Custom Index Creation for orientDB

[pranavasathya]

So, I came across the Custom Index Creation for OrientDB using java and Generate the JAR file and use it as plugin in OrientDB.

But I am unable to create custom index by following the documentation https://orientdb.dev/docs/3.2.x/internals/Custom-Index-Engine.html
Does anyone have created or have any idea in How this can be achieved?

[tglman]

Hi,

It depends on the version you are using, I guess is 3.2.x, and the guide is a bit outdated, because as today we do have a double layer of interfaces to plug under the indexes

[pranavasathya]

Hi @tglman thanks for replaying,
Do you have any resource or latest documentation for implementing one, It will be helpful for reference

[tglman]

Hi,

We do not have an update documentation on that, as you saw on the docs you linked, you need mainly to implement the OIndexFactory and add a resource locator file to make it load, you need also to implement the OBaseIndexEngine interface, and potentially the OIndexInternal (but not required if handling standard unique/notunique case, for that you can use built-in implementations).

Here is the code of the interfaces: https://github.com/orientechnologies/orientdb/blob/3.2.x/core/src/main/java/com/orientechnologies/orient/core/index/engine/OBaseIndexEngine.java https://github.com/orientechnologies/orientdb/blob/3.2.x/core/src/main/java/com/orientechnologies/orient/core/index/OIndexFactory.java

and here is the code of the lucene integration that you can use as example: https://github.com/orientechnologies/orientdb/blob/3.2.x/lucene/src/main/java/com/orientechnologies/lucene/engine/OLuceneIndexEngine.java https://github.com/orientechnologies/orientdb/blob/3.2.x/lucene/src/main/java/com/orientechnologies/lucene/OLuceneIndexFactory.java

I hope this helps

[pranavasathya]

What I am trying is ,
to implement a custom index with HNSW algorithm, however the orient db is supporting storage of vectors with data type embeddingList float, is these achievable

[andrii0lomakin]

Good day, @pranavasathya. We are a fork of ODB developed by JetBrains https://github.com/JetBrains/youtrackdb . I am author of storage engine in ODB that you are trying to hack and development lead of this new database . We worked on a disk-based HNSW algorithm and considired to integrate it in YTDB, though stopped as Lucene provides highly optimized compressed vector indexes. If you intersted to collaborate with us to continue our work we can discuss sharing our achievemnts and ideas so you can proceed.

You can contact us in Zulip chat https://youtrackdb.zulipchat.com

[pranavasathya]

Thanks for the Info @andrii0lomakin, Sure I will take a look at Yourtrack db , I have a project where it's tightly coupled with ODB,
So, Does ODB support Vector data type and storage and any indexes if any and is there any way we can achieve the same performance as the vector DB offers by using the ODB?

[tglman]

As today OrientDB do not support Vector indexes, I would be more than happy to have them from an external contribution, as stated by @andrii0lomakin this could also been achieved using the lucene integration, that would be probably easier and faster to achieve solution.

I know also @lvca (origianl author for OrientDB) worked on a project using HNSW maybe he also have hint on it.