BE-273: HashQL: Profile and benchmark MIR interpreter

[indietyp]

🌟 What is the purpose of this PR?

Replace imbl with rpds for persistent data structures in the MIR interpreter, and optimize interpreter performance.

The interpreter is now only 20-40% slower instead of the previous 100%+.

🔍 What does this change?

• Replaces imbl dependency with rpds for persistent data structures (List and Dict)
• Adds inlining annotations to hot code paths in the interpreter
• Optimizes SwitchTargets::target() with a fast path for common cases
• Adds cold path annotations to error handling code
• Optimizes value comparison in the interpreter
• Adds a new benchmark for the interpreter (fibonacci recursive)
• Improves parameter passing in the interpreter to avoid unnecessary copies
• Optimizes local variable storage in the interpreter

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

🛡 What tests cover this?

• New benchmark for the interpreter (fibonacci recursive)
• Existing tests continue to pass

❓ How to test this?

1. Run the new benchmark to verify performance improvements
2. Run existing tests to ensure functionality is preserved

[cursor]

PR Summary

Medium Risk
Touches core MIR interpreter execution (locals storage, call/arg handling, switch dispatch and collections), so subtle semantic or performance regressions are possible despite being mostly internal changes and benchmark-focused.

Overview
Adds interpreter benchmarking and tunes MIR interpreter hot paths. Introduces a shared run_bencher harness for codspeed/criterion benches and adds a new interpret benchmark (recursive Fibonacci) alongside the existing transform benchmarks.

Performance-focused refactors in the interpreter: replaces persistent collections from imbl to rpds (List/Dict), reduces copying by switching many interpreter APIs to take references and requiring ExactSizeIterator for argument passing, and reworks locals storage to avoid Option overhead by filling missing locals with Unit.

Micro-optimizations and hints: adds #[inline] on hot methods, marks error paths with cold_path/likely_unlikely, and adds a fast-path in SwitchTargets::target() for small target sets; also updates workspace deps/lockfile accordingly and adds a small IdVec::reserve helper used by the runtime.

^{Written by Cursor Bugbot for commit 7476275. This will update automatically on new commits. Configure here.}

[indietyp]

• BE-304: HashQL: Terminator placement based on capabilities #8359
• BE-303: HashQL: Split MIR basic blocks by target support #8354
• BE-364: Rework symbol representation for speed and size #8343
• BE-360: HashQL: Rework compiletest UI #8340
• BE-302: HashQL: Statement cost analysis for execution target placement #8339
• BE-317: HashQL: Implement TraversalExtraction pass #8331
• BE-315: HashQL: Add GraphReadFilter source variant to MIR #8323
• BE-301: HashQL: Implement size estimation analysis for MIR #8278
• BE-273: HashQL: Profile and benchmark MIR interpreter #8251 👈 (View in Graphite)
• BE-272: HashQL: Add numeric addition and subtraction operators to MIR #8250
• BE-257: HashQL: Implement MIR interpreter with runtime value representation #8245
• BE-271: HashQL: Implement PostInline pass #8240
• BE-269: HashQL: Move PreInline out into Canonicalization pass #8239
• BE-268: HashQL: Rename PreInlining to PreInline for consistency #8238
• BE-270: HashQL: Optimize empty tuples to unit constants in MIR #8237
• BE-197: Implement MIR inline pass for HashQL #8236
• BE-267: Use bump scopes instead of full resets across transformation passes #8234
• BE-266: HashQL: Implement PreInlining pass for MIR optimization #8233
• BE-265: HashQL: Make CopyPropagation aware of local re-assignments #8231
• BE-264: HashQL: Rename SROA to ForwardSubstitution for better semantic clarity #8230
• BE-263: HashQL: add body! macro for declarative MIR construction in tests #8229
• BE-261: HashQL: Implement Administrative Reduction pass for MIR #8227
• BE-262: HashQL: Do not intern empty slices #8223
• BE-260: HashQL: Switch to bump allocator with scoping support #8218
• BE-259: HashQL: Add benchmarks for type system operations #8216
• BE-227: HashQL: Implement call graph analysis for MIR #8214
• BE-258: HashQL: Add Changed enum to track MIR transformation pass modifications #8210
• BE-255: HashQL: Implement Copy Propagation (CP) pass for MIR #8207
• BE-231: HashQL: Add InstSimplify pass for MIR instruction simplification and constant folding #8206
• BE-254: Add CodSpeed benchmarking for HashQL MIR transformation passes #8204
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[TimDiekmann]

You probably guessed that I will comment on this file 😅

I'm not thrilled about the raw pointer approach, but it looks sound. PerIteration is indeed critical for soundness to prevent use-after-free and RefCell does not work here.

I have a few concerns (if any of these apply, this will become unsound)

• If DiagnosticIssues ever gains a lifetime parameter
• If criterion/codspeed ever change iter_batched_ref semantics (unlikely, but I don't like relying on implementation details)
• As mentioned above: if BatchSize is changed

Can you assert, that MirDiiagnosticIssues: 'static so we don't accidentally capture the heap-lifetime?

[indietyp]

Yes, me too, but this is the only way I could make this work. I tried multiple other ways, usually you would do this via iter_custom, but that isn't supported by codspeed. So run_bencher was born. This is also why the function is practically only safety comments. FWIW although not explicitely stated, PerIteration docs suggest very strongly that any value will only be held once at a time (even giving the file lock example: https://docs.rs/criterion/latest/criterion/enum.BatchSize.html#variant.PerIteration)

[indietyp]

I have added a static lifetime assertion inside of 7476275

[TimDiekmann]

Great, thank you!

Your organization requires reapproval when changes are made, so Graphite has dismissed approvals. See the output of git range-diff at https://github.com/hashintel/hash/actions/runs/21512891851

[TimDiekmann]

Great, thank you!

[github-actions]

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	$$27.3 \mathrm{ms} \pm 195 \mathrm{μs}\left({\color{red}5.46 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$3.23 \mathrm{ms} \pm 14.4 \mathrm{μs}\left({\color{gray}1.06 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 1001	$$12.1 \mathrm{ms} \pm 97.8 \mathrm{μs}\left({\color{gray}4.01 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 3314	$$43.0 \mathrm{ms} \pm 400 \mathrm{μs}\left({\color{gray}3.94 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$15.6 \mathrm{ms} \pm 98.7 \mathrm{μs}\left({\color{red}15.6 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 1526	$$24.8 \mathrm{ms} \pm 203 \mathrm{μs}\left({\color{red}10.5 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 2078	$$43.0 \mathrm{ms} \pm 318 \mathrm{μs}\left({\color{gray}2.27 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$20.1 \mathrm{ms} \pm 148 \mathrm{μs}\left({\color{gray}2.82 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 1033	$$30.8 \mathrm{ms} \pm 251 \mathrm{μs}\left({\color{red}13.7 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_medium

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 102	$$3.80 \mathrm{ms} \pm 22.0 \mathrm{μs}\left({\color{red}6.24 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$2.92 \mathrm{ms} \pm 15.1 \mathrm{μs}\left({\color{gray}4.32 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 51	$$3.34 \mathrm{ms} \pm 13.6 \mathrm{μs}\left({\color{red}5.65 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 269	$$5.08 \mathrm{ms} \pm 31.4 \mathrm{μs}\left({\color{gray}3.50 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$3.41 \mathrm{ms} \pm 15.0 \mathrm{μs}\left({\color{gray}2.40 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 107	$$3.99 \mathrm{ms} \pm 18.7 \mathrm{μs}\left({\color{gray}2.88 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 133	$$4.57 \mathrm{ms} \pm 37.0 \mathrm{μs}\left({\color{red}9.62 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$3.45 \mathrm{ms} \pm 13.6 \mathrm{μs}\left({\color{red}7.49 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 63	$$4.21 \mathrm{ms} \pm 26.5 \mathrm{μs}\left({\color{red}9.61 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_none

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 2	$$2.36 \mathrm{ms} \pm 10.0 \mathrm{μs}\left({\color{gray}-0.788 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$2.32 \mathrm{ms} \pm 17.0 \mathrm{μs}\left({\color{gray}0.806 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 1	$$2.42 \mathrm{ms} \pm 13.6 \mathrm{μs}\left({\color{gray}0.126 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 8	$$2.62 \mathrm{ms} \pm 11.6 \mathrm{μs}\left({\color{gray}0.219 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$2.48 \mathrm{ms} \pm 10.4 \mathrm{μs}\left({\color{gray}-0.070 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 3	$$2.72 \mathrm{ms} \pm 17.3 \mathrm{μs}\left({\color{gray}2.00 \mathrm{\%}}\right) $$	Flame Graph

policy_resolution_small

Function	Value	Mean	Flame graphs
resolve_policies_for_actor	user: empty, selectivity: high, policies: 52	$$2.75 \mathrm{ms} \pm 15.4 \mathrm{μs}\left({\color{gray}-0.835 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: low, policies: 1	$$2.38 \mathrm{ms} \pm 10.7 \mathrm{μs}\left({\color{gray}-1.273 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: empty, selectivity: medium, policies: 25	$$2.59 \mathrm{ms} \pm 13.3 \mathrm{μs}\left({\color{gray}0.265 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: high, policies: 94	$$3.08 \mathrm{ms} \pm 15.3 \mathrm{μs}\left({\color{gray}0.702 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: low, policies: 1	$$2.64 \mathrm{ms} \pm 13.2 \mathrm{μs}\left({\color{gray}-0.327 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: seeded, selectivity: medium, policies: 26	$$2.87 \mathrm{ms} \pm 11.8 \mathrm{μs}\left({\color{gray}0.545 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: high, policies: 66	$$3.00 \mathrm{ms} \pm 19.7 \mathrm{μs}\left({\color{gray}0.976 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: low, policies: 1	$$2.61 \mathrm{ms} \pm 15.8 \mathrm{μs}\left({\color{gray}-0.156 \mathrm{\%}}\right) $$	Flame Graph
resolve_policies_for_actor	user: system, selectivity: medium, policies: 29	$$2.85 \mathrm{ms} \pm 14.1 \mathrm{μs}\left({\color{gray}0.564 \mathrm{\%}}\right) $$	Flame Graph

read_scaling_complete

Function	Value	Mean	Flame graphs
entity_by_id;one_depth	1 entities	$$39.2 \mathrm{ms} \pm 161 \mathrm{μs}\left({\color{gray}0.540 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	10 entities	$$76.8 \mathrm{ms} \pm 447 \mathrm{μs}\left({\color{gray}1.35 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	25 entities	$$44.2 \mathrm{ms} \pm 198 \mathrm{μs}\left({\color{gray}3.02 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	5 entities	$$46.0 \mathrm{ms} \pm 180 \mathrm{μs}\left({\color{gray}-0.110 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;one_depth	50 entities	$$53.4 \mathrm{ms} \pm 232 \mathrm{μs}\left({\color{gray}-0.588 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	1 entities	$$41.1 \mathrm{ms} \pm 185 \mathrm{μs}\left({\color{gray}-0.673 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	10 entities	$$415 \mathrm{ms} \pm 1.03 \mathrm{ms}\left({\color{gray}0.037 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	25 entities	$$93.6 \mathrm{ms} \pm 401 \mathrm{μs}\left({\color{gray}1.28 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	5 entities	$$84.8 \mathrm{ms} \pm 385 \mathrm{μs}\left({\color{gray}-1.807 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;two_depth	50 entities	$$306 \mathrm{ms} \pm 879 \mathrm{μs}\left({\color{gray}-3.922 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	1 entities	$$14.4 \mathrm{ms} \pm 68.1 \mathrm{μs}\left({\color{lightgreen}-5.116 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	10 entities	$$14.7 \mathrm{ms} \pm 78.5 \mathrm{μs}\left({\color{lightgreen}-5.034 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	25 entities	$$15.3 \mathrm{ms} \pm 81.8 \mathrm{μs}\left({\color{gray}-2.826 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	5 entities	$$14.4 \mathrm{ms} \pm 78.6 \mathrm{μs}\left({\color{lightgreen}-5.975 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id;zero_depth	50 entities	$$17.7 \mathrm{ms} \pm 96.7 \mathrm{μs}\left({\color{gray}-1.027 \mathrm{\%}}\right) $$	Flame Graph

read_scaling_linkless

Function	Value	Mean	Flame graphs
entity_by_id	1 entities	$$14.3 \mathrm{ms} \pm 74.1 \mathrm{μs}\left({\color{lightgreen}-6.792 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10 entities	$$14.5 \mathrm{ms} \pm 89.1 \mathrm{μs}\left({\color{lightgreen}-5.729 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	100 entities	$$14.4 \mathrm{ms} \pm 68.7 \mathrm{μs}\left({\color{lightgreen}-6.517 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	1000 entities	$$15.1 \mathrm{ms} \pm 93.4 \mathrm{μs}\left({\color{lightgreen}-7.372 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10000 entities	$$22.3 \mathrm{ms} \pm 152 \mathrm{μs}\left({\color{lightgreen}-7.324 \mathrm{\%}}\right) $$	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	$$29.7 \mathrm{ms} \pm 283 \mathrm{μs}\left({\color{gray}0.715 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/book/v/1	$$30.2 \mathrm{ms} \pm 317 \mathrm{μs}\left({\color{gray}-2.645 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/building/v/1	$$29.2 \mathrm{ms} \pm 306 \mathrm{μs}\left({\color{gray}0.151 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/organization/v/1	$$29.0 \mathrm{ms} \pm 285 \mathrm{μs}\left({\color{gray}-3.453 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/page/v/2	$$30.1 \mathrm{ms} \pm 342 \mathrm{μs}\left({\color{gray}2.37 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/person/v/1	$$29.1 \mathrm{ms} \pm 284 \mathrm{μs}\left({\color{gray}-3.437 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/playlist/v/1	$$30.2 \mathrm{ms} \pm 289 \mathrm{μs}\left({\color{gray}0.894 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/song/v/1	$$29.6 \mathrm{ms} \pm 337 \mathrm{μs}\left({\color{gray}2.32 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: https://blockprotocol.org/@alice/types/entity-type/uk-address/v/1	$$30.6 \mathrm{ms} \pm 256 \mathrm{μs}\left({\color{gray}1.75 \mathrm{\%}}\right) $$	Flame Graph

representative_read_entity_type

Function	Value	Mean	Flame graphs
get_entity_type_by_id	Account ID: bf5a9ef5-dc3b-43cf-a291-6210c0321eba	$$7.98 \mathrm{ms} \pm 37.3 \mathrm{μs}\left({\color{gray}-0.240 \mathrm{\%}}\right) $$	Flame Graph

representative_read_multiple_entities

Function	Value	Mean	Flame graphs
entity_by_property	traversal_paths=0	0	$$45.4 \mathrm{ms} \pm 242 \mathrm{μs}\left({\color{gray}-0.815 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=255	1,resolve_depths=inherit:1;values:255;properties:255;links:127;link_dests:126;type:true	$$93.0 \mathrm{ms} \pm 450 \mathrm{μs}\left({\color{gray}-0.065 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:0;link_dests:0;type:false	$$51.3 \mathrm{ms} \pm 353 \mathrm{μs}\left({\color{gray}-0.210 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:1;link_dests:0;type:true	$$59.1 \mathrm{ms} \pm 332 \mathrm{μs}\left({\color{gray}-0.357 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:2;links:1;link_dests:0;type:true	$$67.4 \mathrm{ms} \pm 320 \mathrm{μs}\left({\color{gray}-0.592 \mathrm{\%}}\right) $$
entity_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:2;properties:2;links:1;link_dests:0;type:true	$$73.9 \mathrm{ms} \pm 399 \mathrm{μs}\left({\color{gray}-0.476 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=0	0	$$49.5 \mathrm{ms} \pm 288 \mathrm{μs}\left({\color{gray}1.30 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=255	1,resolve_depths=inherit:1;values:255;properties:255;links:127;link_dests:126;type:true	$$77.2 \mathrm{ms} \pm 375 \mathrm{μs}\left({\color{gray}2.05 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:0;link_dests:0;type:false	$$56.3 \mathrm{ms} \pm 386 \mathrm{μs}\left({\color{gray}0.949 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:0;links:1;link_dests:0;type:true	$$63.3 \mathrm{ms} \pm 337 \mathrm{μs}\left({\color{gray}-0.341 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:0;properties:2;links:1;link_dests:0;type:true	$$65.5 \mathrm{ms} \pm 377 \mathrm{μs}\left({\color{gray}-0.107 \mathrm{\%}}\right) $$
link_by_source_by_property	traversal_paths=2	1,resolve_depths=inherit:0;values:2;properties:2;links:1;link_dests:0;type:true	$$65.4 \mathrm{ms} \pm 355 \mathrm{μs}\left({\color{gray}-0.302 \mathrm{\%}}\right) $$

scenarios

Function	Value	Mean	Flame graphs
full_test	query-limited	$$135 \mathrm{ms} \pm 578 \mathrm{μs}\left({\color{gray}-1.131 \mathrm{\%}}\right) $$	Flame Graph
full_test	query-unlimited	$$133 \mathrm{ms} \pm 584 \mathrm{μs}\left({\color{gray}-3.256 \mathrm{\%}}\right) $$	Flame Graph
linked_queries	query-limited	$$106 \mathrm{ms} \pm 492 \mathrm{μs}\left({\color{red}8.27 \mathrm{\%}}\right) $$	Flame Graph
linked_queries	query-unlimited	$$621 \mathrm{ms} \pm 3.99 \mathrm{ms}\left({\color{red}11.0 \mathrm{\%}}\right) $$	Flame Graph