This repository was archived by the owner on Nov 21, 2026. It is now read-only.
Repository navigation
benchmarks: Neo4j 2026.10.0, Memgraph 3.14.0, Qdrant v1.20.0 comparator pins and clients - #482
Merged
Merged
Conversation
…rator pins, with their clients Latest stable releases, pinned by amd64 manifest digest. Neo4j 2026.08.1 -> 2026.10.0 (c3903fa9...), Memgraph 3.13.1 -> 3.14.0 (c7478d17...), Qdrant v1.19.1 -> v1.20.0 (2d1b0bfd...). Clients: neo4j 6.3.1 -> 6.4.1, neo4j-rust-ext 6.3.1.0 -> 6.4.1.0, qdrant-client 1.19.1 -> 1.20.0. Vendor config unchanged (no release note requires a change). neo4j_memory_recommendation.json regenerated on the new image (DECISIONS #202): the recommendation is identical at all eight caps, so NEO4J_MEMORY is unchanged. The 12 stage drafts that bake the digests have a new sha list. Smoke on the laptop (graph, dense, cross-model): result digests equal across Neo4j, Memgraph and the composed stack; Neo4j quantization defaults re-probed (unset BINARY 3.0, NONE 1.0, SCALAR 1.5). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…s_2611.sha256 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…raph-qdrant # Conflicts: # benchmarks/experiments/COMPARATORS.md # benchmarks/experiments/QUANTIZATION.md # benchmarks/experiments/stage_drafts_2611.sha256
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps the Neo4j, Memgraph and Qdrant comparator pins and their clients to the latest stable releases for the 2026-11 comparator subset (SUBSET-READINESS item 3). Vendor configuration is unchanged: no release note requires a change.
sha256:c3903fa96db31479c2a28e5d00e05260b039ff856f8b2cca7280c78299877731sha256:c7478d1737edb77fa2b1fcd4e37e34a68f10272d0ec7d7f9aa92d0afc8e0be9fsha256:2d1b0bfd0a961275481ea09400fc98008216ae54e1b52ed7d24520d63050c644Changed:
runner.BACKENDS,Dockerfile.composed,vendor_memory.NEO4J_IMAGE_DIGEST,build_images.sh[client],smoke_memgraph_server_limit.sh, COMPARATORS.md, QUANTIZATION.md, PROTOCOL.md, code comments that name the pin, andstage_drafts_2611.sha256(12 drafts bake the digests). If the other engine-bump PRs merge first, rewrite that list withpython3 test_stage_generator_2611.py --rewrite.Neo4j memory recommendation (DECISIONS #202):
neo4j_memory_recommendation.jsonregenerated on the new image withvendor_memory.py --regenerate. The tool output is identical at all eight caps, soNEO4J_MEMORYandtest_vendor_memory.pyneed no change.Release notes that matter. Neo4j 2026.09.0 (skipped by our pin) made re-scored binary and scalar quantization search about 5x faster and changed the memory guidance for vector indexes (Lucene files live in the OS file cache); 2026.10.0 has no vector or index change. A probe on the pinned image confirms an index with no
vector.quantization.typeis still BINARY (expansion 3.0), NONE 1.0, SCALAR 1.5, and the arms set and read back the type and 1.0. Memgraph 3.14.0 lists comparison and NaN semantic changes and fixes vector index crash recovery; the lane's Cypher is unchanged and every result digest still matched; the vector index still takes no HNSW parameters (expansion_add 128, expansion_search 64 read back). Qdrant v1.20.0 changes HNSW level assignment (6.25% of points on level 1 at m=16, was 25%), addsturbo8/turbo16andinline_storage(unused); the sparse default is still Float32. Details in DECISIONS #209.Smoke (laptop, micro/e2 scale, one rep, private
dbbench:client-nmqanddbbench:composed-nmqimages built from the new pins): l2 graph oltp and olap on Neo4j and Memgraph, all result digests equal between the two engines; e2 on Neo4j, Memgraph and the composed Qdrant+Neo4j stack, retrieval and filtered-candidate digests equal across all three; l3d dense, six arms (qdrant fp32/int8, neo4j fp32/int8, memgraph fp32/int8) ran, with the Neo4j type and expansion read back. Client image reads backneo4j 6.4.1,neo4j-rust-ext 6.4.1.0,qdrant-client 1.20.0.Tests:
test_vendor_memory, test_engine_pin, test_gen_engines_doc, test_sparse_mp_versions, test_memory_probe, test_fairness_headers, test_stage_generator_2611, test_stage_generator_after, test_milvus_compose, test_jpype_pin: 198 passed, 1 skipped;version_pin_check.py: 0 disagreements.Not done here: the l3s sparse Qdrant smoke (the runner's host lock was held by another agent's run),
docker pulland thedbbench:client/dbbench:composedrebuild on mini (install step).🤖 Generated with Claude Code