Custom Vertex Id (string) on non-existent vertex returns a new vertex

[pdsway]

Potential bug report when using custom vertex id's
Cannot check vertex existence via usual query because a non-existent vertex returns a new vertex instead of null as expected.

• Version: v1.0 branch
• Storage Backend: CQL
• Mixed Index Backend: elasticsearch
• Expected Behavior: gremlin query on non-existent vertex returns null
• Current Behavior: gremlin returns a new empty vertex
• Steps to Reproduce:
  a. create a graph with custom vertex id (as string)
  1. graph options "set-vertex-id" and "allow-custom-vid-types" = true
  2. query for any non-existent vertex: g.V("mycustomid")
  3. gremlin returns a vertex with that id and "label=vertex"
  4. should return null to indicate vertex not found

[FlorianHockmann]

@li-boxuan are you available to look into this? I think you have most knowledge in this area since you've implemented the support for custom string vertex IDs.

[li-boxuan]

@pdsway I cannot reproduce. Do you mind providing some screenshots?

[pdsway]

first. tyvm for trying!
second, yes, i will work on a simple demonstration
third, let me know how you build/test (perhaps its my implementation)

some notes...

• I'm using a custom build of JG 1.0 and TP 3.7.1 on jdk17
• i access JG server via remote gremlin client
• i create a basic ConfiguredGraphFactory graph with cql/elastic plus these custom id options:
  ---- graph.set-vertex-id = true, graph.allow-custom-vid-types = true
• my graph has schema for vertex and edge labels

[pdsway]

forgot to mention...

using the same JG instance, I can switch to a "normal" graph (with Long gremlin id) and perform the same test (searching for a V with non-existing id) and I get null (as expected).

[li-boxuan]

That sounds very weird as custom vertex IDs are not fundamentally different from auto-generated IDs.

What if you set graph.allow-custom-vid-types = false, and use custom numeric IDs?

[pdsway]

What if you set graph.allow-custom-vid-types = false, and use custom numeric IDs?

also strange (graph is empty so no V's should exist)....
g.V(1) -> null
g.V(123) -> null
g.V(1234).elementMap() -> {id=1058, label=vertex}
g.V(12345).elementMap() -> {id=12345, label=vertex}

IMPORTANT: Discovered this is a side-effect of also setting storage.batch-loading = true
I was using that graph setting by default, when I removed it, gremlin functions normally for this use-case

[li-boxuan]

g.V(1234).elementMap() -> {id=1058, label=vertex}

I don't think a vertex is really created. Batch-loading disables consistency checks, so you are likely just seeing a temporary phantom vertex.