Dgraph: Fast graph database written in Go

The talk was given at Gophercon Singapore.

Link to video: https://www.youtube.com/watch?v=cHXbYLNa0qQ

Transcript

1. Dgraph: Fast graph database written
   in Go
   Manish R Jain
   Dgraph Labs
   

   View Slide

2. 1. Introduction: Hi, Diggy!
   

   View Slide

3. What are graph databases
   Optimized for key-value lookups AND edge traversals ( joins ).
   

   View Slide

4. What is Dgraph?
   Dgraph is an open source graph database built for web-scale production environments
   written entirely in Go.
   Fast
   Sharded and Distributed (Distributed Joins, Filters and Sorts)
   Horizontally scalable
   Consistent Replication via Raft
   Highly Available by design
   Fault tolerant
   

   View Slide

5. Why build it?
   Any company doing anything smart is using graphs.
   No native scalable solution.
   

   View Slide

6. Concepts
   SUBJECT - predicate -> OBJECT
   // Facts
   .
   .
   // Half facts
   .
   SUBJECT - predicate -> Value
   "Singapore" .
   

   View Slide

7. Reception
   v0.1 release in Dec '15
   3200 Github stars
   32 contributors
   Next release: v0.8
   First page of HN multiple times.
   news.ycombinator.com/item?id=11322444 (https://news.ycombinator.com/item?id=11322444)
   

   View Slide

8. 2. Benchmarks
   

   View Slide

9. Neo4j
   Neo4j data loading
   Dgraph is 100x faster.
   Neo4j query
   Dgraph is 3x-6x faster on a read-write workload. At least as fast on read-only.
   Link to full benchmark code (https://github.com/dgraph-io/benchmarks/tree/master/data/neo4j)
   

   View Slide

10. Cayley (with Bolt on Macbook)
    Cayley is a graph layer written in Go.
    Loading 21M RDFs
    Dgraph is 9.7x faster
    Queries
    Dgraph is 5x - 36.6x faster
    Link to full benchmark code (https://github.com/ankurayadav/graphdb-benchmarks)
    

    View Slide

11. 3. Using a graph database
    

    View Slide

12. Graph-y questions (1/2)
    Are they an overkill?
    Are they robust?
    Are they scalable?
    

    View Slide

13. Graph-y questions (2/2)
    Are they an overkill? No
    Are they robust? Yes
    Are they scalable? Yes
    

    View Slide

14. Your SQL DB is slowing you down (1/3)
    SQL is like C, Graph query language is like Go or Python.
    Just like high level languages allow you to express logic faster,
    Databases allowing more complex queries allow you to iterate on your application faster.
    They achieve this by cutting down your application code by at least half.
    

    View Slide

15. Your SQL DB is slowing you down (2/3)
    What SQL forces you to do:
    Because SQL does lesser things, you need to do more things.
    Maintain a strict schema.
    Data duplication to avoid slow joins.
    Pre-compute counts for foreign tables to allow e cient sorting.
    

    View Slide

16. Your SQL DB is slowing you down (3/3)
    What Dgraph does:
    Flexible sparse schema, with real time modi able data types.
    Join is a single lookup away, so no data duplication.
    Simplify schema.
    Any counts can be done cheaply in real time.
    

    View Slide

17. 4. Demo: Run Stack Over ow on Dgraph
    

    View Slide

18. User Schema
    SQL | Dgraph
    --- | ------
    Reputation | Reputation
    CreationDate | CreationDate
    DisplayName | DisplayName
    LastAccessDate | LastAccessDate
    Location | Location
    AboutMe | AboutMe
    Age | Age
    Views |
    UpVotes |
    DownVotes |
    

    View Slide

19. Versioning (SQL, Dgraph)
    SQL | Dgraph
    --- | ------
    TypeId | Type
    PostId | Post (point to Post)
    CreationDate | CreationDate
    Author | Author
    Text | Text
    

    View Slide

20. Post
    SQL | Dgraph
    --- | ------
    TypeId | Type
    ParentId | Has.Answer
    AcceptedAnswerId | Chosen.Answer
    Title (duplicate of versioning) | Title (point to Version)
    Body (duplicate of versioning) | Body (point to Version)
    Tags (duplicate of versioning) | Tags (point to Version)
    OwnerUserId | Owner (point to User)
    Score | Score
    LastEditorUserId |
    LastEditDate |
    LastActivityDate |
    CommentCount |
    AnswerCount |
    FavoriteCount |
    CreationDate |
    

    View Slide

21. Comment, Vote
    SQL | Dgraph
    --- | ------
    PostId (Vote/Comment -> Post) | (New edge from Post -> Vote/Comment)
    Timestamp | Timestamp
    Author | Author
    VoteType (for Vote) | Score (for both Vote and Comment)
    Text (for Comment) | Text (for Comment)
    

    View Slide

22. Dgraph Schema
    

    View Slide

23. Live Demo
    

    View Slide

24. Query 1: Top questions
    

    View Slide

25. View Slide

26. Query 2: Question Page
    

    View Slide

27. 5. Final Thoughts
    

    View Slide

28. When to use Dgraph, when SQL
    SQL has a great use case, which is to provide ACID transactions.
    Great solution to store nancial data.
    For everything else, I think a well designed graph database is more ideally suited.
    Cuts down on the amount of coding e ort required, using the power of graphs to run
    complex queries.
    

    View Slide

29. Thank you
    Manish R Jain
    Dgraph Labs
    manish@dgraph.io (mailto:manish@dgraph.io)
    https://dgraph.io (https://dgraph.io)
    

    View Slide

30. View Slide