Schema sync label count query exhausts memory on large graphs when statistics are disabled

[kmcginnes]

Description

Gremlin schema sync fetches edge labels and their counts with a single unbounded aggregation:

g.E().groupCount().by(label)

On a large graph this exhausts Neptune's memory and the whole schema sync fails, so the connection never becomes usable. This is worse than a Schema View problem, because nothing downstream of schema sync can run.

Measured on one Neptune 1.3.5.0 db.t3.medium with DFE enabled and statistics disabled:

Graph size	Result
57,538 edges	succeeds in about 1.2s, fully native per its explain plan
1,057,538 edges	MemoryLimitExceededException in under 10s

This code path only runs when the summary API is unavailable, which on Neptune means statistics are disabled. When the summary API works, fetchSchema takes the summary path and never issues this query. That narrows the blast radius considerably, and I have not checked how common the disabled case is in practice.

Two things I did not measure but expect to share the defect:

• vertexLabelsTemplate is g.V().groupCount().by(label), the same shape, so a graph with a comparable number of vertices should fail the same way.
• openCypher (MATCH ()-[e]-() RETURN type(e) AS label, count(*) AS count) and SPARQL have equivalent unbounded count queries.

Affected files:

• packages/graph-explorer/src/connector/gremlin/fetchSchema/edgeLabelsTemplate.ts
• packages/graph-explorer/src/connector/gremlin/fetchSchema/vertexLabelsTemplate.ts
• packages/graph-explorer/src/connector/gremlin/fetchSchema/index.ts (fetchEdgeLabels, fetchVerticesSchema)

Environment

• OS: N/A (server-side query failure)
• Browser: N/A
• Graph Explorer Version: 3.2.2, and this predates it
• Graph Database & Version: Amazon Neptune 1.3.5.0, db.t3.medium, DFE enabled, cluster statistics disabled

Steps to Reproduce

1. Use a Neptune cluster with statistics disabled, so the summary API returns BadRequestException.
2. Load roughly 1M edges.
3. Connect with Gremlin and synchronize the connection.
4. Schema sync fails. Running g.E().groupCount().by(label) directly returns MemoryLimitExceededException.

Expected Behavior

Schema sync completes on a large graph whether or not statistics are enabled. The label and count discovery needs to be bounded, most likely by chunking per label rather than aggregating the whole edge store in one request.

Related Issues

Related: #2141 (found while investigating it; the fix there does not touch this query), #1677

Important

Internal only — this issue is maintained by the core team and is not accepting external contributions.