Skip to content

feat(hosting): embed concurrent repository hosts behind gateways - #66

Merged
pawelchcki merged 34 commits into
mainfrom
feat/embeddable-repository-host
Nov 16, 2026
Merged

pawelchcki merged 34 commits into
mainfrom
feat/embeddable-repository-host

Conversation

@pawelchcki

@pawelchcki pawelchcki commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor

An embedding product can host ordinary Git repositories with its own users/API keys, browse them over HTTP, manage lifecycle/policy, and catch up on durable push records. The module supports both a separate Worker (HTTP/service binding) and an embedded Worker whose Rust router calls the Durable Object stub directly. All six original milestones remain implemented. The accepted fifth-review admission design and caps remain unchanged. The implementation also protects shared chunk ownership during cleanup, preserves concurrent job recovery scheduling, shares cold initialization between alarms and requests, and budgets signed tree/history pages.

Delivered behavior

  • Scoped read/write/admin grants, constant-time verification at entry and stub arrival, hidden invalid requests, Git-readable write denials, no gateway challenge.
  • Explicit idempotent SHA-1/SHA-256 create/default branch, describe, and bounded/resumable irreversible delete. Gateway requires explicit creation; token implicit creation stays configurable. Delete fences/cancels jobs, waits for their I/O, purges only its own prefix and permanently retires the namespace.
  • Current-reachable refs/history/tree/commit/safe streamed blobs. Ref-rooted reads follow authenticated graph links; returned object proofs enable immediate navigation. Other raw selectors use a persisted bounded outward index; additive publications carry proofs/index progress, destructive updates revoke/rebuild.
  • Atomic generation/time/opaque actor/ref-change records with incarnation-bound paged catch-up.
  • Per-repository force and push/stored-payload policy. Read scratch/cache stay outside write quota. Gateway wants default to current reachability; historic catalog access is explicit.
  • Configurable compiled codec, bindings/routes/key prefix, direct-stub request contract, optional EU jurisdiction, optional/disabled health/test routes. Separate deployment remains thin; reproducible codec pin is an immutable main ancestor.

Grants

X-Git-Zero-Grant: v1.<namespace>.<read|write|admin>.<expiry>.<actor64>.<mac64>
HMAC-SHA-256 over exact UTF-8 git-zero-grant-v1\n<namespace>\n<access>\n<expiry>\n<actor64>\n, including final LF; canonical unpadded base64url fields.

This boring MAC avoids cross-language JSON canonicalization. Use the exact UTF-8 secret (at least 32 bytes), synchronized clocks and about 120-second grants; strict expiry stays now < expiry <= now+300. Authorization is captured at DO arrival. The contract and public test vector define every route, error and forwarded stub field.

Storage and browse correctness

  • Encoded chunk claims and PUTs coordinate with failed-stage deletion per physical key across job views. Ownership is rechecked after waiting; deduplication/accounting remain correct, while unrelated keys proceed concurrently.
  • Job views forward recovery deadlines to the shared Store. A shared scheduling revision also prevents a running recovery pass from clearing a retry scheduled by a protected job while it awaits R2. Warm alarms reclaim failed quarantine and temporary scratch after deferred cleanup. Alarms and requests share one cold initialization promise, keeping the Store private until orphan cleanup finishes; neither arrival order can expose live stages to another cold recovery pass.
  • Tree and history pagination include signed record proofs and their serialized response/cursor envelopes before selecting entries; valid small JSON budgets return shorter pages instead of 413, including long SHA-256 namespaces.

All six defects were independently reproduced. review-bot-regressions.test.mjs has ten cases reproducing the failures on the reviewed heads: concurrent chunk loss, warm encoded/temporary recovery, 8 KiB tree pages in both formats, SHA-256 history with an 80-character namespace, protected encoded/temporary cleanup rescheduled during an alarm, and request-first/alarm-first cold initialization overlap. The latest 22 focused recovery/lifecycle/grant/storage-policy regression tests and complete 270-test host suite pass. A constructor-bypassing cleanup fixture explicitly initializes the shared recovery marker; its assertions and deadlines are unchanged.

Fifth-review fixes

Item Implemented fix and regression
1 Keep one heavy arena for a co-tenant even before it arrives: default two heavy leases/namespace, three/isolate, fourth arena for metadata. Namespaces with no heavy lease precede incumbents. No preemption. admission.test.mjs saturates A, queues another A request, then admits B's clone and push within one second; the three-blob saturation/300-browse-read tests remain. Fairness is per repository, not per team. Admission-group keys are a documented limit; the gateway must enforce team-wide quotas.
2 Reserve cache capacity from a conservative catalog bound before graph preparation. Oversized/full-cache requests prepare directly. shared-pack.test.mjs counts exactly one/three preparations for one/three uncacheable clones and exactly one for a new key against a full cache. No N+1 discarded walk.
3 Explicit token pools validate the fair working-byte allowance. host-module.test.mjs rejects 64/96/99 MiB, performs a real fetch at the exact 100 MiB minimum, and pushes/clones at 101 MiB. Default token mode has no isolate-wide memory bound, plus up to four idle codecs/factory (24 MiB allowance, 22.75 MiB actual linear memory). concurrent-git.test.mjs pauses A's actual pack until B's pack begins, rather than mistaking v2 ls-refs for a pack; overlap, checkout/fsck and the original 30-second assertion remain.
4 Cancel builders with no followers and no output for a response-throughput window. Oversized/full-cache requests never start a builder. shared-pack.test.mjs abandons all followers before output and requires cleanup within its scaled window/bound. Gateway defaults: 256 MiB decoded push cap, 1 GiB codec pack cap (also fetch output), explicitly raisable; token defaults unchanged. host-module.test.mjs checks both modes.
5 Shared readers withhold incomplete pkt-lines, then emit a fatal channel-3 retry error at a complete framing boundary after mid-stream failure. Non-sideband requests bypass sharing. shared-pack.test.mjs cuts a cache chunk inside a packet; shared-failure-git.test.mjs injects a second-chunk PUT failure and requires real Git's “shared pack generation failed; retry fetch” remote message without framing/disconnect diagnostics.
6 Independently compute the expected page/path objects with Git and read their delta depths before the API call. browse-scale.test.mjs requires observed work to equal that reference and SQL < 24 + 5 × reference work, with zero writes and a reference-adjusted 1k-to-25k growth check. API overwork can no longer raise its own allowance.
7 Repeated workerd memory and default-packed 25k full-clone measurements; results below.

Eight targeted regressions copied into an isolated archive of cf3655e9 all failed there and pass with the fixes: heavy admission, invalid token pool, default caps, one/three uncacheable walks, orphan cleanup, partial-frame fatal error and real-Git fatal error. Item 6 strengthens an oracle; ordinary baseline read costs were already correct.

Measurements

The fifth-review verification record contains methods, figures and limits.

  • Co-tenant clone plus push behind A's heavy maximum: 7.74 ms, under the 1000 ms assertion. Existing 300-read fairness run: light tenant maximum 19.69 ms, neither clone cancelled.
  • Workerd memory, two live generation streams +30 cached readers +100 further cache-hit admissions: 60.94 MiB ledger peak /64 MiB, three charged arenas, 17.0625 MiB actual Wasm. After the burst: 9.97 MiB JS used, 31.15 MiB backing storage, 1.56 MiB embedder heap; after GC with streams live: 3.02/5.04/0.26 MiB. Inspector samples, not RSS/worst-case memory or a combined Rust-product measurement.
  • 25k commits /150,028 stored default-packed objects: 17.979 s selection +2.753 s preparation =20.732 s, headers 20.747 s. Strict Git import and complete set comparison verified 150,025 reachable objects, 11,774,780 wire bytes. Templates: 21 compilations/537,840 hits. Previous sample was 26.043 s; this sample is 5.311 s lower, without a causal performance claim. The 30-second setting is a progress-inactivity bound, not a total preparation deadline.
  • Ref-rooted commit/history20/tree/blob remain 14/70/70/76 SQL statements, zero writes. Independent reference work: 1/20/22/32. Raw absence proof: 63 batches; signed browser navigation: zero building retries.
  • Sixteen real clones, one shared build: Wrangler 18.193 s single /18.901 s sixteen, browse 21 ms, push 44 ms; direct workerd 17.521/17.723 s, browse 13 ms, push 53 ms. Snapshot/fsck, same/different-ref races and deletion remain checked.

Decisions and assumptions

  • Fairness rotates repository namespaces, prioritizes zero-heavy owners and does not preempt active jobs. Larger/legacy profiles and several saturated namespaces may still receive retryable 503 after bounded waits. More namespaces give a team more turns; group keys are deliberately deferred and documented.
  • The default 64 MiB pool retains four charged 6 MiB arenas and 13 MiB metadata working headroom. Per-route reservations, separate cached-reader slots, bounded waiters, release-before-wait and existing cancellation/throughput rules remain. Cached reader reservation includes its bounded packet buffer.
  • Token mode bypasses the pool unless explicitly opted in. Default legacy caches require 100 MiB for one fetch, more for push-prefix copies; 112 MiB still fits only one heavy fetch. Lower caches for concurrency and budget idle codecs. Embedders must measure product Rust/Wasm, V8/compiled code, HTTP/SQL and GC together inside 128 MB.
  • Gateway pack transport floor is 16 KiB/s over 30-second credit windows, with no fixed transfer lifetime. Default caps imply about 4.55 hours push input /18.20 hours pack output, plus credit and separately bounded admission/controls; processing/I/O retain their own limits. Raising to 64 GiB permits ~48.5 days at the floor. A 300 MB push needs an explicitly raised cap. Controls/JSON retain their 120-second deadline after admission.
  • Cache estimation includes duplicate/unreachable published representations conservatively; it may skip an otherwise fitting cache selection. It requires an aggregate rather than a graph walk or repository-sized JS memory. A cacheable orphan producing regular output may finish; a stalled orphan is cancelled. Genuine mid-stream failure still fails all followers, now with a readable retry instruction; no transparent resume.
  • Default 32 processing/64 cached-reader slots remain. The uncached scratch test uses its existing low-cache profile with five arenas, allowing two held encoders plus publication while preserving co-tenant capacity; its workload/deadlines/assertions remain.
  • Force-push is not erasure: packs, historical ref IDs and push-log IDs remain. Destructive publication temporarily refuses non-tip raw Git wants until alarm-driven reachability catches up. Erasure requires complete delete and permanent namespace retirement; copies/backups need their own policy.
  • Production total memory/costs, deployed EU/R2 residency and the product's workers-rs build remain unverified. All verification is local. Optional archives/notifications, unreachable-object GC, LFS, SSH, hooks and a native push-capable host remain unimplemented.

Verification

Locked npm installation/build; full local host suite 270/270 in 407.362 s, zero failures/skips/cancellations, plus the strengthened three-blob admission focused rerun. Rust workspace tests, 96 Python tests, 39 isolated warning-denied feature builds, and public-package checks passed. The first full run exposed a test profile needing a third heavy lease; its explicit arena profile was corrected, preserving every assertion and timeout, then the whole suite passed. No Rust source or dependencies changed, and no lint/fuzz/check was weakened.

On head 65ae7457fd423da979365221fc8ea6a3f1ab9c75, all 27 applicable checks passed (both Rust/worker runs, all 16 interop shards, fuzz, benchmark, Runnerless title/commit checks, ci-toolkit and the Codex gate); lifecycle is intentionally skipped. The fifth requested Codex review returned “Didn't find any major issues.” Six VALID inline findings were reproduced, fixed with regression tests, and resolved; none were dismissed as speculative. The branch was not rebased because main stayed an ancestor. No deployment or release is part of this change.

Generated repositories/packs/archives were removed; only small ignored .scratch/ evidence logs remain. No other checkout or process arguments/environments were inspected. Historical timeouts remain the Wrangler/direct-workerd variants of “matches native tree-depth filters across shared paths and history”; old resets were Wrangler tree-depth and direct-workerd wanted-closure. The exact historical cause remains unproven; original tests pass unchanged.

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit e897132d81aafe66e7c827bd9b0ea826b11179db · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.005361 0.001339 -75.023317%
client-discovery-smart/max_rss_kib (bytes) 12177408.000000 2883584.000000 -76.320215%
client-discovery-smart/wall_s (s) 0.006968 0.003002 -56.908601%
server-clone-smart/cpu_s (s) 0.022641 0.024249 7.102160%
server-clone-smart/max_rss_kib (bytes) 12201984.000000 12144640.000000 -0.469956%
server-clone-smart/wall_s (s) 0.032492 0.022174 -31.754082%
server-discovery-smart/cpu_s (s) 0.005340 0.005180 -2.996255%
server-discovery-smart/max_rss_kib (bytes) 12161024.000000 12177408.000000 0.134725%
server-discovery-smart/wall_s (s) 0.006930 0.005335 -23.028085%
server-fetch-smart/cpu_s (s) 0.007390 0.007387 -0.040595%
server-fetch-smart/max_rss_kib (bytes) 12156928.000000 12144640.000000 -0.101078%
server-fetch-smart/wall_s (s) 0.008936 0.007453 -16.595771%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit ce59344f00e41dde12c640d971d5d13b9c71fc81 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.009460 0.002334 -75.327696%
client-discovery-smart/max_rss_kib (bytes) 12177408.000000 2818048.000000 -76.858392%
client-discovery-smart/wall_s (s) 0.011999 0.005068 -57.763290%
server-clone-smart/cpu_s (s) 0.036719 0.040557 10.452354%
server-clone-smart/max_rss_kib (bytes) 12185600.000000 12189696.000000 0.033613%
server-clone-smart/wall_s (s) 0.056116 0.038914 -30.654721%
server-discovery-smart/cpu_s (s) 0.009378 0.009193 -1.972702%
server-discovery-smart/max_rss_kib (bytes) 12144640.000000 12181504.000000 0.303541%
server-discovery-smart/wall_s (s) 0.011866 0.009345 -21.242483%
server-fetch-smart/cpu_s (s) 0.013567 0.013891 2.388148%
server-fetch-smart/max_rss_kib (bytes) 12161024.000000 12201984.000000 0.336814%
server-fetch-smart/wall_s (s) 0.016272 0.014020 -13.841186%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@pawelchcki pawelchcki changed the title feat(hosting): support scoped gateway grants feat(hosting): add scoped grants and repository lifecycle Nov 15, 2026
@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 01b13c70efcd687f2f8088ff587a2b8737c1cd40 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.006510 0.001519 -76.666667%
client-discovery-smart/max_rss_kib (bytes) 12189696.000000 2834432.000000 -76.747312%
client-discovery-smart/wall_s (s) 0.008255 0.003389 -58.947210%
server-clone-smart/cpu_s (s) 0.024502 0.026406 7.770794%
server-clone-smart/max_rss_kib (bytes) 12177408.000000 12185600.000000 0.067272%
server-clone-smart/wall_s (s) 0.035738 0.024691 -30.912757%
server-discovery-smart/cpu_s (s) 0.006041 0.005947 -1.556034%
server-discovery-smart/max_rss_kib (bytes) 12140544.000000 12148736.000000 0.067476%
server-discovery-smart/wall_s (s) 0.007829 0.006104 -22.039064%
server-fetch-smart/cpu_s (s) 0.008804 0.008863 0.670150%
server-fetch-smart/max_rss_kib (bytes) 12189696.000000 12181504.000000 -0.067204%
server-fetch-smart/wall_s (s) 0.010536 0.008967 -14.890268%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@pawelchcki pawelchcki changed the title feat(hosting): add scoped grants and repository lifecycle feat(hosting): add scoped gateway grants and product read APIs Nov 15, 2026
@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit a0e8c1aad5d16ac06a817767ddd4673979de726d · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.008735 0.002157 -75.306239%
client-discovery-smart/max_rss_kib (bytes) 12169216.000000 2809856.000000 -76.910131%
client-discovery-smart/wall_s (s) 0.011117 0.004671 -57.981632%
server-clone-smart/cpu_s (s) 0.034429 0.038598 12.108978%
server-clone-smart/max_rss_kib (bytes) 12165120.000000 12197888.000000 0.269360%
server-clone-smart/wall_s (s) 0.059686 0.042252 -29.210451%
server-discovery-smart/cpu_s (s) 0.008867 0.008886 0.214278%
server-discovery-smart/max_rss_kib (bytes) 12165120.000000 12169216.000000 0.033670%
server-discovery-smart/wall_s (s) 0.011328 0.009098 -19.683881%
server-fetch-smart/cpu_s (s) 0.012017 0.012001 -0.133145%
server-fetch-smart/max_rss_kib (bytes) 12161024.000000 12148736.000000 -0.101044%
server-fetch-smart/wall_s (s) 0.014301 0.012229 -14.483991%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@pawelchcki pawelchcki changed the title feat(hosting): add scoped gateway grants and product read APIs feat(hosting): add scoped repository APIs and push records Nov 15, 2026
@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 0204f708570279a637db405dfc0893237f0461f0 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.005590 0.001383 -75.259392%
client-discovery-smart/max_rss_kib (bytes) 12124160.000000 2850816.000000 -76.486486%
client-discovery-smart/wall_s (s) 0.007211 0.003089 -57.169458%
server-clone-smart/cpu_s (s) 0.022854 0.024382 6.685919%
server-clone-smart/max_rss_kib (bytes) 12156928.000000 12206080.000000 0.404313%
server-clone-smart/wall_s (s) 0.033086 0.022535 -31.889022%
server-discovery-smart/cpu_s (s) 0.005342 0.005270 -1.347810%
server-discovery-smart/max_rss_kib (bytes) 12144640.000000 12165120.000000 0.168634%
server-discovery-smart/wall_s (s) 0.007008 0.005366 -23.436929%
server-fetch-smart/cpu_s (s) 0.007626 0.007616 -0.131130%
server-fetch-smart/max_rss_kib (bytes) 12132352.000000 12140544.000000 0.067522%
server-fetch-smart/wall_s (s) 0.009208 0.007715 -16.208893%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@pawelchcki pawelchcki changed the title feat(hosting): add scoped repository APIs and push records feat(hosting): add scoped repository APIs and deployment policy Nov 15, 2026
@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit c434b802558abf9a82639d5b367787191007177e · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.008360 0.002102 -74.856459%
client-discovery-smart/max_rss_kib (bytes) 12103680.000000 2629632.000000 -78.274112%
client-discovery-smart/wall_s (s) 0.010743 0.004594 -57.235273%
server-clone-smart/cpu_s (s) 0.039370 0.042584 8.163576%
server-clone-smart/max_rss_kib (bytes) 12136448.000000 12120064.000000 -0.134998%
server-clone-smart/wall_s (s) 0.059498 0.040222 -32.398023%
server-discovery-smart/cpu_s (s) 0.008711 0.008810 1.136494%
server-discovery-smart/max_rss_kib (bytes) 12099584.000000 12103680.000000 0.033852%
server-discovery-smart/wall_s (s) 0.011138 0.008965 -19.506316%
server-fetch-smart/cpu_s (s) 0.012951 0.013022 0.548220%
server-fetch-smart/max_rss_kib (bytes) 12107776.000000 12136448.000000 0.236806%
server-fetch-smart/wall_s (s) 0.015372 0.013083 -14.889125%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@pawelchcki pawelchcki changed the title feat(hosting): add scoped repository APIs and deployment policy feat(hosting): make the repository host embeddable Nov 15, 2026
@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit d4471e69d256b0438c3544383156e503a60c8bf7 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.008112 0.002100 -74.112426%
client-discovery-smart/max_rss_kib (bytes) 12161024.000000 2805760.000000 -76.928259%
client-discovery-smart/wall_s (s) 0.010437 0.004596 -55.967429%
server-clone-smart/cpu_s (s) 0.033957 0.036677 8.010130%
server-clone-smart/max_rss_kib (bytes) 12165120.000000 12222464.000000 0.471380%
server-clone-smart/wall_s (s) 0.051814 0.036479 -29.597469%
server-discovery-smart/cpu_s (s) 0.008001 0.007908 -1.162355%
server-discovery-smart/max_rss_kib (bytes) 12115968.000000 12173312.000000 0.473293%
server-discovery-smart/wall_s (s) 0.010410 0.008120 -21.998950%
server-fetch-smart/cpu_s (s) 0.010653 0.010687 0.319159%
server-fetch-smart/max_rss_kib (bytes) 12140544.000000 12161024.000000 0.168691%
server-fetch-smart/wall_s (s) 0.012778 0.010833 -15.222155%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 3d2f1fbf6e1ce4c872e1c14faf69816f53e52a47 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.005561 0.001379 -75.202302%
client-discovery-smart/max_rss_kib (bytes) 12152832.000000 2834432.000000 -76.676778%
client-discovery-smart/wall_s (s) 0.007166 0.003074 -57.097311%
server-clone-smart/cpu_s (s) 0.022575 0.024706 9.439646%
server-clone-smart/max_rss_kib (bytes) 12148736.000000 12189696.000000 0.337154%
server-clone-smart/wall_s (s) 0.032893 0.022642 -31.164300%
server-discovery-smart/cpu_s (s) 0.005365 0.005200 -3.075489%
server-discovery-smart/max_rss_kib (bytes) 12165120.000000 12173312.000000 0.067340%
server-discovery-smart/wall_s (s) 0.006944 0.005323 -23.351397%
server-fetch-smart/cpu_s (s) 0.007490 0.007500 0.133511%
server-fetch-smart/max_rss_kib (bytes) 12132352.000000 12132352.000000 0.000000%
server-fetch-smart/wall_s (s) 0.009040 0.007561 -16.357275%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit cf3655e93bc76b4a97ad8ae780e22b95ce4c56fa · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.008511 0.002065 -75.737281%
client-discovery-smart/max_rss_kib (bytes) 12169216.000000 2801664.000000 -76.977449%
client-discovery-smart/wall_s (s) 0.010761 0.004533 -57.875783%
server-clone-smart/cpu_s (s) 0.033976 0.038363 12.912056%
server-clone-smart/max_rss_kib (bytes) 12206080.000000 12197888.000000 -0.067114%
server-clone-smart/wall_s (s) 0.130822 0.156372 19.530818%
server-discovery-smart/cpu_s (s) 0.008614 0.008555 -0.684932%
server-discovery-smart/max_rss_kib (bytes) 12193792.000000 12185600.000000 -0.067182%
server-discovery-smart/wall_s (s) 0.010901 0.008723 -19.975805%
server-fetch-smart/cpu_s (s) 0.011731 0.011759 0.238684%
server-fetch-smart/max_rss_kib (bytes) 12124160.000000 12173312.000000 0.405405%
server-fetch-smart/wall_s (s) 0.014065 0.011906 -15.349518%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit cd713bd5eaddd71b54ef3773027b78bbf03f1b55 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.006585 0.001578 -76.036446%
client-discovery-smart/max_rss_kib (bytes) 12189696.000000 2842624.000000 -76.680108%
client-discovery-smart/wall_s (s) 0.008352 0.003415 -59.115490%
server-clone-smart/cpu_s (s) 0.024497 0.026362 7.613177%
server-clone-smart/max_rss_kib (bytes) 12152832.000000 12115968.000000 -0.303337%
server-clone-smart/wall_s (s) 0.035577 0.024522 -31.074019%
server-discovery-smart/cpu_s (s) 0.006320 0.006073 -3.908228%
server-discovery-smart/max_rss_kib (bytes) 12169216.000000 12181504.000000 0.100976%
server-discovery-smart/wall_s (s) 0.008158 0.006263 -23.233878%
server-fetch-smart/cpu_s (s) 0.008944 0.008856 -0.983900%
server-fetch-smart/max_rss_kib (bytes) 12148736.000000 12185600.000000 0.303439%
server-fetch-smart/wall_s (s) 0.010646 0.008930 -16.112295%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

Copy link
Copy Markdown
Contributor Author

@codex review

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 16, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-16T10:20:24.338439Z 65ae745 Manual request
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 chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cd713bd5ea

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".

Serialize encoded chunk claims and PUTs against failed-stage deletion while keeping unrelated keys concurrent. Forward job-view recovery deadlines to the shared store, and budget signed proofs plus cursor envelopes before selecting tree pages.

Five regressions reproduce the three Codex findings on the previous head. Locked installation, 19 focused tests and the full 265-test host suite pass.

Copy link
Copy Markdown
Contributor Author

@codex review

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 933c5ce8b48b4b6e0f52af04db2bb219fbbcf0c1 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.007807 0.001974 -74.714999%
client-discovery-smart/max_rss_kib (bytes) 12120064.000000 2834432.000000 -76.613721%
client-discovery-smart/wall_s (s) 0.010126 0.004407 -56.483164%
server-clone-smart/cpu_s (s) 0.032282 0.035674 10.507404%
server-clone-smart/max_rss_kib (bytes) 12156928.000000 12206080.000000 0.404313%
server-clone-smart/wall_s (s) 0.049963 0.036062 -27.823664%
server-discovery-smart/cpu_s (s) 0.007838 0.007699 -1.773412%
server-discovery-smart/max_rss_kib (bytes) 12124160.000000 12181504.000000 0.472973%
server-discovery-smart/wall_s (s) 0.010083 0.007890 -21.750236%
server-fetch-smart/cpu_s (s) 0.010436 0.010580 1.379839%
server-fetch-smart/max_rss_kib (bytes) 12218368.000000 12165120.000000 -0.435803%
server-fetch-smart/wall_s (s) 0.012470 0.010738 -13.892495%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 933c5ce8b4

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".

Reserve the serialized history envelope and an actual signed cursor before selecting commit records. Add a real-Git SHA-256 long-namespace regression that reproduces the former 413 and verifies bounded, complete pagination.

The eight focused browse/regression tests and full 266-test host suite pass. Admission and deployment caps remain unchanged.

Copy link
Copy Markdown
Contributor Author

@codex review

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 606b8dd7058406eeaa6707778c9991a70c2b130b · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.006226 0.001459 -76.566013%
client-discovery-smart/max_rss_kib (bytes) 12181504.000000 2842624.000000 -76.664425%
client-discovery-smart/wall_s (s) 0.007959 0.003259 -59.047203%
server-clone-smart/cpu_s (s) 0.023404 0.025905 10.686207%
server-clone-smart/max_rss_kib (bytes) 12173312.000000 12185600.000000 0.100942%
server-clone-smart/wall_s (s) 0.034464 0.023869 -30.741946%
server-discovery-smart/cpu_s (s) 0.005659 0.005442 -3.834600%
server-discovery-smart/max_rss_kib (bytes) 12201984.000000 12169216.000000 -0.268546%
server-discovery-smart/wall_s (s) 0.007349 0.005591 -23.929172%
server-fetch-smart/cpu_s (s) 0.008190 0.008092 -1.196581%
server-fetch-smart/max_rss_kib (bytes) 12181504.000000 12136448.000000 -0.369872%
server-fetch-smart/wall_s (s) 0.009739 0.008169 -16.114789%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 606b8dd705

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".

Keep a shared scheduling revision so an alarm cannot erase a retry scheduled by a protected job while recovery awaits R2. Cover deferred encoded and temporary cleanup, and initialize the constructor-bypassing cleanup fixture.

Verified both regressions fail on the reviewed head, 20 focused tests pass, and all 268 host tests pass without skips.

Copy link
Copy Markdown
Contributor Author

@codex review

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 85837c6150c320d04f95cecdc84ffb8e4a0a4e22 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.007493 0.001791 -76.097691%
client-discovery-smart/max_rss_kib (bytes) 12197888.000000 2842624.000000 -76.695769%
client-discovery-smart/wall_s (s) 0.009528 0.004002 -57.990960%
server-clone-smart/cpu_s (s) 0.031643 0.034199 8.077616%
server-clone-smart/max_rss_kib (bytes) 12177408.000000 12222464.000000 0.369997%
server-clone-smart/wall_s (s) 0.049571 0.033511 -32.397450%
server-discovery-smart/cpu_s (s) 0.007348 0.007254 -1.279260%
server-discovery-smart/max_rss_kib (bytes) 12177408.000000 12189696.000000 0.100908%
server-discovery-smart/wall_s (s) 0.009420 0.007381 -21.647441%
server-fetch-smart/cpu_s (s) 0.010727 0.010767 0.372891%
server-fetch-smart/max_rss_kib (bytes) 12161024.000000 12193792.000000 0.269451%
server-fetch-smart/wall_s (s) 0.012735 0.010808 -15.132878%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 85837c6150

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".

Keep the Store private until the single request/alarm initialization promise finishes orphan recovery. Otherwise a second initializer can admit live stages that the first pass removes.

Both arrival-order regressions fail on the reviewed head. All 22 focused tests and 270 host tests pass without skips; accepted admission and caps are unchanged.

Copy link
Copy Markdown
Contributor Author

@codex review

@github-actions

Copy link
Copy Markdown

Git versus git-zero

Commit 65ae7457fd423da979365221fc8ea6a3f1ab9c75 · Runnerless report · SVG

Workload / metric Git git-zero Change
client-discovery-smart/cpu_s (s) 0.008815 0.002118 -75.972774%
client-discovery-smart/max_rss_kib (bytes) 12095488.000000 2711552.000000 -77.582120%
client-discovery-smart/wall_s (s) 0.011300 0.004740 -58.058315%
server-clone-smart/cpu_s (s) 0.040727 0.044604 9.519483%
server-clone-smart/max_rss_kib (bytes) 12132352.000000 12120064.000000 -0.101283%
server-clone-smart/wall_s (s) 0.061965 0.041904 -32.375281%
server-discovery-smart/cpu_s (s) 0.008573 0.008539 -0.396594%
server-discovery-smart/max_rss_kib (bytes) 12107776.000000 12103680.000000 -0.033829%
server-discovery-smart/wall_s (s) 0.011019 0.008643 -21.561865%
server-fetch-smart/cpu_s (s) 0.012805 0.012781 -0.187427%
server-fetch-smart/max_rss_kib (bytes) 12066816.000000 12111872.000000 0.373388%
server-fetch-smart/wall_s (s) 0.015209 0.012796 -15.867589%

Matched stored-DEFLATE/no-delta policies; persistent warm servers and OS caches. Wall time, CPU and peak RSS describe client commands, excluding servers. RSS is the largest waited process peak, not aggregate server memory. Deterministic alternating order and shared servers keep these measurements descriptive; no significance or causal claim. All original measurements are retained by Runnerless.

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. Chef's kiss.

Reviewed commit: 65ae7457fd

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".

@pawelchcki
pawelchcki marked this pull request as ready for review November 16, 2026 10:40
@pawelchcki
pawelchcki merged commit 0f98b35 into main Nov 16, 2026
28 checks passed
@pawelchcki
pawelchcki deleted the feat/embeddable-repository-host branch November 16, 2026 10:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant