Skip to content

docs(docs): Operators can see that Fargate x86-64 is qualified for the filter sandbox - #47

Merged
sodre merged 3 commits into
mainfrom
t3/implement-foresight-25
Nov 21, 2026
Merged

sodre merged 3 commits into
mainfrom
t3/implement-foresight-25

Conversation

@sodre

@sodre sodre commented Nov 21, 2026

Copy link
Copy Markdown
Member

Closes #25

Summary

The handbook's Tested platforms page now says which Gateway hosts the filter sandbox has been qualified on: ECS Fargate 1.4.0 on x86-64 is qualified under worker_sandbox: "filter", and Fargate arm64 has passed the maintainers' harness but has had no end-to-end run yet.

 # Tested platforms
   intro: names fbb5aa7 "unless a section names its own revisions",
+         and points Gateway operators to their section
 ## Native identity and desktop packages
 ## End-to-end and deployment evidence
+## Gateway hosts and the worker sandbox
+  | ECS Fargate 1.4.0 | x86-64 | Qualified    | reporter's run on #18 (v0.2.1, 7199c78) · harness run on #24 (30bedc2) |
+  | ECS Fargate 1.4.0 | arm64  | Harness only | harness run on #24 (30bedc2); no end-to-end run                        |
+  link to the harness (qualification/fargate at 7199c78, same files as 30bedc2)
 unqualified list
+  … and an end-to-end run of the filter sandbox on Fargate arm64

Each run is cited with the revision it ran, as the triage brief asks, because the page's opening line names a single revision (fbb5aa7) that these runs don't share. The operator guide is unchanged.

Choices the ticket left open:

  • The new section covers the filter sandbox only. The brief puts a namespace-sandbox host matrix out of scope.
  • The harness link points at 7199c78 on main, not 30bedc2, which is a PR-branch commit. qualification/fargate is identical at both, so the link opens the harness that ran.

After merge, close #18 with a comment linking the v0.2.1 release, the first release that carries the filter sandbox.

Evidence

Criterion Ticked Shown by
The reporter's end-to-end run on Fargate is recorded on #18 (platform version, architecture, Gateway version, worker_sandbox: "filter", outcome) yes gh api repos/zeroae/countersign/issues/comments/7066219942
The tested-platforms page lists ECS Fargate 1.4.0 under the filter sandbox, for the architectures qualified, citing that run and the harness run from the previous ticket yes mkdocs build --strict and the row check on the built page
#18 is closed with a link to the release that carries the filter sandbox no Not run: the owner closes #18 by hand after the merge
Test Red Green
mkdocs build --strict resolves the new section's link into the operator guide Link changed to ../operate.md#choose-the-worker-sandbox: WARNING - Doc file 'reference/support.md' contains a link '../operate.md#choose-the-worker-sandbox', but the doc 'operate.md' does not contain an anchor '#choose-the-worker-sandbox'. Aborted with 1 warnings in strict mode!, exit 1 ok
mkdocs build --strict resolves the intro's link to the new section Link changed to #gateway-hosts-and-the-worker-sandboxx: WARNING - Doc file 'reference/support.md' contains a link '#gateway-hosts-and-the-worker-sandboxx', but there is no such anchor on this page. Aborted with 1 warnings in strict mode!, exit 1 ok

python -m mkdocs build --strict (from requirements-docs.txt in a fresh venv) on d599672: exit 0, Documentation built in 0.27 seconds.

Row check on the built page (site/reference/support/index.html, its table rows that start with "ECS Fargate", and the links in each):

ECS Fargate 1.4.0 | x86-64 | Qualified | …/issues/18#issuecomment-7066219942 …/issues/24#issuecomment-7032499034
ECS Fargate 1.4.0 | arm64 | Harness only | …/issues/24#issuecomment-7032499034
unqualified list has arm64 e2e: True

gh api repos/zeroae/countersign/issues/comments/7066219942, the reporter's run on #18: "Reporter qualification passed on AWS ECS Fargate 1.4.0, Linux x86-64, on November 21, 2026. The shared Gateway used worker_sandbox: "filter"." It links release v0.2.1 and names the release commit 7199c789e2e5….

git diff --stat 30bedc2 7199c78 -- qualification/fargate: empty.

Not run: closing #18 with a link to v0.2.1, which the owner does by hand after the merge. No Rust, Node or runtime suite ran, because the change touches only docs/reference/support.md.

Merge Danger

Door: two-way. A revert restores the page as it was. Nothing else depends on it.

Blast Radius: one handbook page.

The merge changes no image, package or infrastructure. The next version tag publishes the handbook snapshot with this page. No flag.

Reviews

Correctness, Standards, Spec and Security, two rounds: 11 findings, each with its outcome

Correctness, round 1 (pinned a5ce5addad26ebb312af3f34c9757ef484601fa7, against 7199c789e2e514cb6201ba84eed0071d9a0b50e0)

  1. The harness link points at 30bedc2, a branch commit that could be garbage-collected. Fixed: the link now points at 7199c78 on main. qualification/fargate is identical there, and refs/pull/36/head keeps 30bedc2 in any case.
  2. The Qualified row combines a harness run at 30bedc2 and an end-to-end run at 7199c78, and no harness run on v0.2.1 is shown. Pushed back: the ticket's brief asks for each run to be cited with the revision it exercised, and the spec makes the harness run gate the reporter's run, not repeat it on the release. The row names both revisions.
  3. "Maintainers rerun it for every release that touches the grant worker" is a promise, not evidence. Fixed: the sentence is gone.
  4. The intro frames the page for desktop fleets only. Fixed: the intro now points Gateway operators to the new section.
  5. "Production shared infrastructure" stays unqualified while a shared Gateway on Fargate is Qualified. Pushed back: the new row qualifies the host for the filter sandbox, under its own column. The existing line is about production deployments, and the ticket leaves it alone.
  6. The row leaves out that the chat front end ran in Compose, outside Fargate. Pushed back: the row says exactly what ran in the Fargate task (the Gateway, the Broker and the mTLS site). The page qualifies the Gateway host, not where the Integrator runs.

Standards, round 1

  1. The page spells processors x64/ARM64, and the new rows use x86-64/arm64. Pushed back: the Gateway docs, ADR 0001 and the harness use x86-64/arm64. The desktop mentioned in the row keeps the page's desktop spelling.
  2. "Chat replies" and "the site's flag" are never explained. Other rows state the check that was made. Fixed: the row now says that the mTLS site answered each grant's request with its protected flag, and that the demo Integrator's chat showed it in both replies.
  3. The harness sentence is repeated in both rows. Pushed back: each table row stands on its own evidence, as the page's other tables do, and the paragraph below describes the harness, not its results.

Spec, round 1

  1. Same as 3: the rerun sentence is a policy that neither cited run shows. Fixed: the sentence is gone.

Security, round 1: no findings (the change is documentation only).

Correctness, round 2 (pinned d5996722524c72c61dde23cdb73ebd94fe6aa83b, against a5ce5addad26ebb312af3f34c9757ef484601fa7)

  1. The harness link opens 7199c78, but the cells say the harness ran at 30bedc2. Pushed back: git diff 30bedc2 7199c78 -- qualification/fargate is empty, so the link shows the harness that ran, and the cells still name the revision the run used.

Standards, Spec and Security, round 2: no findings.

Tracking

Epic: E1 Run grant workers in a filter sandbox on hosts without namespaces
Spec: #21 Run grant workers in a filter sandbox on hosts without namespaces
Ticket: #25 List Fargate as a tested platform
Flag: none

🤖 Generated with Claude Code

sodre and others added 3 commits November 21, 2026 15:41
…the filter sandbox

The tested-platforms page gains a section on Gateway hosts. ECS Fargate
1.4.0 x86-64 is qualified under worker_sandbox "filter", citing the
reporter's end-to-end run on #18 and the harness run on #24; arm64 is
harness-tested only, and its end-to-end run joins the unqualified list.

Refs #25

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…eck the run made

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@sodre
sodre marked this pull request as ready for review November 21, 2026 23:48
@sodre
sodre merged commit 59cfd01 into main Nov 21, 2026
38 checks passed
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.

List Fargate as a tested platform Gateway v0.2.0 cannot start a grant worker on ECS Fargate 1.4.0

1 participant