Skip to content

perf(kits): reuse pinned fetches within one invocation - #631

Merged
wz-gsa merged 1 commit into
mainfrom
perf/586-invocation-kit-fetch
Oct 26, 2026
Merged

perf(kits): reuse pinned fetches within one invocation#631
wz-gsa merged 1 commit into
mainfrom
perf/586-invocation-kit-fetch

Conversation

@mogul

@mogul mogul commented Oct 24, 2026

Copy link
Copy Markdown
Contributor

Summary

Implements the delegated within-invocation fetch-reuse increment in #586. A private invocation root shares one verified source per URL + full lowercase 40-character commit ID; callers receive independent kit directories materialized from committed objects. Symbolic and abbreviated refs keep per-call fetch behavior.

The foreground ACQ process retains its PID, stdin, TTY, signal behavior, and backend exec semantics. A private guardian cleans sources when EXIT handlers are replaced or the process execs; normal exits clean synchronously. Fallback cleanup normally occurs within the one-second polling interval. Nested invocations receive fresh roots. No persistent cache or new dependency is introduced.

Anonymous fetch, authenticated retry, credential-URL rejection, subdirectory validation, and suppressed Git diagnostics remain in place. Commit identity is checked before publishing the invocation-local verified marker. Materialization preserves executable bits, symlinks, tracked index/marker filenames, and export-ignore payloads; consumer edits do not modify the shared source.

Evidence and limitations

  • Supplied original host trace: eight remote kit fetches totaled 11.81 seconds, individually 1.28–1.84 seconds. This is prior host evidence, not a measurement made by this PR.
  • Measured offline with real Git repositories: eight same-pin materializations require one fetch; a ninth read of another kit subdirectory also uses that source. URL and revision changes fetch separately.
  • No live wall-clock saving is claimed. Host MSB create/reattach timing and authenticated transport remain unverified. Track the post-PR sandbox-capable-host run under feat(acq): investigate routinely fast new sandbox launches #586: compare the same pinned configuration, count fetches, and retain verification overhead in timing.
  • Bash 3.2/macOS/BSD runtime validation remains for CI/host verification. Local execution used Linux Bash 5.3.
  • Reuse targets the existing sequential ACQ fetch flow; this does not introduce parallel fetch coordination.

Plan and architecture

Changes: acq lifecycle ownership; acq.backends/kit-translate.sh reuse/materialization; offline transport fixture adaptation; focused real-Git and lifecycle tests. This is an implementation optimization under ADR-0011 and ADR-0030, with no new composition or distribution architecture.

Verification transcript

  • PASS: ./scripts/test-acq-bats test/bats/102-kit-fetch-reuse.bats test/bats/103-kit-fetch-lifecycle.bats test/bats/100-kit-translate.bats test/bats/50-kit-list-completeness.bats test/bats/140-opencode-migration-gate.bats — 76 tests passed, including 21 Python lifecycle scenarios covering success/failure, trap replacement, exec, signals, spinner, stdin, PTY, nesting, and owner SIGKILL.
  • PASS after final marker/BSD-sed hardening: ./scripts/test-acq-bats test/bats/102-kit-fetch-reuse.bats test/bats/103-kit-fetch-lifecycle.bats — 10 tests passed, including the 21 lifecycle scenarios.
  • PASS: shellcheck --severity=warning acq acq.backends/kit-translate.sh scripts/test-acq-lib.sh test/bats/102-kit-fetch-reuse.bats test/bats/103-kit-fetch-lifecycle.bats.
  • PASS: separate bash -n checks for acq, kit-translate.sh, and test-acq-lib.sh; git diff --check and staged diff whitespace check.
  • Security scan: gitleaks unavailable locally; used the code-review skill fallback, searching tracked diff and new tests for secrets/token/password/credential/API-key/private-key patterns and reviewing matches. Only explanatory text and synthetic credential-rejection fixtures were found. No dependency changes; ShellCheck supplied shell static analysis.
  • Full suite attempt timed out at 240 seconds, with no complete verdict. User then directed focused testing only; it was not repeated. CI remains authoritative for broader checks.
  • Adversarial review found lifecycle and subdirectory blockers in the initial version. They were fixed, focused tests added, and an independent re-review plus final-delta review found no blockers.

Security impact and rollback

Trust boundaries and secret mechanisms remain unchanged. Only verified committed objects are reused within the invocation; no raw ref is added to debug output. Roll back by reverting this commit to restore per-call fetching.

AI assistance

AI-assisted implementation, tests, and review by OpenCode (gpt-6-astra), including independent adversarial sub-agent review. Human review is pending and required before merge. No merge requested.

@mogul
perf(kits): reuse pinned fetches within one invocation
be830b9 
Materialize independent kit directories from verified Git objects and clean\ninvocation-owned sources through a foreground-safe lifecycle guardian.\n\nRefs: #586\n\nCo-authored-by: OpenCode [gpt-6-astra] <[email protected]>
@mogul
mogul requested a review from a team as a code owner October 24, 2026 19:10
@mogul

mogul commented Oct 24, 2026

Copy link
Copy Markdown
Contributor Author

Host measurement update

Human-run host measurement completed from an isolated checkout of this PR at be830b95f1b0bfacead75f339bab51e261626a4c, using a normal MSB acq run opencode flow with the configured kit set.

Observed result

  • Exactly one fetch: remote kit -> ... debug event occurred.
  • The same pinned Patterns revision subsequently supplied provider facts, built-in agent selection, configured extra-kit resolution, and all applied kit directories.
  • Sandbox creation, provisioning, USAi validation, and Paseo web UI readiness completed successfully.
  • The first fetch began at 12:39:00; provisioning reported all kits applied at 12:39:38. That 38-second interval includes picker interaction, provider validation, create/boot, file staging, kit startup, and other work. It is not a fetch-only or comparable end-to-end timing claim.

Comparison boundary

The prior trace used to motivate this PR observed eight remote fetches totaling 11.81 seconds. That run used a different Quickstart/Patterns revision and host configuration state. This PR run proves the fetch-count reduction from eight to one in the normal flow; it does not independently prove an exact wall-clock saving. The one observed remote fetch took roughly two wall-clock seconds based on terminal timestamps, but that is not a controlled comparison.

No credential values were displayed or recorded. AI-assisted measurement analysis by OpenCode; the human executed the host command.

@mogul mogul mentioned this pull request Oct 24, 2026
Open
2 of 8 tasks
wz-gsa
wz-gsa approved these changes Oct 26, 2026
View reviewed changes

@wz-gsa wz-gsa left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Adversarial fixed-head static review of be830b95f1b0bfacead75f339bab51e261626a4c: reuse is limited to full immutable commit refs within an invocation-local private root; the fetched HEAD is checked before marking it verified, and consumer payloads are materialized from committed objects into independent directories. I inspected the lifecycle guardian, signal/exec cleanup, credential-URL rejection, fallback fetch path, and the real-Git/PTY/signal offline fixtures. No blocker found. All three required current-head checks pass. This is not a claim of measured live MSB speedup or authenticated-transport verification; those remain tracked under #586. No reviewed-repository code was executed on the host. AI-assisted static review (OpenCode); approval does not authorize merge.

@wz-gsa
wz-gsa added this pull request to the merge queue Oct 26, 2026
Merged via the queue into main with commit b433a61 Oct 26, 2026
5 checks passed
@wz-gsa
wz-gsa deleted the perf/586-invocation-kit-fetch branch October 26, 2026 21:59
@github-actions github-actions Bot mentioned this pull request Oct 26, 2026
Merged
mogul pushed a commit that referenced this pull request Nov 5, 2026
@github-actions
chore(main): release 4.2.0 (#617)
f75c87b 
🤖 I have created a release *beep* *boop*
---


##
[4.2.0](v4.1.1...v4.2.0)
(2026-11-05)


### Features

* **acq:** project provider inputs read-only
([#608](#608))
([bd11c77](bd11c77))
* **sbx:** trust host state for built-in kit heal; add MSB forged-marker
and legacy-sandbox coverage
([#637](#637))
([ef24965](ef24965))
* **scripts:** add summarize-bats-timing for bats TAP timings
([#652](#652))
([5bf9398](5bf9398))


### Bug Fixes

* **agents:** move the default OpenCode image to devenv-shell 2.1.0
([#668](#668))
([dbd8a76](dbd8a76))
* **common:** canonicalize macOS /var,/tmp,/etc aliases before
O_NOFOLLOW traversal
([#622](#622))
([1135fcc](1135fcc))
* **install:** package provider validator runtime
([#619](#619))
([5192fa1](5192fa1))
* **msb:** distinguish denied probe writes from writable mounts
([#670](#670))
([049931c](049931c))
* **msb:** strip CR from jq output in acq ports (Windows)
([#663](#663))
([7ad6e56](7ad6e56))
* **release:** use supported macOS wheelhouse runners
([#674](#674))
([9dbc9aa](9dbc9aa))


### Performance Improvements

* **acq:** stop the kit-fetch guardian tick from delaying exit
([#664](#664))
([fe686ea](fe686ea))
* **kits:** reuse pinned fetches within one invocation
([#631](#631))
([b433a61](b433a61))
* **msb:** reduce foreground node kit invocation latency
([#616](#616))
([ec73abd](ec73abd))
* **test:** eliminate redundant bats fetches and waits
([#621](#621))
([dda5fe1](dda5fe1))
* **test:** shorten hostport and cli-kits test tail
([#654](#654))
([a3acc7b](a3acc7b))
* **test:** shorten port fixtures and parallelize shellcheck
([#624](#624))
([8eb3ff8](8eb3ff8))

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wz-gsa wz-gsa wz-gsa approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@mogul @wz-gsa