Django CQRS

)

django-cqrs is an Django application, that implements CQRS data synchronisation between several Django microservices.

CQRS

In Connect we have a rather complex Domain Model. There are many microservices, that are decomposed by subdomain and which follow database-per-service pattern. These microservices have rich and consistent APIs. They are deployed in cloud k8s cluster and scale automatically under load. Many of these services aggregate data from other ones and usually API Composition is totally enough. But, some services are working too slowly with API JOINS, so another pattern needs to be applied.

The pattern, that solves this issue is called CQRS - Command Query Responsibility Segregation. Core idea behind this pattern is that view databases (replicas) are defined for efficient querying and DB joins. Applications keep their replicas up to data by subscribing to Domain events published by the service that owns the data. Data is eventually consistent and that's okay for non-critical business transactions.

Documentation

Full documentation is available at https://django-cqrs.readthedocs.org.

Examples

Integration

• Setup RabbitMQ
• Install django-cqrs
• Apply changes to master service, according to RabbitMQ settings

# models.py

from django.db import models
from dj_cqrs.mixins import MasterMixin, RawMasterMixin


class Account(MasterMixin, models.Model):
    CQRS_ID = 'account'
    CQRS_PRODUCE = True  # set this to False to prevent sending instances to Transport
    
    
class Author(MasterMixin, models.Model):
    CQRS_ID = 'author'
    CQRS_SERIALIZER = 'app.api.AuthorSerializer'


# For cases of Diamond Multiinheritance the following approach could be used:
from mptt.models import MPTTModel
from dj_cqrs.metas import MasterMeta

class ComplexInheritanceModel(MPTTModel, RawMasterMixin):
    pass

MasterMeta.register(ComplexInheritanceModel)

# settings.py

CQRS = {
    'transport': 'dj_cqrs.transport.rabbit_mq.RabbitMQTransport',
    'host': RABBITMQ_HOST,
    'port': RABBITMQ_PORT,
    'user': RABBITMQ_USERNAME,
    'password': RABBITMQ_PASSWORD,
}

• Apply changes to replica service, according to RabbitMQ settings

from django.db import models
from dj_cqrs.mixins import ReplicaMixin


class AccountRef(ReplicaMixin, models.Model):
    CQRS_ID = 'account'
    
    id = models.IntegerField(primary_key=True)
    

class AuthorRef(ReplicaMixin, models.Model):
    CQRS_ID = 'author'
    CQRS_CUSTOM_SERIALIZATION = True
    
    @classmethod
    def cqrs_create(cls, sync, **mapped_data):
        # Override here
        pass
        
    def cqrs_update(self, sync, **mapped_data):
        # Override here
        pass

# settings.py

CQRS = {
    'transport': 'dj_cqrs.transport.RabbitMQTransport',
    'queue': 'account_replica',
    'host': RABBITMQ_HOST,
    'port': RABBITMQ_PORT,
    'user': RABBITMQ_USERNAME,
    'password': RABBITMQ_PASSWORD,
}

• Apply migrations on both services
• Run consumer worker on replica service. Management command: python manage.py cqrs_consume -w 2

Notes

• When there are master models with related entities in CQRS_SERIALIZER, it's important to have operations within atomic transactions. CQRS sync will happen on transaction commit.
• Please, avoid saving different instances of the same entity within transaction to reduce syncing and potential racing on replica side.
• Updating of related model won't trigger CQRS automatic synchronization for master model. This needs to be done manually.
• By default update_fields doesn't trigger CQRS logic, but it can be overridden for the whole application in settings:

settings.CQRS = {
    ...
    'master': {
        'CQRS_AUTO_UPDATE_FIELDS': True,
    },
    ...
}

or a special flag can be used in each place, where it's required to trigger CQRS flow:

instance.save(update_fields=['name'], update_cqrs_fields=True)

• When only needed instances need to be synchronized, there is a method is_sync_instance to set filtering rule. It's important to understand, that CQRS counting works even without syncing and rule is applied every time model is updated.

Example:

class FilteredSimplestModel(MasterMixin, models.Model):
    CQRS_ID = 'filter'

    name = models.CharField(max_length=200)

    def is_sync_instance(self):
        return len(str(self.name)) > 2

Utilities

Bulk synchronizer without transport (usage example: it may be used for initial configuration). May be used at planned downtime.

• On master service: python manage.py cqrs_bulk_dump --cqrs-id=author -> author.dump
• On replica service: python manage.py cqrs_bulk_load -i=author.dump

Filter synchronizer over transport (usage example: sync some specific records to a given replica). Can be used dynamically.

• To sync all replicas: python manage.py cqrs_sync --cqrs-id=author -f={"id__in": [1, 2]}
• To sync all instances only with one replica: python manage.py cqrs_sync --cqrs-id=author -f={} -q=replica

Set of diff synchronization tools ()

• To get diff and synchronize master service with replica service in K8S:

kubectl exec -i MASTER_CONTAINER -- python manage.py cqrs_diff_master --cqrs-id=author | 
    kubectl exec -i REPLICA_CONTAINER -- python manage.py cqrs_diff_replica |
    kubectl exec -i MASTER_CONTAINER -- python manage.py cqrs_diff_sync

• If it's important to check sync and clean up deleted objects within replica service in K8S:

kubectl exec -i REPLICA_CONTAINER -- python manage.py cqrs_deleted_diff_replica --cqrs-id=author | 
    kubectl exec -i MASTER_CONTAINER -- python manage.py cqrs_deleted_diff_master |
    kubectl exec -i REPLICA_CONTAINER -- python manage.py cqrs_deleted_sync_replica

Development

1. Python 3.6+
2. Install dependencies requirements/dev.txt

Testing

Unit testing

1. Python 3.6+
2. Install dependencies requirements/test.txt
3. export PYTHONPATH=/your/path/to/django-cqrs/

Check code style: flake8 Run tests: pytest

Tests reports are generated in tests/reports.

• out.xml - JUnit test results
• coverage.xml - Coverage xml results

To generate HTML coverage reports use: --cov-report html:tests/reports/cov_html

Integrational testing

1. docker-compose
2. cd integration_tests
3. docker-compose run master