Skip to content

Name the versions the storage page was tested on - #70

Merged
kisielewski merged 1 commit into
mainfrom
docs/the-storage-page-names-what-it-was-tested-on
Sep 20, 2026
Merged

kisielewski merged 1 commit into
mainfrom
docs/the-storage-page-names-what-it-was-tested-on

Conversation

@kisielewski

Copy link
Copy Markdown
Member

/install/storage names the two S3 implementations the project is tested against and gave no version for either, so a reader could not tell whether exercised here meant a release candidate or a stable build. It now says RustFS 1.0.0 and SeaweedFS 4.47, in both languages, matching the pins in AlgoJudge/AlgoJudge-Server#125.

The paragraph below it was wrong in three separate ways, and had been since 2026-09-07:

  • it described an item that writes a known string and looks for it in the store's own files. That check was deleted: measured against SeaweedFS, grep found a 44-character prefix of a 48-character needle that od showed lying complete and contiguous on disk — so it answered absent about bytes that were present, and it failed in the direction that reads as encrypted.
  • it said SeaweedFS refuses the encryption call. It does not. It accepts the call and then fails every write to that bucket, which is worse, because an operator who applies a default encryption rule to a SeaweedFS bucket gets a store that agreed and stopped working — and a write made just before the rule succeeds, so nothing warns them at the moment they set it.
  • it said the item cannot be run against either implementation available here. It runs on RustFS, on the suite's own default endpoint.

All three are replaced by what was measured on 2026-09-20 against RustFS 1.0.0 and SeaweedFS 4.47.

The /v0.1/ snapshots are deliberately untouched. They are the frozen record of what 0.1 documented, and an operator still running 0.1 should read what shipped rather than what is true of the next version.

The Polish wants a human pass before it is trusted. It was written beside the English in one sitting, which is exactly how a calque gets in.

Checks

npm run build produces 524 pages, and check:links, check:versions, check:translations, check:structure, check:glossary, check:section-links and check:no-playground all pass. Editing the English invalidated the Polish page's sourceSha256; it was rewritten with check:translations -- --update after re-reading the Polish against the English, which is the only thing that should move a fingerprint. Both rendered pages were read out of out/ to confirm the new versions appear and no sentence of the old paragraph survives anywhere in the export.

RustFS 1.0.0 and SeaweedFS 4.47, in both languages.

The paragraph about bucket-default encryption described a check that was
deleted on 2026-09-07 and said SeaweedFS refuses the call. It does not: it
accepts the call and then fails every write to that bucket, which is worth
knowing before an operator sets the rule.
@kisielewski
kisielewski merged commit fed583e into main Sep 20, 2026
5 checks passed
@kisielewski
kisielewski deleted the docs/the-storage-page-names-what-it-was-tested-on branch September 20, 2026 17:10
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