![•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?](https://image.slidesharecdn.com/cloudbrewcloudcamp-181017123357/85/Cloud-brew-cloudcamp-5-320.jpg)
Cloud brew cloudcamp
- 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: [email protected] 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
- 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: [email protected] T: @henry_been L: linkedin.com/in/henrybeen W: henrybeen.nl
Editor's Notes
- #2: Licence notice for those parts I copied in from the WingTips presentations MIT License Copyright (c) Microsoft Corporation. All rights reserved. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE
- #11: https://www.tubefilter.com/2018/08/20/twitch-private-messages-bug/ https://help.twitter.com/en/account-activity-api
- #16: "O’ the Mysteries of Colorado" (CC BY-NC 2.0) by WarzauWynn
- #21: The Masterplan
- #22: The Masterplan
- #24: Image from Wikipedia, https://upload.wikimedia.org/wikipedia/commons/c/ca/DART_UML_DART_2011_2013_RAW.svg
- #28: "tools" (CC BY-NC-ND 2.0) by ᴾᴴᴵᴸ
- #30: Real telemetry from an accounting customer’s set of pools (when this data was recorded they had over 300 pools, and each pool had nearly 400 databases sharing 200 DTUs). As you can see, their aggregate tenant workload is easily supported by 200 DTU pool. [The pool at 200 DTUs is twice the per-database max of 100 DTUs so would allow 2 databases to be active]
- #36: https://github.com/Microsoft/WingtipTicketsSaaS-DbPerTenant