Cosmos DB

Fundamentals
 Jes Borland

Software Engineer @
Microsoft
[email protected]
@grrl_geek
LessThanDot.com
Cosmos DB isn’t another SQL Server flavor in “the
cloud”!
 Cosmos DB is for NoSQL data
 Not relational, not highly structured

 This one data store has a common model for multiple types of data
structures
What kind of data can be stored?

Document Document
(SQL) (MongoDB)

Table Etcd
(Key/Value) (Key/Value)

Wide column Graph

(Cassandra) (Gremlin)
Benefits of Cosmos DB
 Globally distributed
 Offers varying levels of throughput
 Offers multiple consistency models
 Has performance SLAs
 Performs automatic indexing
 Fully managed Azure fabric
APIs
SQL
 The SQL API started life as DocumentDB - a JSON database engine with a
query language that will look familiar to those used to T-SQL
SQL API Use Cases
 Anything that generates JSON data!
 IOT
 Retail
 Gaming
MongoDB
 Mongo is an open-source document database which uses JSON-like
documents
 Existing Mongo databases can have data imported directly, then the
applications that use them can be pointed to Cosmos DB
 Existing tools like Mongo Shell can be used with Cosmos DB
MongoDB API Use Cases
 Content Management Systems (CMS)
 Timekeeping
 IOT
Table
 The Table API directly interacts with Azure Table Storage, which is
key/value pairs
 Gives additional features that Storage alone can't such as global
distribution, higher throughput
Table API Use Cases
 Replace Redis (an application caching program)
 Games and scores per player
 Shopping carts
 Analytics materialized view for specific dates
Etcd
 In Kubernetes, etcd is used to store the state and the configuration of the
Kubernetes clusters.
 Ensuring availability, reliability, and performance of etcd is crucial to the
overall cluster health, scalability, elasticity availability, and performance of
a Kubernetes cluster.
Cassandra
 Apache Cassandra is a distributed, wide column store database
 “Wide column store” has tables with rows and columns, but names and
format of columns vary from row to row in the same table
 Developed at Facebook to power the inbox search feature
Cassandra API Use Cases
 Systems where writes exceed reads
 Transaction logging
 Package tracking
 IoT
Graph
 Graph is a non-relational data store
that lets you define entities (vertices),
and the relationships between them
(edges)
 Uses the Apache TinkerPop language
"Gremlin"
 It's very different from SQL!
Graph API Use Cases
 Social networks
 Vertices - people
 Edges - relationships

 Recommendations
 Vertices - restaurant, location, cuisine type
 Edges - cuisine type, distance to specific location

 Logistics
 Vertices - airports
 Edges - flights
Cosmos DB Infrastructure
Resource Model
Horizontal Partitioning
 Partitioning maps data to physical
storage
 You choose a logical partitioning
key
 Choose wisely
 The partition key will determine how data is
returned when you query it later
Partition guidance and performance
 Queries that access data in a single logical partition are more cost-effective
than queries that access multiple.
 Transactions are allowed only against a single logical partition.
 The best choice is something with high cardinality
 What is high cardinality? More uniqueness.
 Binary is low cardinality - true/false
 Unique ID is high cardinality
 First name may be somewhere between
Partitioning questions
 Can I change the partitioning key?
 No. You would have to create a new collection with the new partition key, and move the data.
Global Distribution
 Global distribution is how the data is
replicated to the regions you choose
 You can choose automatic or manual
failover between regions
 This can be set in the Azure Portal, or
in the application's ConnectionPolicy
parameter
Multi-master writes
 Write-anywhere model
 < 10 ms latency and 99.999 % write availability
Consistency Levels
 SQL Server has isolation levels  Cosmos DB has consistency levels
 Read Uncommitted  Eventual
 Read Committed  Consistent prefix
 Repeatable Read  Session
 Snapshot  Bounded stateless
 Serializable  Strong
Strong
 A write is only visible after it's committed to a majority quorum of replicas
 Reads are guaranteed to return the most recent version
 Can only be used in one region
 Examples
 Banking accounts
 Payment processing
 Reservation systems

“Absolute truth every time”

Bounded stateless
 Reads can lag behind writes in a "staleness windows"
 K versions
 t time interval

 Global total order outside of that window

 Recommended for globally-distributed applications
 Examples
 Flight status app
 Package tracking app

“Not too old”

Session
 Consistency is scoped to a single session
 Guarantees RYW - read your own writes
 Examples
 Social profile updates
 Retail shopping cart
 Social music app preferences

“Logical, real-time experiences for a user”

Constant Prefix
 If there are no more writes, all replicas will have the same data
 ….and in the correct order
 Examples
 Baseball score updates
 Social media comments

“The order of the data tells as much of a story as the end result”
 Client Jes
1 sees: Erin

Jes
App Erin Jes
Client
writes: Sarah
2 sees:

Jes
Client Erin
3 sees: Sarah
Eventual
 Eventually, with no further writes, all replicas in a group eventually
converge
 A session may get values older than what it saw on a previous read
 Examples
 Product reviews
 Social media following posts
 Receipts

“Apps that live and die according to their availability”

Throughput
 Request Units
 CPU + memory + IOPs
 Similar to a SQL Database DTU

 A 1-RU throughput corresponds to the throughput of the GET of a

1-KB document
 Each operation (create, update, delete, read, query) requires X RUs for each
consistency level
 CRUD are consistent
 SELECT returning X rows - these depend on X
 The query explorer in the Portal will show you how many RUs a query used
Viewing Metrics
Working with Cosmos DB
I'm the DBA...hold me, I'm scared
 High Availability
 All writes are durably committed to local disks by a quorum of replicas within a local data
center before acknowledging to the client
 Each partition is made highly available via a replica set

 Disaster Recovery
 Choose multiple regions for DR
Backup and restore
 Backups
 Full backups taken approximately every 4 hours
 The last 2 backups are stored at all times in Azure Blob storage
 Stored in same region as your current write region
 Replicated to another region via GRS
 "If the data is accidentally dropped or corrupted, contact Azure support within eight hours."

 Restores
 File a support ticket or call Azure support
Security
 Policy-driven IP-based access controls for inbound firewall support
 Using the master key for the account, you can create user resources and
permission resources per database
 Data is encrypted at rest
 All client-to-service Azure Cosmos DB interactions are SSL/TLS 1.2
capable. All intra datacenter and cross datacenter replication is SSL/TLS
1.2 enforced.
I'm the developer...hold me back!
 Quickstarts are available for all APIs in multiple languages

SQL MongoDB Graph Table

• .NET • Node.js • .NET • PHP
• Java • .NET • Gremlin console • .NET
• Node.js • Java • Java • Java
• Python • Golang • Node.js • Node.js
• Xamarin • Python • Python • Python
Let's play
 Azure Cosmos DB Emulator
https://docs.microsoft.com/en-us/azure/cosmos-db/local-emulator
 Quickstart: SQL API + .NET
https://docs.microsoft.com/en-us/azure/cosmos-db/create-sql-api-dotnet
Demo
 Open Azure Cosmos DB Emulator
 Open Visual Studio
 Emulator – click Explorer, click New Collection
 Database id – Create new – Tasks
 Collection id – Items
 Storage – Fixed
 Throughput – 400
 OK
Demo 2
 Click Items > Documents > New  {

Document  "id": "2",

 {  "category": "workout",
 "id": "1",  "name": "spin",
 "category": "personal",  "description": "12:10 spin class",
 "name": "groceries",  "isComplete": false
 "description": "Pick up apples and  }
strawberries.",  One more, category “personal”
 "isComplete": false
 }

 Save
 New
Demo 3
 Click Edit Filter button.
 Add ORDER BY c._ts DESC

 Click New SQL Query

 SELECT * FROM c WHERE c.category = "workout"
Demo 4
 Visual Studio > Open Project/Solution > C:\git-samples\documentdb-
dotnet-todo-app > todo.sln
 Open web.config
 <add key="endpoint“ – add URI
 <add key="authKey“ – add Primary Key
 <add key="database" – add Database name Tasks
 <add key="collection“ – add Collection name Items

 Run app – F5 or green arrow with “Microsoft Edge”

 Click Create New
 Baking
 Make zucchini bread
 Personal
 Create
Demo 5
 Emulator > Explorer > New SQL Query
 SELECT c.id, c.name, c.description, c.isComplete FROM c WHERE c.category = "personal"

 Close Edge
 Stop program
 Close VS
 Close Emulator
Why Cosmos DB?

Scalable up/down
Distributed
for surges and
worldwide
dips in usage

Flexible with Meets or exceeds

consistency performance
models SLAs
Resources
• Azure Cosmos DB product page -
https://azure.microsoft.com/en-us/services/cosmos-db/
• Documentation - https://docs.microsoft.com/en-us/azure/cosmos-db/
• eBook - https://discover.microsoft.com/apps-for-cats/ (Highly
recommended, less than half an hour to go through.)
• SLAs
https://azure.microsoft.com/en-us/support/legal/sla/cosmos-db/v1_2/
Questions?
 Jes Borland
Premier Field Engineer
Microsoft
[email protected]
@grrl_geek
LessThanDot.com