call-error#

Full Event Name:

'after-call-error.service-name.operation-name'

Description:

This event is emitted upon receiving an error after making an API call. This event provides information about any errors encountered during the operation and allows listeners to take corrective actions if necessary.

Keyword Arguments Emitted:
type exception:

Exception

param exception:

The exception raised after making an api call.

Expected Return Value:

None

Example:

Here is an example we use the before-send to mimic a bad response which triggers the after-call-error event and prints the exception:

import boto3

s3 = boto3.client('s3')

# Access the event system on the S3 client
event_system = s3.meta.events

# Prints the detected exception.
def print_after_call_error_args(exception, **kwargs):
    if exception is not None:
        print(f'Exception Detected: {exception}')

# Mocks an exception raised when making an API call.
def list_objects_v2_bad_response(**kwargs):
    raise Exception("This is a test exception.")

event_system.register('before-send.s3.ListBuckets', list_objects_v2_bad_response)
event_system.register('after-call-error.s3.ListBuckets', print_after_call_error_args)

s3.list_buckets()

This should output:

Exception Detected: This is a test exception.
# Stack Trace
Exception: This is a test exception.