[puskuruk]

Adds nest-gea and nest-node rows to bench/http-matrix.py and the Nest build to bench/goal-http-build.sh, aligns apps/hello-nest with the other benchmark apps (gea binary in dist/, tsc oracle in dist-node/, a cluster.node.mjs), and records the runs with BENCHMARKS.md sections.

• 1bd2de7 Add hello-nest to the HTTP benchmark matrix as nest-gea and nest-node
• 22eba03 Record the hello-nest Node against compiled run of 2026-10-19 on an arm64 Linux VM
• 2bfd355 Record the full HTTP matrix of 2026-10-19 on the arm64 Linux VM
• e61a006 Record the 2026-10-25 runs on the arm64 Linux VM with platform-node 12.2.0 and compiler 1.0.47

The Node row runs the tsc output of the same server.ts (Node's type stripping has no decorators). The current run, 2026-10-25: arm64 Ubuntu 24.04 container (clang 18.1.3, Node 24.20.0, wrk 4.1.0, cargo 1.90) on an 8-vCPU Colima VM on an Apple M1, VM and host otherwise idle, house protocol, all ten servers three complete rounds, 72 wire checks passed. The three gea binaries were built in the container by the published compiler 1.0.47, hello-nest on platform-node 12.2.0; brotli 1.2.1 was supplied through BROTLI_DIR because Ubuntu's 1.1.0 fails the build's CVE check. Not the bench host: compare ratios, not absolute sizes, across runs.

Server	1w /	1w /json	1w PSS MiB	1w startup ms	4w /	4w /json	4w PSS MiB	4w startup ms	4w p50	4w p99
hono-gea	153,784	121,440	6.4	7	376,158	323,931	8.5	8	128 µs	844 µs
hono-node	32,085	29,943	115.9	65	99,174	90,035	255.5	103	571 µs	2.02 ms
nest-gea	121,585	114,648	9.6	12	324,514	314,454	11.2	12	148 µs	736 µs
nest-node	54,336	52,106	118.5	498	175,532	165,650	350.7	924	318 µs	1.66 ms
gea-raw	232,639	229,654	4.5	9	459,725	474,422	5.9	7	96 µs	1.68 ms
node-raw	58,188	57,713	88.1	53	180,682	177,656	182.4	76	302 µs	1.17 ms
cpp-drogon	184,309	177,426	10.3	12	396,290	394,825	10.5	6	116 µs	635 µs
rust-axum	212,382	216,768	3.2	6	320,962	321,643	4.2	6	163 µs	498 µs
rust-hyper	245,773	249,293	2.3	6	340,853	341,511	2.8	6	152 µs	442 µs
cpp-epoll	281,368	280,232	2.3	6	473,147	472,736	3.4	6	90 µs	1.00 ms

GET /, compiled over Node at one and four workers: hono 4.79× and 3.79×, nest 2.24× and 1.85×, raw 4.00× and 2.54×. At four workers nest-gea is 0.86× hono-gea and 1.80× bare Node; gea-raw is 0.97× cpp-epoll and 1.35× rust-hyper. PSS 1/30 to 1/31 for all three pairs. Binaries: hono-gea 5,283,152 bytes, nest-gea 8,016,904, gea-raw 3,669,280.

Records: bench/results/http-colima-arm64-2026-10-25{,-startup}.json and a nest-only run the same day, nest-colima-arm64-2026-10-25{,-startup}.json, within 2% of the full matrix. The 2026-10-19 records (platform-node 12.1.3, compiler checkout 1.0.37 plus the four fixes) stay alongside with their own section; every row, the controls included, ran faster on 2026-10-25, so the two days compare by ratio only.

Verification on e61a006 with 1.0.47: all 33 scripts/test-*.mjs pass except test-hello-nest.mjs at the expected HELLO_NEST_PARITY: 10 of 11 identical (the JSON.parse wording; 11 of 11 with https://github.com/geastack/compiler/pull/22); tools/nest check 24 tables; python3 -m py_compile bench/http-matrix.py; bash -n bench/goal-http-build.sh; bench/summarize-http-matrix.py reproduces the table from the records.

Related

Stack: #6 (1/4) → #7 (2/4) → #8 (3/4, the app this measures) → this PR. Compiler for the current run: the published @geastack/compiler 1.0.47.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[puskuruk]

@coderabbitai review

[coderabbitai]

⚠️ Action not completed

Review rate limited.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.