Graph Database in Azure Cosmos

March 21, 2023    Azure Cosmos GraphDatabase Presentation

Graph Database in Azure Cosmos

I’m fortunate to be a part of a talented team that will be building a new product. We’ve compared using a graph database with a relational database. We’ve decided to use a graph database. We’ve been learning about graph databases and using GremlinDB inside of Azure Cosmos. It was important to prototype what the graph might look (we used like before starting.

Here’s what the graph looks like in the Azure Portal (using Learning Azure Cosmos DB | Pluralsight as a guide)

graph DB screenshot from Azure Portal

What is a Graph Database?

A graph database is a type of database that uses graph theory to store, manage, and query data. In a graph database, data is represented as nodes (also known as vertices) and edges. Nodes represent entities, while edges represent the relationships between them.

Each node and edge in a graph database can store properties, which are key-value pairs that provide additional information about the node or edge. These properties can be used to filter and query the data in the graph.

Graph databases are particularly useful for managing data that has complex relationships, such as social networks, recommendation engines, and fraud detection systems. They are also well-suited for applications that require real-time analysis of large datasets.

(thanks ChatGPT for the answer)

when should I use a graphDb? answer by

“the main thing I love about it is, the ability to be extremely flexible in the modeling of relationships” ~ Kyle after experience

Prototyping Questions

  • What will the relationships be? Does a graph make sense?
  • What are the edges and vertices?
  • What data do we need to store?
  • Would a relational database be better?


I started with this excelent course Learning Azure Cosmos DB | Pluralsight .

  • Graph Container
  • Vertex (entity) and edge (1 way relationship)
  • CRUD - GraphSON (convention to define vertices and edges and persist) and Gremlin (functional step by step language create and query), based on Apache TinkerPop
  • NoSQL storage - the data is stored in JSON


Complex, many to many, excessive joins, analyze interconnected data/relationships social, recommendations, knowledge graphs, conference


  • Id (within partition key) - like others
  • label (type) Edge
  • cardinality
    • 2 vertices being connected.
  • ways to represent the relationships
    • examples: know, interested, runsOs, users, located
    • use domain language to match with business terminology

Gremlin Querying

g.addV('person').property('id', 'John')

.property('weekends', true)
.to(g.V().has('id', 'Acme'))

a query from

  by('firstName', decr)

Gremlin Steps for Querying

  • Map
  • Filter
  • Side-effect
    • walking on the graph, collecting information
    • .store()… .cap()
    • makes a union/unique
  • Barrier
  • Branch

filter out as soon as you can will be faster and cheaper.action-area

“you have in there the 1 way relationship but that is kinda true, for instance an vertext in partition A will store all of its outward edges in partition A but incoming ones could be stored in a different partition. This can cause issues when you are doing a lot of different relationship queries. It’s a pretty complex but subtle thing that will kill you at scale.” ~Kyle from recent experience coding an api

Pay attention to your RU usage (in Azure Cosmos) / performance

to demo the subtle differences in the partitioned scenario, this query would be excellent in the partitioned graph.

string query = g.V("thomas").OutE("knows").Where(__.InV().Has("id", "mary")).Drop().ToGremlinQuery();

but this would could absolutely kill your RU’s

string query = g.V("thomas").InE("knows").Where(__.InV().Has("name", "mary")).Drop().ToGremlinQuery();

Demo code from Pluralsight

When I’m presenting, I’ll share a few snippets from Learning Azure Cosmos DB | Pluralsight .

Azure Cosmos DB

"Azure Cosmos DB for Apache Gremlin is a graph database service that can be used to store massive graphs with billions of vertices and edges. You can query the graphs with millisecond latency and evolve the graph structure easily. The API for Gremlin is built based on Apache TinkerPop, a graph computing framework that uses the Gremlin query language."

from Azure Cosmos has a Gremlin implementation . The data is provisioned, etc. just the same as Table or Document or others in Cosmos DB. There is a long list of benefits. We see Azure Cosmos as a great fit for our app. There is a cost, but we think it will fit our product well.

You need to choose the Partition Key when you create the graph. This is using a partitioned graph, which just means that vertexes are stored in different partitions so you have to be careful about your modeling.

serverless or provisioned? Serverlesss is more restricted “Unavailable (serverless accounts can only run in a single Azure region)”, billed per hour for number or Rus

Some differences when using Azure Cosmos DB

Azure Cosmos DB Graph engine runs breadth-first traversal while TinkerPop Gremlin is depth-first. This behavior achieves better performance in horizontally scalable system like Azure Cosmos DB.

Watch out for compatibility issues with the Gremlin .Net Nuget package by Apache TinkerPop - Azure Cosmos DB for Gremlin support and compatibility with TinkerPop features | Microsoft Learn 3.4.13 was released on Jan 13, 2022 :-(. Also the constructor new GremlinClient(gremlinServer, mimeType: GremlinClient.GraphSON2MimeType) to avoid error: "Gremlin Malformed Request: GraphSON v3 IO is not supported."

LINQ Possibilities

You may want to avoid using strings in the queries. If you’re up for an abstraction and another dependency, check out GitHub - ExRam/ExRam.Gremlinq: A .NET object-graph-mapper for Apache TinkerPop™ Gremlin enabled databases. was the one that was the most up to date that I found on Github.

Here’s an example from their page:

var g = AnonymousTraversalSource.Traversal();
string query = g.V("thomas").OutE("knows").Where(__.InV().Has("id", "mary")).Drop().ToGremlinQuery();

// query will be "g.V('thomas').outE('knows').where(inV().has('id', 'mary')).drop()"

// with strong typing
string query = g.V<Person>("thomas").OutE<Knows>().Where(__ => __.InV<Person>("mary")).Drop().ToGremlinQuery();

After using Gremlinq in our project, Kyle says “Gremlinq is awesome” :-).

GitHub - csharpsi/Gremlin.Net.Extensions looks to be another option.

I recommend doing some prototyping/experimenting to see what works best for you.


In Azure you have to use the primary key/password, in the [].Net Conf 2022 video]( ) at minute 16 she says “that doesn’t work for the database level”. That may change in the future, check the docs

Connect to GraphQL

More Videos

Cheatsheets from .Net Conf 2022

note: Kyle is a great innovative team member, software engineer, architect and leader. He said I could quote him in this article.

A Power Point Presentation for lunch and learn

My Omnitech Lunch and Learn Power Point

Watch the Story for Good News
I gladly accept BTC Lightning Network tips at [email protected]

Please consider using Brave and adding me to your BAT payment ledger. Then you won't have to see ads! (when I get to $100 in Google Ads for a payout, I pledge to turn off ads)

Use Brave

Also check out my Resources Page for referrals that would help me.

Swan logo
Use Swan Bitcoin to onramp with low fees and automatic daily cost averaging and get $10 in BTC when you sign up.