Dgraph vs JanusGraph

Dgraph

JanusGraph

Native GraphQL Support

Yes

Only DB to natively support GraphQL resulting in capacity to process GraphQL queries in parallel with high performance

No

JanusGraph query language is Gremlin. Reference

Distributed Graph Database

Distributed ACID Transactions

• Supported and Jepsen-tested
• Synchronous replication with immediate consistency meaning any client can read the latest write
• Open source
• Reference

Sharding

• Predicate-based sharding. Avoids N+1 problem and network broadcasts when running a query in high fanout scenarios. This ensures low-latency query execution, irrespective of the size of the cluster or the number of intermediate results. Reference
• Consistent production level latencies and consistent queries. Reference
• Automatic sharding
• Sharding a single predicate on the roadmap

• When JanusGraph is deployed on a cluster of multiple storage backend instances, the graph is partitioned across those machines. By default, JanusGraph uses a random partitioning strategy that randomly assigns vertices to machines.
• When the graph is small or accommodated by a few storage instances, it is best to use random partitioning for its simplicity. As a rule of thumb, one should strongly consider enabling explicit graph partitioning and configure a suitable partitioning heuristic when the graph grows into the 10s of billions of edges.
• Reference

Consistent Replication

• Only Hbase has native support for strong consistency at row level ( Reference ). Even so JanusGraph documentation explains use of locks for data consistency on HBase here .
• Cansandra has specific configurations for replication. In general, higher levels are more consistent and robust but have higher latency. Reference

Linearizable Reads

• Apache Cassandra or Apache HBase are both eventual consistency storage backends that means JanusGraph must obtain locks in order to ensure consistency. Because of the additional steps required to acquire a lock when committing a modifying transaction, locking is a fairly expensive way to ensure consistency and can lead to deadlock when very many concurrent transactions try to modify the same elements in the graph. Reference
• JanusGraph first persists all graph mutations to the storage backend. If the primary persistence into the storage backend succeeds but secondary persistence into the indexing backends or the logging system fail, the transaction is still considered to be successful because the storage backend is the authoritative source of the graph. This can create inconsistencies with the indexes and logs. To automatically repair such inconsistencies, JanusGraph can maintain a transaction write-ahead log which is enabled through the configuration. Reference

Correctness and Durability Testing

• Jespen-tested
• Newer report will be published soon

High Availability

Yes

• HA Cluster Setup is explained here
• HA Cluster setup is available in Community Edition

• High availability depends on the backend configuration. Both Hbase and Cassandra can be highly available.
• If an instance fails, i.e. is not properly shut down, JanusGraph considers it to be active and expects its participation in cluster-wide operations which subsequently fail because this instances did not participate in or did not acknowledge the operation. In this case, the user must manually remove the failed instance record from the cluster and then retry the operation. Reference

Transparent Data Encryption

Yes

Database files are encrypted at rest with a user-specified key

Query Languages

• GraphQL
• GraphQL± (Variation of GraphQL supporting advanced features)

Management of Runaway Queries

• Context cancellation which works across clients and servers. So, a context cancellation at the client level would automatically cancel the query at all involved servers.
• OpenCensus integration, which allows distributed tracing all the way from app to Dgraph cluster and back.
• Open standards for query context cancellation and tracking

• There is nothing in Gremlin Server that will list running queries. As for cancellation, according to standard TinkerPop semantics a Traversal should respect a request for interruption on a thread. These semantics are enforced by the TinkerPop process test suite. That said, it is still up to the graph provider to properly allow for that behavior. Reference

Backups

• Binary format
• Both full and incremental backups to files, S3 and Google storage via Minio.
• Live backups with no downtime
• Reference

• JanusGraph acts as an abstraction layer on top of the storage backends and defers to the storage backends for administrative best practices. As a result, there is a lack of centralized documentation on backend administrative tasks. Reference
• Cassandra offers Snapshot, incremental, and commit-log backups. Reference
• Hbase backup offerings are summarized here

Pricing and Free Trial

• Open source version is under Apache 2.0, so free to use and modify.
• Enterprise version pricing is based on the number of instances of Dgraph, not the number of cores / RAM / Disk, etc…

Appropriate as primary database to build apps/data platform on

Open Source

Yes

• Apache 2.0. GitHub
• Enterprise features are NOT Apache 2.0. But, users can still read the source.
• Dgraph open source version and enterprise version provide the same performance. They’re only different in that enterprise version has more features.
• Dgraph supports many open standards, like Grpc, Protocol Buffers, Go contexts, Open Census integration for distributed tracing.

Protocols

• HTTP/HTTPS
• gRPC
• Protocol Buffers

• HTTP/HTTPS
• WebSockets
• Reference

Point in Time Recovery

Multi-region Deployments

Yes

SQL Migration Tool

Yes

No

There are some suggestions on the resources to do this with Cassandra here.

Authentication and authorization

• JSON web tokens
• ACLs for enterprise
• Integration with LDAP on the roadmap

Drivers

• Dgraph’s drivers use gRPC not REST
• Any GraphQL compatible client can be used
• Dgraph’s supported drivers are the same as Neo4J’s supported drivers: Java, JavaScript , Go, Python, .Net
• Dgraph’s unofficial drivers are: Rust, Dart, Elixir
• Reference

• A list of TinkerPop drivers is available on TinkerPop’s homepage
• In addition to drivers, there exist query languages for TinkerPop that make it easier to use Gremlin in different programming languages like Java, Python, or C#.

Multi-database Features

Graph Database As A Service (DBaaS)

No

Query Execution Plans

Support for graph algorithms

• Shortest k-paths
• Edge traversal limit to determine cycles in graphs
• Others requested from community listed here

Apache Spark integration

No

Kafka integration

Import/export

• Using BulkLoader or LiveLoader, Dgraph can read the data as is with no modification needed
• Supported data formats are JSON and RDF
• Exporting database is explained here

• Export: GraphML or GraphSon, Gremlin I/O library
• Import: Bulkloading, Gremlin I/O library