An introduction into creating a multi tenant SaaS application, creating a database per tenant architecture. Incluiding a case study, example and general pointers

1. Henry Been
How to build a multi-tenant SaaS application using 60.000
databases

3. Server 1 Server 2 Server 3 Server 4 Server n
Database
loadbalancing

4. WHY?

5. •Single point of failure
•More expensive
•Cannot grow in small steps
•Scaling up is failing up [1]
[1] Scalability rules: 50 principles for Scaling Web Sites
WHY?

6. MANY, MANY
SMALLER
DATABASES
to the resue!

7. Server 1 Server 2 Server 3 Server 4 Server n
DB
loadbalancing
DB DB DB DB DB

8. •Unlimited scaleability
•Isolation of failure
•Roughly 2,5 times cheaper
Database per tenant provides…

9. 100% tenant isolation
•Snapshot backup/restore
•Recover to point in time
•Compliance
•Limited authorization risks
100% tenant isoluation

11. •Designed for single tenant, single db
•Move to the cloud, one by one
•Use case I will present
Cloud strategy for legacy applications

12. WONDERING WHO
IS THAT GUY?
HENRY BEEN
Independent Devops & Azure Architect
E: consultancy@henrybeen.nl
T: @henry_been
L: linkedin.com/in/henrybeen
W: henrybeen.nl

13. A (very) short history of multi-tenancy
Tenant C
Standalone App
Tenant B
Tenant
DB
App
Tenant A

14. A (very) short history of multi-tenancy
App
Sharded Multi-tenant Database
Catalog
Tenant
C
Tenant
B
App
Tenant
M
Catalog
Tenants
A,B,C,D
Tenant
A
Tenants
E,F,J,K
Share everything Database per Tenant
Tenant
DB
App
Tenant
L

15. "Incidental Expenses" (CC BY-NC 2.0) by tim ellis
"O’ the Mysteries of Colorado" (CC BY-NC 2.0) by WarzauWynn

18. SnelStart
A
SnelStart
AA

19. SnelStart
SnelStart
A

20. Web-
application A
SnelStart
SnelStart

21. Web-
application A

24. User
Directory

25. Catalog
Cust 1 Cust 2 Cust 3 Cust N
2. App uses key to get
connection from catalog
Tenant Databases
4. On subsequent requests, use a
cache!
Tenant Catalog
1. User connects to the app
Selfservice tenant creation and destruction
A. User signs up for a new
tenant
Cust 4
B. App provisions new
tenant db
C. …registers tenant key,
db location in catalog
3. …then connects to correct
tenant database
ARM / SQL Database
Tenant Onboarding Application Connection
Base
TenantDB
Tenant
bacpac
Apps

26. Catalog
Cust 1 Cust 2 Cust 3 Cust N
Tenant Databases
Tenant Catalog
More simple approach: no selfservice
Apps
Management scripts
AAD

27. So…. what do I need?
"tools" (CC BY-NC-ND 2.0) by ᴾᴴᴵᴸ

28. Individually unpredictable
share
resources
Collectively stable and predictable
Cost effective,
scale pool up/down based on trends
Expensive,
must provision each db for peaks
Pools are cost-effective for unpredictable workloads

29. Elastic pool usage example
385 dbs in 200 DTU pool, per-database max at 100 DTUs

30. Elastic
Database
Client Library
Elastic
Database
Jobs
Elastic
Transactions
Elastic
Query
More elastic capabilities for Azure SQL DB

31. Elastic Database Client Library
using (SqlConnection conn = customerShardMap.OpenConnectionForKey(
customerId,
Configuration.GetCredentialsConnectionString(),
ConnectionOptions.Validate))
{
// Execute a simple command.
SqlCommand cmd = conn.CreateCommand();
cmd.CommandText = @"UPDATE Sales.Customer ….";
// … and more boring stuff ..
}

32. Elastic Transactions
using (var scope = new TransactionScope())
{
using (var conn1 = new SqlConnection(connStrDb1))
{ // … boring stuff … }
using (var conn2 = new SqlConnection(connStrDb2))
{ // … boring stuff }
scope.Complete();
}

33. Cust 1 Cust 2 Cust 3 Cust N
Schema management at scale
Cust 4
Catalog
Manage 1000s of databases as one
Apps
T-SQLT-SQLT-SQL T-SQLT-SQL
Azure Portal
Job
Account
T-SQL
CREATE TABLE…
CREATE INDEX…
INSERT INTO…
SELECT * FROM…
T-SQL Job
SELECT * FROM…
Elastic Jobs
Tenant Databases
Tenant Catalog
Jobs, target groups,
schedules, credentials

34. Cust 1 Cust 2
Tenant Databases
Cust 3 Cust N
Distributed query across tenant databases
Tenant Catalog
Adhoc
Analytics
Cust 4
Catalog
Database locations are
retrieved from catalog
Distributed query plan
External tables
project data from
tenant dbs
PowerBI
Query from any client
Query all tenants as if they are in a single database
Apps
Elastic Query

35. demo
Wingtip Tickets SaaS app

37. Standalone App Database per tenant Sharded Multi-tenant
Scale High
1-1000s
Very high
1-100,000s
Unlimited
1-1,000,000s
Database cost–per tenant High (sized for peaks) Low (using pools) Lowest (small tenants)
Tenant isolation Very High High Low (high for singletons)
Performance monitoring/mgt. Per-tenant Aggregate + per-tenant Aggregate, per-tenant for
singletons only
Restore single tenant Easy Easy Hard (easy for singletons)
Disaster recovery (all tenants) Moderate
Many components to
recover
Moderate
Patterns address complexity at
scale
Easy (for multi-tenant dbs)
Patterns address singleton
complexity at scale
Development complexity Low Low Medium (sharding)
Operational complexity Medium-High
Individually simple, complex
at scale
Low-Medium
Patterns address complexity at
scale
Low-High
Individual tenant management
is complex
Per-tenant customization Easy Easy Complex
Comparing the models, what we see…

39. DO TRY THIS AT HOME!
HENRY BEEN
Independent Devops & Azure Architect
E: consultancy@henrybeen.nl
T: @henry_been
L: linkedin.com/in/henrybeen
W: henrybeen.nl