Multitenant applications: How and Why
0 likes516 views
The document discusses multitenant applications, emphasizing the architecture where a single software instance serves multiple tenants to reduce infrastructure costs and simplify maintenance. It explains implementation strategies, including middleware for tenant identification and querying methods, and examines the pros and cons of database routing versus application multitenancy. Additionally, it highlights challenges like ensuring data isolation and managing schema limits while optimizing performance and development efficiency.
![Single Shared Schema
[or how the big guys do it]](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-16-320.jpg)
![Routing - ibm.spinnertracking.com
def tenant_middleware(get_response):
def middleware(request):
host = request.get_host().split(':')[0]
subdomain = host.split('.')[0]
try:
customer = Customer.objects.get(name=subdomain)
except Customer.DoesNotExist:
customer = None
request.customer = customer
response = get_response(request)
return response
return middleware](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-19-320.jpg)
![Querying
avg_duration = (
Spin.objects
.filter(user_spinner__user__customer=request.customer)
.aggregate(avg=Avg('duration')))['avg']](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-20-320.jpg)
![Simpler querying
avg_duration = (
Spin.objects
.filter(customer=request.customer)
.aggregate(avg=Avg('duration')))['avg']](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-22-320.jpg)
![The router
class TenantRouter(object):
def db_for_read(self, model, **hints):
return get_thread_local('using_db', 'default')
def db_for_write(self, model, **hints):
return get_thread_local('using_db', 'default')
# …
# settings.py
DATABASE_ROUTERS = ['multitenancy.routers.TenantRouter']](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-30-320.jpg)
![Routing - settings
MIDDLEWARE_CLASSES = [
'tenant_schemas.middleware.TenantMiddleware',
# …
]
DATABASES = {
'default': {
'ENGINE': 'tenant_schemas.postgresql_backend',
'NAME': 'mydb',
}
}](https://image.slidesharecdn.com/multitenantapplications-howandwhy-170729150805/85/Multitenant-applications-How-and-Why-40-320.jpg)
Multitenant applications: How and Why
- 1. Multitenant applications: how and why @xima
- 3. Who am I? ● Filipe Ximenes ● Recife / Brazil ● Aussie for 1 year (2008 - 2009)
- 6. FLOSS Django React boilerplate https://github.com/vintasoftware/django-react-boilerplate Django Role Permissions https://github.com/vintasoftware/django-role-permissions Tapioca https://github.com/vintasoftware/tapioca-wrapper
- 7. Context
- 12. "How do you protect our data?"
- 14. "... refers to a software architecture in which a single instance of software runs on a server and serves multiple tenants." - Wikipedia
- 15. What we want to achieve? ● Reduce infrastructure costs by sharing hardware resources ● Simplify software maintenance by keeping a single code base ● Simplify infrastructure maintenance by having fewer nodes
- 16. Single Shared Schema [or how the big guys do it]
- 19. Routing - ibm.spinnertracking.com def tenant_middleware(get_response): def middleware(request): host = request.get_host().split(':')[0] subdomain = host.split('.')[0] try: customer = Customer.objects.get(name=subdomain) except Customer.DoesNotExist: customer = None request.customer = customer response = get_response(request) return response return middleware
- 22. Simpler querying avg_duration = ( Spin.objects .filter(customer=request.customer) .aggregate(avg=Avg('duration')))['avg']
- 23. Case study: Salesforce ● 1:5000 ratio ● Double checking ● Transparent to developers
- 24. Drawbacks ● Guaranteeing isolation is hard ● Might lead to complexity to the codebase ● 3rd party library integration
- 27. Routing DATABASES = { 'default': { 'ENGINE': ..., 'NAME': ..., }, 'ibm': { 'ENGINE': ..., 'NAME': ..., } }
- 28. The `.using()` approach spinners = ( Spinner.objects .using(request.customer.name) .annotate( avg_duration=Avg('owned_spinners__spins__duration')) .order_by('-avg_duration'))
- 29. The threadlocal middleware approach def multidb_middleware(get_response): def middleware(request): subdomain = get_subdomain(request) customer = get_customer(subdomain) request.customer = customer @thread_local(using_db=customer.name) def execute_request(request): return get_response(request) response = execute_request(request) return response return middleware
- 30. The router class TenantRouter(object): def db_for_read(self, model, **hints): return get_thread_local('using_db', 'default') def db_for_write(self, model, **hints): return get_thread_local('using_db', 'default') # … # settings.py DATABASE_ROUTERS = ['multitenancy.routers.TenantRouter']
- 35. What are schemas in the first place? SELECT id, name FROM user WHERE user.name LIKE 'F%';
- 36. What are schemas in the first place? CREATE SCHEMA ibm; SELECT id, name FROM ibm.user WHERE ibm.user.name LIKE 'F%';
- 37. The `search_path` SET search_path TO ibm; SELECT id, name FROM user WHERE user.name LIKE 'F%';
- 39. Routing - middleware # ... connection.set_schema_to_public() hostname = self.hostname_from_request(request) TenantModel = get_tenant_model() try: tenant = self.get_tenant(TenantModel, hostname, request) assert isinstance(tenant, TenantModel) except TenantModel.DoesNotExist: # ... request.tenant = tenant connection.set_tenant(request.tenant) # ...
- 40. Routing - settings MIDDLEWARE_CLASSES = [ 'tenant_schemas.middleware.TenantMiddleware', # … ] DATABASES = { 'default': { 'ENGINE': 'tenant_schemas.postgresql_backend', 'NAME': 'mydb', } }
- 41. Routing - db backend # ... try: cursor_for_search_path.execute( 'SET search_path = {0}'.format(','.join(search_paths))) except (django.db.utils.DatabaseError, psycopg2.InternalError): self.search_path_set = False else: self.search_path_set = True if name: cursor_for_search_path.close() # ...
- 42. The Command Line ./manage.py tenant_command shell ./manage.py createsuperuser ./manage.py migrate_schemas
- 44. SELECT id, duration FROM ibm.spinner_spin WHERE duration > 120 UNION SELECT id, duration FROM vinta.spinner_spin WHERE duration > 120; Querying across schemas
- 45. SELECT uuid, duration FROM ibm.spinner_spin WHERE duration > 120 UNION SELECT uuid, duration FROM vinta.spinner_spin WHERE duration > 120; Querying across schemas
- 46. Upsides ● Querying looks same as standard application ● New schemas created automatically ● Knows how to handle migrations ● Simpler infrastructure
- 47. Drawbacks ● Be carefull with too many schemas (maybe not more than 100's clients?) ● Tests need some setup and might get slower ● Harder to query across schemas
- 48. multitenancy is not discrete, it is a continuous spectrum