个人对drf-extentions 的英文文档的部分整理翻译与保存（个人用）

Posted 2021-11-27 daysn

Caching

To cache something is to save the result of an expensive calculation so that you don‘t have to perform the calculation next time. Here‘s some pseudocode explaining how this would work for a dynamically generated api response:

given a URL, try finding that API response in the cache
if the response is in the cache:
    return the cached response
else:
    generate the response
    save the generated response in the cache (for next time)
    return the generated response
缓存一些东西是为了保存一个昂贵计算的结果，这样下次就不必执行计算了。下面是一些伪代码，解释了这将如何为动态生成的api响应工作:

给定一个URL，
如果响应在缓存中，尝试在缓存中找到该API响应返回缓存的响应:否则生成响应并将生成的响应保存到缓存中(下次使用)然后返回

Cache response

DRF-extensions allows you to cache api responses with simple @cache_responsedecorator. There are two requirements for decorated method:

• It should be method of class which is inherited from rest_framework.views.APIView
• It should return rest_framework.response.Response instance

翻译：DRF-extensions允许您使用简单的@cache_responsedecorator缓存api响应。装饰方法有两个要求:

它应该是类的方法，继承自rest_frame .views. apiview

它应该返回rest_frame .response。响应实例

Usage example:

from rest_framework.response import Response
from rest_framework import views
from rest_framework_extensions.cache.decorators import (
    cache_response
)
from myapp.models import City

class CityView(views.APIView):
    @cache_response()
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

If you request view first time you‘ll get it from processed SQL query. (~60ms response time):

# Request
GET /cities/ HTTP/1.1
Accept: application/json

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8

[‘Moscow‘, ‘London‘, ‘Paris‘]

Second request will hit the cache. No sql evaluation, no database query. (~30 ms response time):

# Request
GET /cities/ HTTP/1.1
Accept: application/json

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8

[‘Moscow‘, ‘London‘, ‘Paris‘]

Reduction in response time depends on calculation complexity inside your API method. Sometimes it reduces from 1 second to 10ms, sometimes you win just 10ms.

Timeout

You can specify cache timeout in seconds, providing first argument:

class CityView(views.APIView):
    @cache_response(60 * 15)
    def get(self, request, *args, **kwargs):
        ...

In the above example, the result of the get() view will be cached for 15 minutes.

If you don‘t specify cache_timeout argument then value from REST_FRAMEWORK_EXTENSIONS settings will be used. By default it‘s None, which means "cache forever". You can change this default in settings:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_CACHE_RESPONSE_TIMEOUT‘: 60 * 15
}

Usage of the specific cache

New in DRF-extensions 0.2.3

@cache_response can also take an optional keyword argument, cache, which directs the decorator to use a specific cache (from your CACHES setting) when caching results. By default, the default cache will be used, but you can specify any cache you want:

class CityView(views.APIView):
    @cache_response(60 * 15, cache=‘special_cache‘)
    def get(self, request, *args, **kwargs):
        ...

You can specify what cache to use by default in settings:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_USE_CACHE‘: ‘special_cache‘
}

Cache key

By default every cached data from @cache_response decorator stored by key, which calculated with DefaultKeyConstructor.

You can change cache key by providing key_func argument, which must be callable:

def calculate_cache_key(view_instance, view_method,
                        request, args, kwargs):
    return ‘.‘.join([
        len(args),
        len(kwargs)
    ])

class CityView(views.APIView):
    @cache_response(60 * 15, key_func=calculate_cache_key)
    def get(self, request, *args, **kwargs):
        ...

You can implement view method and use it for cache key calculation by specifying key_func argument as string:

class CityView(views.APIView):
    @cache_response(60 * 15, key_func=‘calculate_cache_key‘)
    def get(self, request, *args, **kwargs):
        ...

    def calculate_cache_key(self, view_instance, view_method,
                            request, args, kwargs):
        return ‘.‘.join([
            len(args),
            len(kwargs)
        ])

Key calculation function will be called with next parameters:

• view_instance - view instance of decorated method
• view_method - decorated method
• request - decorated method request
• args - decorated method positional arguments
• kwargs - decorated method keyword arguments

Default key function

If @cache_response decorator used without key argument then default key function will be used. You can change this function in settings:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_CACHE_KEY_FUNC‘:
      ‘rest_framework_extensions.utils.default_cache_key_func‘
}

default_cache_key_func uses DefaultKeyConstructor as a base for key calculation.

Caching errors

New in DRF-extensions 0.2.7

By default every response is cached, even failed. For example:

class CityView(views.APIView):
    @cache_response()
    def get(self, request, *args, **kwargs):
        raise Exception("500 error comes from here")

First request to CityView.get will fail with 500 status code error and next requests to this endpoint will return 500 error from cache.

You can change this behaviour by turning off caching error responses:

class CityView(views.APIView):
    @cache_response(cache_errors=False)
    def get(self, request, *args, **kwargs):
        raise Exception("500 error comes from here")

You can change default behaviour by changing DEFAULT_CACHE_ERRORS setting:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_CACHE_ERRORS‘: False
}

CacheResponseMixin

It is common to cache standard viewset retrieve and list methods. That is why CacheResponseMixin exists. Just mix it into viewset implementation and those methods will use functions, defined in REST_FRAMEWORK_EXTENSIONS settings:

• "DEFAULT_OBJECT_CACHE_KEY_FUNC" for retrieve method
• "DEFAULT_LIST_CACHE_KEY_FUNC" for list method

By default those functions are using DefaultKeyConstructor and extends it:

• With RetrieveSqlQueryKeyBit for "DEFAULT_OBJECT_CACHE_KEY_FUNC"
• With ListSqlQueryKeyBit and PaginationKeyBit for "DEFAULT_LIST_CACHE_KEY_FUNC"

You can change those settings for custom cache key generation:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_OBJECT_CACHE_KEY_FUNC‘:
      ‘rest_framework_extensions.utils.default_object_cache_key_func‘,
    ‘DEFAULT_LIST_CACHE_KEY_FUNC‘:
      ‘rest_framework_extensions.utils.default_list_cache_key_func‘,
}

Mixin example usage:

from myapps.serializers import UserSerializer
from rest_framework_extensions.cache.mixins import CacheResponseMixin

class UserViewSet(CacheResponseMixin, viewsets.ModelViewSet):
    serializer_class = UserSerializer

You can change cache key function by providing object_cache_key_func orlist_cache_key_func methods in view class:

class UserViewSet(CacheResponseMixin, viewsets.ModelViewSet):
    serializer_class = UserSerializer

    def object_cache_key_func(self, **kwargs):
        return ‘some key for object‘

    def list_cache_key_func(self, **kwargs):
        return ‘some key for list‘

Ofcourse you can use custom key constructor:

from yourapp.key_constructors import (
    CustomObjectKeyConstructor,
    CustomListKeyConstructor,
)

class UserViewSet(CacheResponseMixin, viewsets.ModelViewSet):
    serializer_class = UserSerializer
    object_cache_key_func = CustomObjectKeyConstructor()
    list_cache_key_func = CustomListKeyConstructor()

If you want to cache only retrieve method then you could use rest_framework_extensions.cache.mixins.RetrieveCacheResponseMixin.

If you want to cache only list method then you could use rest_framework_extensions.cache.mixins.ListCacheResponseMixin.

Key constructor

As you could see from previous section cache key calculation might seem fairly simple operation. But let‘s see next example. We make ordinary HTTP request to cities resource:

# Request
GET /cities/ HTTP/1.1
Accept: application/json

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8

[‘Moscow‘, ‘London‘, ‘Paris‘]

By the moment all goes fine - response returned and cached. Let‘s make the same request requiring XML response:

# Request
GET /cities/ HTTP/1.1
Accept: application/xml

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8

[‘Moscow‘, ‘London‘, ‘Paris‘]

What is that? Oh, we forgot about format negotiations. We can add format to key bits:

def calculate_cache_key(view_instance, view_method,
                        request, args, kwargs):
    return ‘.‘.join([
        len(args),
        len(kwargs),
        request.accepted_renderer.format  # here it is
    ])

# Request
GET /cities/ HTTP/1.1
Accept: application/xml

# Response
HTTP/1.1 200 OK
Content-Type: application/xml; charset=UTF-8

<?xml version="1.0" encoding="utf-8"?>
<root>
    <list-item>Moscow</list-item>
    <list-item>London</list-item>
    <list-item>Paris</list-item>
</root>

That‘s cool now - we have different responses for different formats with different cache keys. But there are many cases, where key should be different for different requests:

• Response format (json, xml);
• User (exact authorized user or anonymous);
• Different request meta data (request.META[‘REMOTE_ADDR‘]);
• Language (ru, en);
• Headers;
• Query params. For example, jsonp resources need callback param, which rendered in response;
• Pagination. We should show different data for different pages;
• Etc...

Of course we can use custom calculate_cache_key methods and reuse them for different API methods, but we can‘t reuse just parts of them. For example, one method depends on user id and language, but another only on user id. How to be more DRYish? Let‘s see some magic:

from rest_framework_extensions.key_constructor.constructors import (
    KeyConstructor
)
from rest_framework_extensions.key_constructor import bits
from your_app.utils import get_city_by_ip

class CityGetKeyConstructor(KeyConstructor):
    unique_method_id = bits.UniqueMethodIdKeyBit()
    format = bits.FormatKeyBit()
    language = bits.LanguageKeyBit()

class CityHeadKeyConstructor(CityGetKeyConstructor):
    user = bits.UserKeyBit()
    request_meta = bits.RequestMetaKeyBit(params=[‘REMOTE_ADDR‘])

class CityView(views.APIView):
    @cache_response(key_func=CityGetKeyConstructor())
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

    @cache_response(key_func=CityHeadKeyConstructor())
    def head(self, request, *args, **kwargs):
        city = ‘‘
        user = self.request.user
        if user.is_authenticated() and user.city:
            city = Response(user.city.name)
        if not city:
            city = get_city_by_ip(request.META[‘REMOTE_ADDR‘])
        return Response(city)

Firstly, let‘s revise CityView.get method cache key calculation. It constructs from 3 bits:

• unique_method_id - remember that default key calculation? Here it is. Just one of the cache key bits. head method has different set of bits and they can‘t collide with get method bits. But there could be another view class with the same bits.
• format - key would be different for different formats.
• language - key would be different for different languages.

The second method head has the same unique_method_id, format and language bits, buts extends with 2 more:

• user - key would be different for different users. As you can see in response calculation we use request.user instance. For different users we need different responses.
• request_meta - key would be different for different ip addresses. As you can see in response calculation we are falling back to getting city from ip address if couldn‘t get it from authorized user model.

All default key bits are listed in this section.

Default key constructor

DefaultKeyConstructor is located in rest_framework_extensions.key_constructor.constructors module and constructs a key from unique method id, request format and request language. It has next implementation:

class DefaultKeyConstructor(KeyConstructor):
    unique_method_id = bits.UniqueMethodIdKeyBit()
    format = bits.FormatKeyBit()
    language = bits.LanguageKeyBit()

How key constructor works

Key constructor class works in the same manner as the standard django forms and key bits used like form fields. Lets go through key construction steps for DefaultKeyConstructor.

Firstly, constructor starts iteration over every key bit:

• unique_method_id
• format
• language

Then constructor gets data from every key bit calling method get_data:

• unique_method_id - u‘your_app.views.SometView.get‘
• format - u‘json‘
• language - u‘en‘

Every key bit get_data method is called with next arguments:

• view_instance - view instance of decorated method
• view_method - decorated method
• request - decorated method request
• args - decorated method positional arguments
• kwargs - decorated method keyword arguments

After this it combines every key bit data to one dict, which keys are a key bits names in constructor, and values are returned data:

{
    ‘unique_method_id‘: u‘your_app.views.SometView.get‘,
    ‘format‘: u‘json‘,
    ‘language‘: u‘en‘
}

Then constructor dumps resulting dict to json:

‘{"unique_method_id": "your_app.views.SometView.get", "language": "en", "format": "json"}‘

And finally compresses json with md5 and returns hash value:

‘b04f8f03c89df824e0ecd25230a90f0e0ebe184cf8c0114342e9471dd2275baa‘

Custom key bit

We are going to create a simple key bit which could be used in real applications with next properties:

• High read rate
• Low write rate

The task is - cache every read request and invalidate all cache data after write to any model, which used in API. This approach let us don‘t think about granular cache invalidation - just flush it after any model instance change/creation/deletion.

Lets create models:

# models.py
from django.db import models

class Group(models.Model):
    title = models.CharField()

class Profile(models.Model):
    name = models.CharField()
    group = models.ForeignKey(Group)

Define serializers:

# serializers.py
from yourapp.models import Group, Profile
from rest_framework import serializers

class GroupSerializer(serializers.ModelSerializer):
    class Meta:
        model = Group

class ProfileSerializer(serializers.ModelSerializer):
    group = GroupSerializer()

    class Meta:
        model = Profile

Create views:

# views.py
from yourapp.serializers import GroupSerializer, ProfileSerializer
from yourapp.models import Group, Profile

class GroupViewSet(viewsets.ReadOnlyModelViewSet):
    serializer_class = GroupSerializer
    queryset = Group.objects.all()

class ProfileViewSet(viewsets.ReadOnlyModelViewSet):
    serializer_class = ProfileSerializer
    queryset = Profile.objects.all()

And finally register views in router:

# urls.py
from yourapp.views import GroupViewSet,ProfileViewSet

router = DefaultRouter()
router.register(r‘groups‘, GroupViewSet)
router.register(r‘profiles‘, ProfileViewSet)
urlpatterns = router.urls

At the moment we have API, but it‘s not cached. Lets cache it and create our custom key bit:

# views.py
import datetime
from django.core.cache import cache
from django.utils.encoding import force_text
from yourapp.serializers import GroupSerializer, ProfileSerializer
from rest_framework_extensions.cache.decorators import cache_response
from rest_framework_extensions.key_constructor.constructors import (
    DefaultKeyConstructor
)
from rest_framework_extensions.key_constructor.bits import (
    KeyBitBase,
    RetrieveSqlQueryKeyBit,
    ListSqlQueryKeyBit,
    PaginationKeyBit
)

class UpdatedAtKeyBit(KeyBitBase):
    def get_data(self, **kwargs):
        key = ‘api_updated_at_timestamp‘
        value = cache.get(key, None)
        if not value:
            value = datetime.datetime.utcnow()
            cache.set(key, value=value)
        return force_text(value)

class CustomObjectKeyConstructor(DefaultKeyConstructor):
    retrieve_sql = RetrieveSqlQueryKeyBit()
    updated_at = UpdatedAtKeyBit()

class CustomListKeyConstructor(DefaultKeyConstructor):
    list_sql = ListSqlQueryKeyBit()
    pagination = PaginationKeyBit()
    updated_at = UpdatedAtKeyBit()

class GroupViewSet(viewsets.ReadOnlyModelViewSet):
    serializer_class = GroupSerializer

    @cache_response(key_func=CustomObjectKeyConstructor())
    def retrieve(self, *args, **kwargs):
        return super(GroupViewSet, self).retrieve(*args, **kwargs)

    @cache_response(key_func=CustomListKeyConstructor())
    def list(self, *args, **kwargs):
        return super(GroupViewSet, self).list(*args, **kwargs)

class ProfileViewSet(viewsets.ReadOnlyModelViewSet):
    serializer_class = ProfileSerializer

    @cache_response(key_func=CustomObjectKeyConstructor())
    def retrieve(self, *args, **kwargs):
        return super(ProfileViewSet, self).retrieve(*args, **kwargs)

    @cache_response(key_func=CustomListKeyConstructor())
    def list(self, *args, **kwargs):
        return super(ProfileViewSet, self).list(*args, **kwargs)

As you can see UpdatedAtKeyBit just adds to key information when API models has been update last time. If there is no information about it then new datetime will be used for key bit data.

Lets write cache invalidation. We just connect models to standard signals and change value in cache by key api_updated_at_timestamp:

# models.py
import datetime
from django.db import models
from django.db.models.signals import post_save, post_delete

def change_api_updated_at(sender=None, instance=None, *args, **kwargs):
    cache.set(‘api_updated_at_timestamp‘, datetime.datetime.utcnow())

class Group(models.Model):
    title = models.CharField()

class Profile(models.Model):
    name = models.CharField()
    group = models.ForeignKey(Group)

for model in [Group, Profile]:
    post_save.connect(receiver=change_api_updated_at, sender=model)
    post_delete.connect(receiver=change_api_updated_at, sender=model)

And that‘s it. When any model changes then value in cache by key api_updated_at_timestamp will be changed too. After this every key constructor, that used UpdatedAtKeyBit, will construct new keys and @cache_response decorator will cache data in new places.

Key constructor params

New in DRF-extensions 0.2.3

You can change params attribute for specific key bit by providing params dict for key constructor initialization function. For example, here is custom key constructor, which inherits from DefaultKeyConstructor and adds geoip key bit:

class CityKeyConstructor(DefaultKeyConstructor):
    geoip = bits.RequestMetaKeyBit(params=[‘GEOIP_CITY‘])

If you wanted to use GEOIP_COUNTRY, you could create new key constructor:

class CountryKeyConstructor(DefaultKeyConstructor):
    geoip = bits.RequestMetaKeyBit(params=[‘GEOIP_COUNTRY‘])

But there is another way. You can send params in key constructor initialization method. This is the dict attribute, where keys are bit names and values are bit params attribute value (look at CountryView):

class CityKeyConstructor(DefaultKeyConstructor):
    geoip = bits.RequestMetaKeyBit(params=[‘GEOIP_COUNTRY‘])

class CityView(views.APIView):
    @cache_response(key_func=CityKeyConstructor())
    def get(self, request, *args, **kwargs):
        ...

class CountryView(views.APIView):
    @cache_response(key_func=CityKeyConstructor(
        params={‘geoip‘: [‘GEOIP_COUNTRY‘]}
    ))
    def get(self, request, *args, **kwargs):
        ...

If there is no item provided for key bit then default key bit params value will be used.

Constructor‘s bits list

You can dynamically change key constructor‘s bits list in initialization method by altering bits attribute:

class CustomKeyConstructor(DefaultKeyConstructor):
    def __init__(self, *args, **kwargs):
        super(CustomKeyConstructor, self).__init__(*args, **kwargs)
        self.bits[‘geoip‘] = bits.RequestMetaKeyBit(
            params=[‘GEOIP_CITY‘]
        )

Default key bits

Out of the box DRF-extensions has some basic key bits. They are all located in rest_framework_extensions.key_constructor.bits module.

FormatKeyBit

Retrieves format info from request. Usage example:

class MyKeyConstructor(KeyConstructor):
    format = FormatKeyBit()

LanguageKeyBit

Retrieves active language for request. Usage example:

class MyKeyConstructor(KeyConstructor):
    user = LanguageKeyBit()

UserKeyBit

Retrieves user id from request. If it is anonymous then returnes "anonymous" string. Usage example:

class MyKeyConstructor(KeyConstructor):
    user = UserKeyBit()

RequestMetaKeyBit

Retrieves data from request.META dict. Usage example:

class MyKeyConstructor(KeyConstructor):
    ip_address_and_user_agent = bits.RequestMetaKeyBit(
        [‘REMOTE_ADDR‘, ‘HTTP_USER_AGENT‘]
    )

You can use * for retrieving all meta data to key bit:

New in DRF-extensions 0.2.7

class MyKeyConstructor(KeyConstructor):
    all_request_meta = bits.RequestMetaKeyBit(‘*‘)

HeadersKeyBit

Same as RequestMetaKeyBit retrieves data from request.META dict. The difference is that HeadersKeyBit allows to use normal header names:

class MyKeyConstructor(KeyConstructor):
    user_agent_and_geobase_id = bits.HeadersKeyBit(
        [‘user-agent‘, ‘x-geobase-id‘]
    )
    # will process request.META[‘HTTP_USER_AGENT‘] and
    #              request.META[‘HTTP_X_GEOBASE_ID‘]

You can use * for retrieving all headers to key bit:

New in DRF-extensions 0.2.7

class MyKeyConstructor(KeyConstructor):
    all_headers = bits.HeadersKeyBit(‘*‘)

ArgsKeyBit

New in DRF-extensions 0.2.7

Retrieves data from the view‘s positional arguments. A list of position indices can be passed to indicate which arguments to use. For retrieving all arguments you can use *which is also the default value:

class MyKeyConstructor(KeyConstructor):
    args = bits.ArgsKeyBit()  # will use all positional arguments

class MyKeyConstructor(KeyConstructor):
    args = bits.ArgsKeyBit(‘*‘)  # same as above

class MyKeyConstructor(KeyConstructor):
    args = bits.ArgsKeyBit([0, 2])

KwargsKeyBit

New in DRF-extensions 0.2.7

Retrieves data from the views‘s keyword arguments. A list of keyword argument names can be passed to indicate which kwargs to use. For retrieving all kwargs you can use *which is also the default value:

class MyKeyConstructor(KeyConstructor):
    kwargs = bits.KwargsKeyBit()  # will use all keyword arguments

class MyKeyConstructor(KeyConstructor):
    kwargs = bits.KwargsKeyBit(‘*‘)  # same as above

class MyKeyConstructor(KeyConstructor):
    kwargs = bits.KwargsKeyBit([‘user_id‘, ‘city‘])

QueryParamsKeyBit

Retrieves data from request.GET dict. Usage example:

class MyKeyConstructor(KeyConstructor):
    part_and_callback = bits.QueryParamsKeyBit(
        [‘part‘, ‘callback‘]
    )

You can use * for retrieving all query params to key bit which is also the default value:

New in DRF-extensions 0.2.7

class MyKeyConstructor(KeyConstructor):
    all_query_params = bits.QueryParamsKeyBit(‘*‘)  # all qs parameters

class MyKeyConstructor(KeyConstructor):
    all_query_params = bits.QueryParamsKeyBit()  # same as above

PaginationKeyBit

Inherits from QueryParamsKeyBit and returns data from used pagination params.

class MyKeyConstructor(KeyConstructor):
    pagination = bits.PaginationKeyBit()

ListSqlQueryKeyBit

Retrieves sql query for view.filter_queryset(view.get_queryset()) filtering.

class MyKeyConstructor(KeyConstructor):
    list_sql_query = bits.ListSqlQueryKeyBit()

RetrieveSqlQueryKeyBit

Retrieves sql query for retrieving exact object.

class MyKeyConstructor(KeyConstructor):
    retrieve_sql_query = bits.RetrieveSqlQueryKeyBit()

UniqueViewIdKeyBit

Combines data about view module and view class name.

class MyKeyConstructor(KeyConstructor):
    unique_view_id = bits.UniqueViewIdKeyBit()

UniqueMethodIdKeyBit

Combines data about view module, view class name and view method name.

class MyKeyConstructor(KeyConstructor):
    unique_view_id = bits.UniqueMethodIdKeyBit()

Conditional requests

This documentation section uses information from RESTful Web Services Cookbook 10-th chapter.

Conditional HTTP request allows API clients to accomplish 2 goals:

• Conditional HTTP GET saves client and server time and bandwidth.
• For unsafe requests such as PUT, POST, and DELETE, conditional requests provide concurrency control.

HTTP Etag

An ETag or entity tag, is part of HTTP, the protocol for the World Wide Web. It is one of several mechanisms that HTTP provides for web cache validation, and which allows a client to make conditional requests. - Wikipedia

For etag calculation and conditional request processing you should use rest_framework_extensions.etag.decorators.etag decorator. It‘s similar to native django decorator.

from rest_framework_extensions.etag.decorators import etag

class CityView(views.APIView):
    @etag()
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

By default @etag would calculate header value with the same algorithm as cache keydefault calculation performs.

# Request
GET /cities/ HTTP/1.1
Accept: application/json

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8
ETag: "e7b50490dc546d116635a14cfa58110306dd6c5434146b6740ec08bf0a78f9a2"

[‘Moscow‘, ‘London‘, ‘Paris‘]

You can define custom function for Etag value calculation with etag_func argument:

from rest_framework_extensions.etag.decorators import etag

def calculate_etag(view_instance, view_method,
                   request, args, kwargs):
    return ‘.‘.join([
        len(args),
        len(kwargs)
    ])

class CityView(views.APIView):
    @etag(etag_func=calculate_etag)
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

You can implement view method and use it for Etag calculation by specifying etag_funcargument as string:

from rest_framework_extensions.etag.decorators import etag

class CityView(views.APIView):
    @etag(etag_func=‘calculate_etag_from_method‘)
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

    def calculate_etag_from_method(self, view_instance, view_method,
                                   request, args, kwargs):
        return ‘.‘.join([
            len(args),
            len(kwargs)
        ])

Etag calculation function will be called with next parameters:

• view_instance - view instance of decorated method
• view_method - decorated method
• request - decorated method request
• args - decorated method positional arguments
• kwargs - decorated method keyword arguments

Default etag function

If @etag decorator used without etag_func argument then default etag function will be used. You can change this function in settings:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_ETAG_FUNC‘:
      ‘rest_framework_extensions.utils.default_etag_func‘
}

default_etag_func uses DefaultKeyConstructor as a base for etag calculation.

Usage with caching

As you can see @etag and @cache_response decorators has similar key calculation approaches. They both can take key from simple callable function. And more then this - in many cases they share the same calculation logic. In the next example we use both decorators, which share one calculation function:

from rest_framework_extensions.etag.decorators import etag
from rest_framework_extensions.cache.decorators import cache_response
from rest_framework_extensions.key_constructor import bits
from rest_framework_extensions.key_constructor.constructors import (
    KeyConstructor
)

class CityGetKeyConstructor(KeyConstructor):
    format = bits.FormatKeyBit()
    language = bits.LanguageKeyBit()

class CityView(views.APIView):
    key_constructor_func = CityGetKeyConstructor()

    @etag(key_constructor_func)
    @cache_response(key_func=key_constructor_func)
    def get(self, request, *args, **kwargs):
        cities = City.objects.all().values_list(‘name‘, flat=True)
        return Response(cities)

Note the decorators order. First goes @etag and after goes @cache_response. We want firstly perform conditional processing and after it response processing.

There is one more point for it. If conditional processing didn‘t fail then key_constructor_func would be called again in @cache_response. But in most cases first calculation is enough. To accomplish this goal you could use KeyConstructor initial argument memoize_for_request:

>>> key_constructor_func = CityGetKeyConstructor(memoize_for_request=True)
>>> request1, request1 = ‘request1‘, ‘request2‘
>>> print key_constructor_func(request=request1)  # full calculation
request1-key
>>> print key_constructor_func(request=request1)  # data from cache
request1-key
>>> print key_constructor_func(request=request2)  # full calculation
request2-key
>>> print key_constructor_func(request=request2)  # data from cache
request2-key

By default memoize_for_request is False, but you can change it in settings:

REST_FRAMEWORK_EXTENSIONS = {
    ‘DEFAULT_KEY_CONSTRUCTOR_MEMOIZE_FOR_REQUEST‘: True
}

It‘s important to note that this memoization is thread safe.

Saving time and bandwith

When a server returns ETag header, you should store it along with the representation data on the client. When making GET and HEAD requests for the same resource in the future, include the If-None-Match header to make these requests "conditional".

For example, retrieve all cities:

# Request
GET /cities/ HTTP/1.1
Accept: application/json

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8
ETag: "some_etag_value"

[‘Moscow‘, ‘London‘, ‘Paris‘]

If you make same request with If-None-Match and there is the cached value for this request, then server will respond with 304 status code without body data.

# Request
GET /cities/ HTTP/1.1
Accept: application/json
If-None-Match: some_etag_value

# Response
HTTP/1.1 304 NOT MODIFIED
Content-Type: application/json; charset=UTF-8
Etag: "some_etag_value"

After this response you can use existing cities data on the client.

Concurrency control

Concurrency control ensures the correct processing of data under concurrent operations by clients. There are two ways to implement concurrency control:

• Pessimistic concurrency control. In this model, the client gets a lock, obtains the current state of the resource, makes modifications, and then releases the lock. During this process, the server prevents other clients from acquiring a lock on the same resource. Relational databases operate in this manner.
• Optimistic concurrency control. In this model, the client first gets a token. Instead of obtaining a lock, the client attempts a write operation with the token included in the request. The operation succeeds if the token is still valid and fails otherwise.

HTTP, being a stateless application control, is designed for optimistic concurrency control.

                                PUT
                                 |
                          +-------------+
                          |  Etag       |
                          |  supplied?  |
                          +-------------+
                           |           |
                          Yes          No
                           |           |
        +--------------------+       +-----------------------+
        |  Do preconditions  |       |  Does the             |
        |  match?            |       |  resource exist?      |
        +--------------------+       +-----------------------+
            |           |                   |              |
            Yes         No                  Yes            No
            |           |                   |              |
+--------------+  +--------------------+  +-------------+  |
|  Update the  |  |  412 Precondition  |  |  403        |  |
|  resource    |  |  failed            |  |  Forbidden  |  |
+--------------+  +--------------------+  +-------------+  |
                                                           |
                                     +-----------------------+
                                     |  Can clients          |
                                     |  create resources     |
                                     +-----------------------+
                                           |           |
                                          Yes          No
                                           |           |
                                     +-----------+   +-------------+
                                     |  201      |   |  404        |
                                     |  Created  |   |  Not Found  |
                                     +-----------+   +-------------+

Delete:

                               DELETE
                                 |
                          +-------------+
                          |  Etag       |
                          |  supplied?  |
                          +-------------+
                           |           |
                          Yes          No
                           |           |
        +--------------------+       +-------------+
        |  Do preconditions  |       |  403        |
        |  match?            |       |  Forbidden  |
        +--------------------+       +-------------+
            |           |
            Yes         No
            |           |
+--------------+  +--------------------+
|  Delete the  |  |  412 Precondition  |
|  resource    |  |  failed            |
+--------------+  +--------------------+

Here is example of implementation for all CRUD methods (except create, because it doesn‘t need concurrency control) wrapped with etag decorator:

from rest_framework.viewsets import ModelViewSet
from rest_framework_extensions.key_constructor import bits
from rest_framework_extensions.key_constructor.constructors import (
    KeyConstructor
)

from your_app.models import City
from your_app.key_bits import UpdatedAtKeyBit

class CityListKeyConstructor(KeyConstructor):
    format = bits.FormatKeyBit()
    language = bits.LanguageKeyBit()
    pagination = bits.PaginationKeyBit()
    list_sql_query = bits.ListSqlQueryKeyBit()
    unique_view_id = bits.UniqueViewIdKeyBit()

class CityDetailKeyConstructor(KeyConstructor):
    format = bits.FormatKeyBit()
    language = bits.LanguageKeyBit()
    retrieve_sql_query = bits.RetrieveSqlQueryKeyBit()
    unique_view_id = bits.UniqueViewIdKeyBit()
    updated_at = UpdatedAtKeyBit()

class CityViewSet(ModelViewSet):
    list_key_func = CityListKeyConstructor(
        memoize_for_request=True
    )
    obj_key_func = CityDetailKeyConstructor(
        memoize_for_request=True
    )

    @etag(list_key_func)
    @cache_response(key_func=list_key_func)
    def list(self, request, *args, **kwargs):
        return super(CityViewSet, self).list(request, *args, **kwargs)

    @etag(obj_key_func)
    @cache_response(key_func=obj_key_func)
    def retrieve(self, request, *args, **kwargs):
        return super(CityViewSet, self).retrieve(request, *args, **kwargs)

    @etag(obj_key_func)
    def update(self, request, *args, **kwargs):
        return super(CityViewSet, self).update(request, *args, **kwargs)

    @etag(obj_key_func)
    def destroy(self, request, *args, **kwargs):
        return super(CityViewSet, self).destroy(request, *args, **kwargs)

Etag for unsafe methods

From previous section you could see that unsafe methods, such update (PUT, PATCH) or destroy (DELETE), have the same @etag decorator wrapping manner as the safe methods.

But every unsafe method has one distinction from safe method - it changes the data which could be used for Etag calculation. In our case it is UpdatedAtKeyBit. It means that we should calculate Etag:

• Before building response - for If-Match and If-None-Match conditions validation
• After building response (if necessary) - for clients

@etag decorator has special attribute rebuild_after_method_evaluation, which by default is False.

If you specify rebuild_after_method_evaluation as True then Etag will be rebuilt after method evaluation:

class CityViewSet(ModelViewSet):
    ...
    @etag(obj_key_func, rebuild_after_method_evaluation=True)
    def update(self, request, *args, **kwargs):
        return super(CityViewSet, self).update(request, *args, **kwargs)

    @etag(obj_key_func)
    def destroy(self, request, *args, **kwargs):
        return super(CityViewSet, self).destroy(request, *args, **kwargs)

# Request
PUT /cities/1/ HTTP/1.1
Accept: application/json

{"name": "London"}

# Response
HTTP/1.1 200 OK
Content-Type: application/json; charset=UTF-8
ETag: "4e63ef056f47270272b96523f51ad938b5ea141024b767880eac047d10a0b339"

{
  id: 1,
  name: "London"
}

As you can see we didn‘t specify rebuild_after_method_evaluation for destroymethod. That is because there is no sense to use returned Etag value on clients if object deletion already performed.

With rebuild_after_method_evaluation parameter Etag calculation for PUT/PATCHmethod would look like:

             +--------------+
             |    Request   |
             +--------------+
                    |
       +--------------------------+
       |  Calculate Etag          |
       |  for condition matching  |
       +--------------------------+
                    |
          +--------------------+
          |  Do preconditions  |
          |  match?            |
          +--------------------+
              |           |
              Yes         No
              |           |
  +--------------+  +--------------------+
  |  Update the  |  |  412 Precondition  |
  |  resource    |  |  failed            |
  +--------------+  +--------------------+
         |
+--------------------+
|  Calculate Etag    |
|  again and add it  |
|  to response       |
+--------------------+
         |
   +------------+
   |  Return    |
   |  response  |
   +------------+

If-None-Match example for DELETE method:

# Request
DELETE /cities/1/ HTTP/1.1
Accept: application/json
If-None-Match: some_etag_value

# Response
HTTP/1.1 304 NOT MODIFIED
Content-Type: application/json; charset=UTF-8
Etag: "some_etag_value"

If-Match example for DELETE method:

# Request
DELETE /cities/1/ HTTP/1.1
Accept: application/json
If-Match: another_etag_value

# Response
HTTP/1.1 412 PRECONDITION FAILED
Content-Type: application/json; charset=UTF-8
Etag: "some_etag_value"