Neo4j - Cypher vs Gremlin query language

We have to traverse thousands of nodes in our queries. Cypher was slow. Neo4j team told us that implementing our algorithm directly against the Java API would be 100-200 times faster. We did so and got easily factor 60 out of it. As of now we have no single Cypher query in our system due to lack of confidence. Easy Cypher queries are easy to write in Java, complex queries won't perform. The problem is when you have multiple conditions in your query there is no way in Cypher to tell in which order to perform the traversals. So your cypher query may go wild into the graph in a wrong direction first. I have not done much with Gremlin, but I could imagine you get much more execution control with Gremlin.

I initially started using Gremlin. However, at the time, the REST interface was a little unstable, so I switched to Cypher. It has much better support for Neo4j. However, there are some types of queries that are simply not possible with Cypher, or where Cypher can't quite optimize the way you can with Gremlin.

Gremlin is built over Groovy, so you can actually use it as a generic way to get Neo4j to execute 'Java' code and perform various tasks from the server, without having to take the HTTP hit from the REST interface. Among others, Gremlin will let you modify data.

However, when all I want is to query data, I go with Cypher as it is more readable and easier to maintain. Gremlin is the fallback when a limitation is reached.

The Neo4j team's efforts on Cypher have been really impressive, and it's come a long way. The Neo team typically pushes people toward it, and as Cypher matures, Gremlin will probably get less attention. Cypher is a good long-term choice.

That said- Gremlin is a Groovy DSL. Using it through its Neo4j REST endpoint allows full, unfettered access to the underlying Neo4j Java API. It (and other script plugins in the same category) cannot be matched in terms of low-level power. Plus, you can run Cypher from within the Gremlin plugin.

Either way, there's a sane upgrade path where you learn both. I'd go with the one that gets you up and running faster. In my projects, I typically use Gremlin and then call Cypher (from within Gremlin or not) when I need tabular results or expressive pattern matching- both are a pain in the Gremlin DSL.

For general querying, Cypher is enough and is probably faster. The advantage of Gremlin over Cypher is when you get into high level traversing. In Gremlin, you can better define the exact traversal pattern (or your own algorithms) whereas in Cypher the engine tries to find the best traversing solution itself.

I personally use Cypher because of its simplicity and, to date, I have not had any situations where I had to use Gremlin (except working with Gremlin graphML import/export functions). I expect, however, that even if i would need to use Gremlin, I would do so for a specific query I would find on the net and never come back to again.

You can always learn Cypher really fast (in days) and then continue with the (longer-run) general Gremlin.