Scaling Multi-tenant Applications Using the Django ORM & Postgres | PyCon Canada 2018 | Sai Srirampur

In the real-world there are 10000s of B2B companies. Their app-stack fits the multi-tenant model - each tenant(customer) deals with it’s own data. It is super critical to build scalable applications which gives the company leeway to grow as more customers get on-boarded. Let’s learn how to do that!

Transcript

1. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 Scaling Multi-Tenant Applications Using the Django ORM & Postgres Sai Srirampur PyCon Canada | Toronto | Nov 2018
2. Sai Srirampur | PyConCA 2018 • Sai Srirampur a.k.a Sai • Engineer at Citus Data • Joined Citus to make it so developers never have to worry about scaling their database • Creator django multi-tenant • Follow me @saisrirampur @citusdata
3. Sai Srirampur | PyConCA 2018
4. Sai Srirampur | PyConCA 2018 Talk about my favorite things Sai Srirampur | PyConCA 2018
5. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 PYTHON & DJANGO POSTGRES SCALING MULTI-TENANT APPLICATIONS
6. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 Why Postgres? TLDR; Open source Constraints Extensions PostGIS / Geospatial HLL, TopN, Citus Foreign data wrappers Rich SQL CTEs Window functions Full text search Datatypes JSONB
7. Sai Srirampur | PyConCA 2018 Today… Scaling Multi-Tenant Apps Using the Django ORM & Postgres 3 architectures to build Multi-tenant apps django_multitenant Scaling multi-tenant apps with distributed postgres
8. Sai Srirampur | PyConCA 2018 • Multiple customers (“tenants”) • Each with own data • SaaS • Shopify, Salesforce Multi-Tenant Apps Sai Srirampur | PyConCA 2018
9. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 Why talk about scalable multi-tenant applications? tens 100’s 1000’s
10. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 Architectures To Build Multi-Tenant Applications 3
11. Sai Srirampur | PyConCA 2018 [1] One database per tenant Sai Srirampur | PyConCA 2018
12. Sai Srirampur | PyConCA 2018 What defines a Database? • Organized collection of interrelated data • Don’t share resources. • Username and password • Connections • Memory
13. Sai Srirampur | PyConCA 2018 One database per tenant Tenant 1251 Tenant 1252 Tenant 5 [1]
14. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 CREATE DATABASE tenant_100; ./manage.py migrate --database=tenant_100; One database per tenant / Onboarding database_routing: db_for_read, db_for_write etc.
15. Sai Srirampur | PyConCA 2018 • Start quickly • Isolate customer (tenant) data • Compliance is a bit easier • Time for DBA/developer to manage • Maintain consistency (ex: create index across all databases) • Longer running migrations • Performance degrades as # customers (tenants) goes up PROS CONS [1] One Database Per Tenant Sai Srirampur | PyConCA 2018
16. Sai Srirampur | PyConCA 2018 [2] One schema per tenant Sai Srirampur | PyConCA 2018
17. Sai Srirampur | PyConCA 2018 What defines a database Schema? • Logical namespaces to hold a set of tables • Share resources: • Username and password • Connections • Memory
18. Sai Srirampur | PyConCA 2018 One schema per tenant Tenant 5 Tenant 1251 Tenant 1252 Database [2]
19. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 CREATE schema tenant_100; ./manage.py migrate --schema=tenant_100; app_code for schema routing based on tenant One schema per tenant / Onboarding
20. Sai Srirampur | PyConCA 2018 • Better resource utilization vs. one database per tenant • Start quickly • Logical isolation • Hard to manage (ex: add column across all schemas) • Longer running migrations • Performance degrades as # customers (tenants) goes up PROS CONS [2] One Schema Per Tenant Sai Srirampur | PyConCA 2018
21. Sai Srirampur | PyConCA 2018 [3] Shared table architecture Sai Srirampur | PyConCA 2018
22. Sai Srirampur | PyConCA 2018 Accounts Shared table architecture Campaigns Leads TenantId 5 5 5 1251 Database [3]
23. Sai Srirampur | PyConCA 2018 Example
24. Sai Srirampur | PyConCA 2018 • Easy maintenance • Faster running migrations • Best resource utilization • Faster performance • Scales to 1k-100k tenants • Application code to guarantee isolation • Make sure ORM calls are always scoped to a single tenant PROS CONS [3] Shared Table Architecture :-) Sai Srirampur | PyConCA 2018
25. Sai Srirampur | PyConCA 2018 Comparing the 3 architectures for scaling multi-tenant Django applications ONE DATABASE PER TENANT ONE SCHEMA PER TENANT SHARED TABLE ARCHITECTURE Database Database [1] [2] [3] start quickly & isolation guarantees, bad resource utilization & not scalable better resource util, logical isolation not scalable Scales to over 100K tenants! explicitly handle isolation
26. Sai Srirampur | PyConCA 2018
27. Sai Srirampur | PyConCA 2018 django_multitenant to the rescue Sai Srirampur | PyConCA 2018
28. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 django_multitenant Automates all ORM calls to be scoped to a single tenant
29. Sai Srirampur | PyConCA 2018 Today... Purchase.objects.filter(id=1) <=> "SELECT* from purchase where id=1"
30. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 With django_multitenant Purchase.objects.filter(id=1) <=> "SELECT* from purchase where id=1 and store_id=<current_tenant>"
31. Sai Srirampur | PyConCA 2018 Components of django_multitenant Sai Srirampur | PyConCA 2018
32. Sai Srirampur | PyConCA 2018 django_multitenant usage — 3 steps 1. Inherit all models with TenantModel 2. Change ForeignKey to TenantForeignKey 3. Define tenant scoping: set_current_tenant(t)
33. Sai Srirampur | PyConCA 2018 TenantModel defines new Manager Sai Srirampur | PyConCA 2018
34. Sai Srirampur | PyConCA 2018 TenantForeignKey mimics composite foreign key behavior • Adds tenant_id filter to referenced model. Handles: • Reference lookups — ex: (Purchase.product.name) • select_related() / prefetch_related() • Explicit Joins (product__name)
35. Sai Srirampur | PyConCA 2018 set_current_tenant(t) • Specifies which tenant the APIs should be scoped to • Set at authentication logic via middleware • Set explicitly at top of function (ex. view, external tasks/jobs)
36. Sai Srirampur | PyConCA 2018 Example Code Sai Srirampur | PyConCA 2018
37. Sai Srirampur | PyConCA 2018 Models without django_multitenant class Purchase(models.Model): store = models.ForeignKey(Store) product_purchased = models.ForeignKey(Product) ordered_at = models.DateTimeField(default=timezone.now) billing_address = models.TextField() Sai Srirampur | PyConCA 2018
38. Sai Srirampur | PyConCA 2018 Post django_multitenant class Purchase(TenantModel): store = models.ForeignKey(Store) product_purchased = TenantForeignKey(Product) ordered_at = models.DateTimeField(default=timezone.now) billing_address = models.TextField() Sai Srirampur | PyConCA 2018
39. Sai Srirampur | PyConCA 2018 Setting the tenant at authentication class SetCurrentTenantFromUser(object): def process_request(self, request): if not hasattr(self, 'authenticator'): from rest_framework_jwt.authentication import JSONWebTokenAuthentication self.authenticator = JSONWebTokenAuthentication() try: user, _ = self.authenticator.authenticate(request) except: return try: #Assuming your app has a function to get the tenant associated for a user current_tenant = get_tenant_for_user(user) except: # TODO: handle failure return set_current_tenant(current_tenant) Sai Srirampur | PyConCA 2018
40. Sai Srirampur | PyConCA 2018
41. Sai Srirampur | PyConCA 2018 Benefits of django_multitenant Drop-in implementation of shared tables architecture Guarantees isolation Ready to scale with distributed Postgres (Citus)
42. Sai Srirampur | PyConCA 2018
43. Sai Srirampur | PyConCA 2018 Infinite scale with Citus
44. Sai Srirampur | PyConCA 2018 django_multitenant with Postgres
45. Sai Srirampur | PyConCA 2018 django_multitenant with Citus
46. Sai Srirampur | PyConCA 2018 Sai Srirampur | PyConCA 2018 django_multitenant with Citus
47. Sai Srirampur | PyConCA 2018 also one of my favorites: Citus
48. Scale out Django! github.com/citusdata/django-multitenant @saisrirampur @citusdata sai@citusdata.com citusdata.com/newsletter Thank You