Repository navigation
Conversation
The CacheRuntimeClass and the read jobs are written with an unquoted
`cat<<EOF`, so the local shell expands `${CURVINE_HOME}`,
`$FLUID_DATALOAD_DATA_PATH` and `$(cat /data/minio/bar)` before the
manifest is written, and the resulting scripts are broken. Quote the
delimiter for those three heredocs.
The image tag `curvine/curvine:0.3.6-alpha` does not exist on Docker Hub.
Use `v0.3.6-alpha`, the tag the e2e test already uses.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Signed-off-by: btxu-db <btxu-db@outlook.com>
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #6252 +/- ##
=======================================
Coverage 65.81% 65.81%
=======================================
Files 486 486
Lines 34454 34454
=======================================
Hits 22677 22677
Misses 10024 10024
Partials 1753 1753 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues are recorded for the documentation updates to heredoc quoting and Curvine image tags.
No open issues 🎉
What changed in this PR
Fixes executable examples in the English and Chinese Curvine CacheRuntime guides so generated manifests preserve container-side shell expressions and use the correct image tag.
Changes:
- Quotes the CacheRuntimeClass and both read-job heredoc delimiters to prevent local shell expansion.
- Updates master, worker, and client images to
curvine/curvine:v0.3.6-alpha, matching the integration fixtures. - Leaves the bucket-creation heredoc unchanged because it already escapes its variables.
| File | Description |
|---|---|
| docs/zh/samples/cacheruntime/curvine_cache_runtime.md | Corrects heredoc quoting and image tags in the Chinese guide. |
| docs/en/samples/cacheruntime/curvine_cache_runtime.md | Applies the same corrections to the English guide. |
Review effort: Balanced
…nce (root .gitignore excludes *.out) Evidence files referenced by the README/manifest were silently skipped by the repo-wide *.out ignore rule. Co-Authored-By: Claude Code <noreply@anthropic.com>
cheyang
left a comment
There was a problem hiding this comment.
Both claims of this docs fix verified end-to-end, independently. Reproduced on the merge-base first: the unquoted heredocs really let the local shell eat the pod-side variables — CURVINE_CONF_DIR comes out as "/conf", the read jobs degenerate to test -n "" after a local cat /data/minio/bar — and curvine/curvine:0.3.6-alpha is a 404 on Docker Hub. At the PR head, every heredoc block in both languages regenerates its intended file verbatim, every generated YAML parses, v0.3.6-alpha resolves as a multi-arch (amd64/arm64) index, and the doc-generated cacheruntimeclass.yaml is byte-identical to the manifest kind-e2e actually deploys. The one heredoc left unquoted (step 2's minio bucket job) is correct as-is: its \$ escaping is load-bearing, and quoting it would leave a literal $MINIO_ROOT_USER that fails YAML parsing — mutation-tested in both directions. Only the three intended generated files change; the other seven are byte-identical, and en/zh stay in lockstep. Required CI is green (13/13 checks). No blockers or majors from either reviewer, no disagreement survived cross-critique, and Copilot's advisory pass raised no findings.
Next round
[MUST] (none — no PR-introduced blockers; merge checklist green)
[SHOULD] Follow-up PR: fix step 6's kubectl get pod -l cacheruntime.fluid.io/runtime-name=... in both language docs — the code only ever sets cacheruntime.fluid.io/name/component-name, so that command shows no pods (pre-existing).
[SHOULD] Consider wiring a doc-samples contract check (execute the heredocs, parse the YAML) into CI — the tag drift survived precisely because nothing runs the samples (pre-existing gap; both verify branches are working prototypes).
[Filed] (none)
Non-blocking observations (3) — not part of this round's ask; fix now, follow up, or file an issue as you see fit
- docs/en/samples/cacheruntime/curvine_cache_runtime.md:393 [minor] — pre-existing, not introduced by this PR — Step-6 pod selector uses a label the code never sets
F1 (Claude, confirmed by Codex in cross-critique):kubectl get pod -l cacheruntime.fluid.io/runtime-name=curvine-demomatches no pods. pkg/common/label.go:154,156 defines only cacheruntime.fluid.io/name and cacheruntime.fluid.io/component-name (set on component pods via GetComponentMatchLabels in pkg/ddc/cache/component/component_manager.go:78), and test/gha-e2e/curvine/test.sh selects pods only with those keys; repo-wide grep finds the runtime-name form only in these two doc files, so no counter-path exists. The zh doc carries the identical command at the identical line. Pre-existing on base (identical command at line 393 of ff99142) and outside this PR's declared scope — recommend a one-line follow-up (-l cacheruntime.fluid.io/name=curvine-demo), not a change to this PR. - [minor] — pre-existing, not introduced by this PR — Sample docs are never executed in CI, so doc/manifest drift goes unnoticed
F2 (Codex, confirmed by Claude in cross-critique): no workflow parses, executes, or lints docs/{en,zh}/samples/** — the only curvine coverage in CI is test/gha-e2e/curvine/ via kind-e2e, which is a separate copy of the manifests, not the doc. This drift survived because of it: the doc pinned curvine/curvine:0.3.6-alpha (404 on Docker Hub) while test/gha-e2e/curvine/cacheruntimeclass.yaml:58,84,108 has carried v0.3.6-alpha all along, and three heredocs were broken from the day they were written. Not a merge condition for this docs fix — docs PRs do not owe test infrastructure — but a cheap guard exists: both reviewer harnesses (e.g. docs/verification/curvine-cache-sample-doc-codex/scripts/run_doc_contracts.sh) execute every heredoc in the sample and YAML-parse the output; wiring that style of check into CI would catch the next drift and the step-6 label-selector class of bug. - docs/en/samples/cacheruntime/curvine_cache_runtime.md:173 [nit] — nit: mixed cat<<EOF / cat<<'EOF' style across the sample's ten heredocs
F1 (Codex, confirmed by Claude in cross-critique): the doc now mixes unquoted (7 blocks) and quoted (3 blocks) heredoc openers, introduced by this PR quoting exactly the three blocks that needed it. Both forms are correct where used — six unquoted blocks contain no $ at all, and step 2's bucket job MUST stay unquoted (its $ escapes are collapsed by the local shell for the pod; quoting it leaves literal$MINIO_ROOT_USER, an invalid YAML double-quote escape — proven by mutation in both reviewers' rounds, results/harness-bites-mutation.txt). Optionally quoting the six $ -free blocks too would harden the doc against future edits that add $; keeping quotes only where needed instead documents intent. Taste either way — both reviewers agree: do not hold the PR on this.
Verified review delta
- curvine-doc-step6-label: open (preexisting)
- mt-sample-docs-not-executed-in-ci: open (preexisting)
- nit-heredoc-quoting-uniformity: open (initial-pr)
Code: clear. CI: success; attribution: unknown. Required CI checks still apply.



Ⅰ. Describe what this PR does
The Curvine sample does not work when followed as written, in either language:
cat<<EOF, so the local shell expands${CURVINE_HOME},$FLUID_DATALOAD_DATA_PATHand$(cat /data/minio/bar)before the file is written.${CURVINE_HOME}/confends up as/conf, for example. These three heredocs now usecat<<'EOF'.curvine/curvine:0.3.6-alphadoes not exist on Docker Hub. The tag isv0.3.6-alpha, which is whattest/gha-e2e/curvine/uses.Ⅱ. Does this pull request fix one issue?
NONE
Ⅲ. List the added test cases (unit test/integration test) if any, please explain if no tests are needed.
No tests; docs only.
Ⅳ. Describe how to verify it
docker manifest inspect curvine/curvine:0.3.6-alphareturnsno such manifest, whilev0.3.6-alpharesolves. The heredoc expansion can be seen by running the step 4 command in any shell and looking at the generatedcacheruntimeclass.yaml.Ⅴ. Special notes for reviews
The bucket job in step 2 is left alone. It already escapes its variables with
\$, so it needs the unquoted heredoc.