Skip to content

Pack small Cell publications and verify durability performance - #66

Merged
forhappy merged 8 commits into
mainfrom
codex/packed-write-publication
Oct 7, 2026
Merged

forhappy merged 8 commits into
mainfrom
codex/packed-write-publication

Conversation

@forhappy

@forhappy forhappy commented Oct 7, 2026 •

Copy link
Copy Markdown
Contributor

Small Cell publication required separate native-body, page-index and directory uploads before selecting the root. This change packs small native LTX/index dependencies and inlines bounded directory metadata, reducing ordinary publication from six successful PUTs to four while preserving fenced Cell selection and byte-identical recovery. Composed compaction plus append still requires five PUTs.

The pinned Docker SQL application comparison adds windowed telemetry, exact command retry and all-acknowledgment warm/cold audits, bounded JSONL journals, source-content build caches and provider lifecycle/capacity evidence. Raw performance material stays outside the repository. See docs/write-performance-delivery.md and the proposal's unchanged gates.

Qualification remains incomplete. The final matched 2,000/s bucket stress completed 163.03/s with 8,906.2-ms scheduled p99, versus baseline 104.71/s and celld 905.08/s. All arms dropped offers and were unqualified; celld's cold audit also failed through provider OOM. Cellule acknowledgment/retry audits passed, but publication debt remained unstable. These overload completion rates do not establish sustainable capacity or celld parity.

Shared publication, signed append grants and production bundle coverage proofs are subsequent work and are not part of this merged PR.

Add bounded packed LTX/index objects and inline directory leaves across publication, restore, sparse reads, compaction, backup and collection. Share one verified packed compaction fetch between body and index spools.

Add response/proof/worker/peer/storage telemetry and pinned Docker workload, warm/cold exact-retry audits, failure-preserving reports and explicit delivery gaps. Keep raw evidence outside Git.
Keep late arrivals in the original offered cohort, support immediate overload recovery, and stream every successful command through bounded collection and GET/retry audits. Require client, fixture, runner and host provenance for new comparisons. Retain failed read saturation results and regression evidence.
An empty object coverage queue requires no new writer CAS. Preserve pending-ticket fencing, contiguous rotation, member receipts and authority closure. The fenced-owner fleet test now drains in 1.3 seconds instead of waiting forever.

Record provider byte and inode headroom throughout Docker runs. Retained volumes can exhaust inodes despite free bytes; refuse such startup and invalidate deficient measurements.
Record the final matched bucket point, all-ACK cold recovery, and remaining M0–M5 gaps. Capture provider state through cold audits and preserve OOM or missing evidence as qualification failures.
@forhappy
forhappy marked this pull request as ready for review October 7, 2026 06:37
@forhappy
forhappy merged commit 831877c into main Oct 7, 2026
34 checks passed
@forhappy forhappy changed the title Pack small Cell publications and verify durability performance Share Cell publication and use signed follower append grants Oct 7, 2026
@forhappy forhappy changed the title Share Cell publication and use signed follower append grants Pack small Cell publications and verify durability performance Oct 7, 2026
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