Should I use Janus server or embed it as a library?

[melgenek]

Hi, Janus community!

I am going to deploy Janus with Scylla backend.
The graph is going to be used by:

1. an http app that does simple queries by node ids
2. long-running traversals in spark

It seems that there are two options to run Janus:

1. as a cluster of docker containers
2. as java library

Approach 1 works better for me because it allows hosting Janus as a regular database where the clients simply send requests via a Gremlin api.

There are some restrictions imposed by the remote gremlin connection as described here https://tinkerpop.apache.org/docs/current/reference/#connecting-gremlin-server-limitations. Though, it's not really clear in which use cases Janus server or embedded mode should be used.

Is it a viable option to host a set docker containers for my use case or should I consider using Janus as a library?

Thank you in advance for your help!

[li-boxuan]

FYI, there is a related discussion: #1108

[melgenek]

Thank you a lot. It's a great resource that describes pros and cons of both approaches.

As a summary for myself and others who come to this issue later, I'll post the @FlorianHockmann's recommendation from this post https://groups.google.com/g/janusgraph-users/c/t7gNBeWC844/m/gW7QzowECAAJ:

The overall recommendation is to use JanusGraph Server, unless you have really good reasons to use JanusGraph embedded.