fix(tests): reject unreadable RD preset schemas - #4191

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

Merge Risk: ⚪ Minimal · up to 47fe0

The checker retains safe preset matching and reports schema-read failures; the added tests cover these behaviors. No concrete merge-blocking regression is established.

• Fix all pre-merge checks with AI

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

The Broken pipe in those logs is not the jq pipe. It's line 39, the preset loop:

if ! printf '%s\n' "$enums" | grep -Fqx -- "$want"; then

grep -q exits on the first match and closes the pipe, printf takes EPIPE, pipefail promotes it into the pipeline status, and ! reads that as "not found". The value has to be present for the check to call it missing.

bash names the last line of a multi-line command in that message, so the jq substitution would have reported line 32:

$ cat -n t.sh
     5  out=$(printf '%s\n' "$big" \
     6    | head -n 1 \
     7    || true)
     9  printf '%s\n' "$big" | grep -Fqx -- "1"
$ bash t.sh
t.sh: line 7: printf: write error: Broken pipe
t.sh: line 9: printf: write error: Broken pipe

The one-preset FAIL says the same. Truncated jq output is a prefix, so everything past the cut goes missing at once, and rabbitmq.yaml holds a single copy of the enum with s1.xlarge near the middle, so that run would have listed about half the set. It listed one. Four reds, one preset and one Broken pipe in each: rabbitmq/s1.xlarge and postgres/m1.small on 09-09, harbor/c1.medium and http-cache/t1.2xlarge in run 34454715175 attempts 1 and 2 today. The loop runs 21 files x 47 values per run and one of those 987 pipelines loses the race.

Line 39 needs the same here-string you gave jq:

if ! grep -Fqx -- "$want" <<<"$enums"; then

Without that the check keeps going red after this merges, with a different file and preset each time.

This is now stopping E2E from starting on other PRs, not only reddening the unit lane. On #3802 two runs died the same way with a different victim each time, and E2E (in-tree) was skipped both times because the unit lane runs first:

run 34560955738   opensearch-rd     missing: c1.4xlarge
run 34565433782   tcp-balancer-rd   missing: m1.large

Both presets are present in those files and hack/check-rd-presets.sh passes on the same tree locally, so that PR has no way to reach a green run while this sits on main.

This PR and #4194 rewrite the same membership check, so whichever lands second conflicts on that line. Which of the two should go in?

The here-string part landed in #4204, so after a rebase that line resolves to what main has. The jq status guard and the bats suite are still new. Main has no test for this script, so both are worth keeping.