Use adaptive HTTP/2 windows for the remote cache connection

[senamakel]

The remote client kept hyper's fixed HTTP/2 flow-control windows (2 MiB per stream, 5 MiB per connection). Against grpcs://cache.tinyhumans.ai from GitHub-hosted runners (~120 ms away), that caps reads at about 5 MiB / 0.12 s ≈ 40 MiB/s per connection, however fast the link is. A warm chottu-core job fetches gigabytes of hits. Endpoint::http2_adaptive_window(true) makes hyper probe the bandwidth-delay product with pings and grow the windows to match.

The server side of the same problem (nginx's 64k request-body window, which capped uploads at ~0.5 MiB/s per stream) is fixed in gh-hosted-runner#18.

Behavior change: none beyond flow control: the client may send HTTP/2 BDP pings.

Validation: cargo fmt --check, clippy -D warnings, test --all-features (exit 0) and cargo doc -D warnings are all clean.

Not covered by an automated test: Endpoint exposes no getter for its window settings, and the in-process fake has no latency to show a window limit. The effect will show as fetch throughput in chottu-core's next warm run, which I'll report on the chottu PR. No bench/bench.py numbers either: the harness measures local compiles, not WAN transfer.

Summary by CodeRabbit

• New Features
  • Remote blob reads now use HTTP/2 adaptive flow-control windows. The existing connection timeout remains in place, and ByteStream reads continue to avoid the gRPC message-size limit.
• Documentation
  • Updated the remote lookup documentation to describe adaptive flow control for blob reads and the existing ByteStream read behavior.

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: b2997fc3-03f7-4422-be05-809ee5bad991

📥 Commits

Reviewing files that changed from the base of the PR and between 2a7ef4b and 06094be.

📒 Files selected for processing: 2

• crates/tinymold-remote/src/client/mod.rs
• docs/specs/remote.md

+-----------------------------------------------------+
| Merge conflicts fear my conflict-resolution skills. |
+-----------------------------------------------------+
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

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

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

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-15T22:33:03.318085Z	6824ac2	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6824ac2371

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[tinysweeper]

Tiny Sweeper review

This pull request tunes the remote cache client's HTTP/2 connection to use adaptive flow-control windows, adding a single builder call on the tonic endpoint and a matching documentation sentence in the remote spec. Review found no defects across the critique, security, tests, commits, and description lanes; the change only alters HTTP/2 flow-control sizing without introducing new input handling, auth changes, network targets, or subprocesses.

State: Ready for maintainer review
Priority: none
Reviewed head: 06094be00138
Updated: 2026-10-15T22:41:50Z

Review snapshot

Change surface	Files	Review signal	Count
Production	1	Active findings	0
Tests	0	Noted findings	0
Documentation	1	Resolved findings	0
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: Test coverage is assessed from changed tests and lane evidence; execution is not claimed without trusted check data.

What changed

Two files changed: `crates/tinymold-remote/src/client/mod.rs` adds `.http2_adaptive_window(true)` to the tonic `Endpoint` construction in `Client::new`, with an explanatory comment about hyper's fixed windows versus bandwidth-delay product growth; `docs/specs/remote.md` adds a sentence describing that the connection uses HTTP/2 adaptive flow-control windows so reads from a distant cache are not held to a fixed window per round trip.

Features

• Added — Adaptive HTTP/2 flow-control windows for the remote cache connection: The remote client's tonic endpoint is now built with `.http2_adaptive_window(true)`, so flow-control windows can grow toward the bandwidth-delay product instead of being limited to hyper's fixed per-stream and per-connection windows, allowing a read to move more than one window per round trip. This is transport tuning only; it adds no new input handling, auth change, network target, or subprocess. (crates/tinymold-remote/src/client/mod.rs#impl Client {)
• Modified — Remote spec documentation for adaptive flow-control windows: The remote spec now states that the connection uses HTTP/2 adaptive flow-control windows, so reads from a distant cache are not held to a fixed window per round trip, keeping the documented behavior aligned with the client tuning change. (docs/specs/remote.md#sharing a key.)

Tests

• no_new_test — The change adds one builder call, `.http2_adaptive_window(true)`, to the gRPC endpoint plus a two-line doc update; the effect is remote-server window growth that no unit test in this repository can observe, and the existing `Client::allowance` tests already pin the only thing the client computes locally.: No new test is appropriate. No assertion in the diff is unverifiable and no branch is added, so nothing here can regress silently in a way a test would catch. I could not independently verify hyper's default window sizes or that `http2_adaptive_window` exists on this tonic's `Endpoint`; a wrong method name would surface only at compile time, which is not visible from the diff. (crates/tinymold-remote/src/client/mod.rs#impl Client {)

Findings

No active actionable findings.

Before merge

None.

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 2 files; 0 findings. _The code index is behind this pull request (indexed at `6824ac237142`), so retrieved context may be out of date._

security

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The diff adds a single builder call, `.http2_adaptive_window(true)`, to the tonic endpoint construction. It changes only HTTP/2 flow-control sizing and introduces no new input handling, auth change, network target, or subprocess. Nothing in the change is a reportable defect; the only caveat I could not verify is whether this tonic version exposes `http2_adaptive_window`, but the diff itself does not prove a missing symbol and the dependency is a released tonic 0.14. 1 file was not security-reviewed: docs/specs/remote.md (prose or tabular data). _The code index is behind this pull request (indexed at `6824ac237142`), so retrieved context may be out of date._

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The change adds one builder call, `.http2_adaptive_window(true)`, to the gRPC endpoint and a two-line doc update. This is a transport tuning flag whose effect is a remote-server behaviour (window growth) that no unit test in this repository can observe; the existing `Client::allowance` tests already pin the only thing the client computes locally. No assertion in the diff is unverifiable, no branch is added, and nothing here can regress silently in a way a test would catch — so the absence of a new test is appropriate. The diff introduces no defect I can identify; note that I could not independently verify hyper's default window sizes or that `http2_adaptive_window` exists on this tonic's `Endpoint` (the trait-building chain means a wrong method name would surface only at compile time, which I cannot see from the diff alone). _The code index is behind this pull request (indexed at `6824ac237142`), so retrieved context may be out of date._

commits

• Conclusion: Neutral
• Scope reviewed: all assigned evidence
• Lane summary: Nothing sensitive found in what this pull request commits.

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The change is a one-line client tuning addition (`.http2_adaptive_window(true)` on the tonic endpoint) plus a matching docs/specs/remote.md sentence; the description accurately states the behaviour change (client may send HTTP/2 BDP pings), that the server-side nginx fix lives in another PR, and that no automated test or bench numbers cover it. No defects found. _The code index is behind this pull request (indexed at `6824ac237142`), so retrieved context may be out of date._

Evidence and run details

• Models: ladder/vectors-oai3, glm-5.3-flash, deepseek-v4.1-flash
• Spend: $0.000585
• Tokens: 93284 input · 4520 output · 9472 cached · 87 embedding

Head	State	Pass summary
06094be00138	ready for maintainer review	0 active finding(s), 0 resolved finding(s) (at 2026-10-15T22:41:50Z)

[tinysweeper]

tinysweeper found nothing blocking. Approving.

             $0.0006 · 93,284 in / 4,520 out · 9,472 cached (10%) · ladder/vectors-oai3, glm-5.3-flash, deepseek-v4.1-flash · 87 embedded
critique:    $0.0003 · 26,907 in / 1,478 out · 0 cached (0%)      · glm-5.3-flash
security:    $0.0002 · 27,202 in / 1,540 out · 1,792 cached (7%)  · deepseek-v4.1-flash, glm-5.3-flash
tests:       $0.0000 · 14,120 in / 201 out   · 2,176 cached (15%) · deepseek-v4.1-flash
description: $0.0001 · 23,184 in / 680 out   · 5,504 cached (24%) · deepseek-v4.1-flash