Yeah . `apache/gremlin" seems like a better option then. Does anyone have anything against it? I think we are pretty happy with accepting "other apache" projects as providers, so I see no issue with Gremlin - knowing that we can always reach out to our friendly Apache Community in case of any issues. So - unless we do not hear any "opposition" in a few days, I think it would make sense if you start `[LAZY CONSENSUS]` thread - without a need for `[VOTE]` thread.
One thing though that I would love to have - is to also have an integration test if possible (we had it with apache.kafka for example) - those are tests that could run **some** graphdb database locally (via docker-compose) and run a very rudimentary checks against a "real" database, not a mocked call. That would make it more robust. More about integration tests, how to build, run, test them and integrate them in our CI can be found here: https://github.com/apache/airflow/blob/main/contributing-docs/testing/integration_tests.rst - happy to help if you are stuck with it. J. On Wed, Feb 26, 2025 at 1:25 PM Ahmad Farhan <ahmad.farhan9...@gmail.com> wrote: > I pushed changes to move the provider into the “apache” directory. After > updating the class references across the project, I re-tested and all tests > passed. > > Regarding the use of Gremlin (or another graph query language like Cypher > and SPARQL) for a common package approach, here are my thoughts on the pros > and cons: > > pros (I can see only one): > > - Gremlin has been widely adopted by different cloud vendors (e.g. Azure > Cosmos DB with Apache Gremlin and AWS Neptune) as well as in self-hosted > environments. > > cons: > > - Gremlin, Cypher (native for Neo4j) and SPARQL each have their own > drivers for executing queries. > - To achieve a common abstraction, a wrapper around each driver would be > required. Each driver has its own connection parameters, underlying > protocols, and may need method overrides for compatibility with > different > Python versions. > - Not all vendors support every query language; for instance, Gremlin > for Neo4j has been deprecated in recent releases, while Cosmos DB does > not > support Cypher or SPARQL. > > While it would be ideal to have a unified graph query language and driver > that works seamlessly across different vendors, such a solution does not > exist at the moment. In my opinion, implementing provider-specific > solutions for each query language (Gremlin, Cypher, SPARQL) is more > realistic and practical given the current landscape. > > Happy to discuss further or answer any questions! > > Farhan > > On Mon, Feb 24, 2025 at 11:33 AM Ahmad Farhan <ahmad.farhan9...@gmail.com> > wrote: > > > I have worked with two different graph database vendors—Azure Cosmos DB > > and Neo4j. During our migration to Neo4j, we discovered that using the > > Gremlin language wasn’t possible; we were forced to rewrite all our > queries > > into Cypher, which is the native language for Neo4j and, in my > experience, > > much simpler for querying. > > > > This situation highlights a key challenge for a common abstraction: the > > underlying query languages and connection/authentication mechanisms vary > > significantly. Gremlin is not only different from Cypher in syntax but is > > also deprecated for Neo4j (see > > https://tinkerpop.apache.org/docs/3.7.3/reference/#neo4j-gremlin). > > > > The question would be how can the common approach accommodate these > > different query languages? > > > > On Fri, Feb 21, 2025 at 7:36 PM Jarek Potiuk <ja...@potiuk.com> wrote: > > > >> Without deep looking at the code I love the idea - it's very similar to > >> what we have for common.sql and common.io - and soon common.messaging > - I > >> also - long time ago - suggested common.dataframe that someone could > >> submit > >> using Apache Ibis: > >> https://lists.apache.org/thread/qx3yh6h0l6jb0kh3fz9q95b3x5b4001l - > >> similarly I believe there was an idea about common.llm ... > >> > >> I think the "common" pattern is a great one for Airflow, to build on top > >> of > >> "other giants" who build those common abstractions that you can easily > >> switch between different implementations of various data access layers. > >> > >> My suggestion and question - would be however (not very strong on it, I > >> would love to hear what others think, I know it's been somewhat > >> contentious > >> when I started the ibis discussion) - would be to make it > "common.graph", > >> "common.dataframe" - instead of "apache.gremlin" or "apache.ibis" - just > >> to > >> stress that those are not implementations of particular service but > >> opinionated choice of particular technology to do "common" operations. > >> This > >> is what essentially "common.io" is . - it should be named "fsspec" > >> provider > >> if we were to name it by the "library" that implemented it. > >> > >> J. > >> > >> > >> On Fri, Feb 21, 2025 at 8:22 PM Ahmad Farhan < > ahmad.farhan9...@gmail.com> > >> wrote: > >> > >> > Hi Everyone, > >> > > >> > I’ve created a draft PR (https://github.com/apache/airflow/pull/46977 > ) > >> to > >> > introduce and discuss a new provider for using Gremlin—the graph > >> traversal > >> > language of Apache TinkerPop (more details here: > >> > https://tinkerpop.apache.org/gremlin.html). Gremlin is supported by > >> > various > >> > graph database vendors such as Azure Cosmos DB and Amazon Neptune. > >> > Previously, I had to develop a custom hook to query data from Azure > >> Cosmos > >> > DB using Apache Gremlin. > >> > > >> > I managed to create a provider and run it locally on the main branch. > >> > However, I ran into the BaseHook issue ( > >> > https://github.com/apache/airflow/issues/45233) on that branch, so I > >> ended > >> > up testing it fully on the v2-10-test branch. The PR should be > complete, > >> > but I’ve kept it as a draft for now while we discuss the provider. > >> > > >> > I’m a new contributor, so I’m especially eager to hear your feedback. > >> > Comments on the PR is very welcome, and please feel free to reach out > >> with > >> > any questions via email or Slack. > >> > > >> > Thanks, > >> > Ahmad Farhan > >> > > >> > > >
