Skip to content

Revert #1166: un-mute the Advisor check now that #1165 is fixed - #1170

Open
bjagg wants to merge 1 commit into
LIF-Initiative:mainfrom
bjagg:revert-1166-unmute-advisor
Open

Revert #1166: un-mute the Advisor check now that #1165 is fixed#1170
bjagg wants to merge 1 commit into
LIF-Initiative:mainfrom
bjagg:revert-1166-unmute-advisor

Conversation

@bjagg

@bjagg bjagg commented Aug 28, 2026

Copy link
Copy Markdown
Contributor
Description of Change

Reverts #1166. The Advisor check pages again.

The mute existed because the check failed 100% of the time on a known, tracked data defect, and a check that always fails just trains people to filter the alert topic. #1167 fixed the defect and #1169 promoted it to demo, so the reason is gone. Dereck called this revert out as item 2 in his #1167 review.

Verified on demo before reverting — the Advisor happy path passes unmuted, and the payload that broke it is gone:

before after
response bytes for the query the Advisor issues 6,271,948 10,386
approx tokens (model limit 1,047,576) ~1,567,987 2,596
Name / Contact / Identifier (learner 100001) 1,294 / 1,296 / 1,876 2 / 2 / 3
CourseLearningExperience 5,096 items 8
CredentialAward 5,733 items 9

The query is the exact GraphQL the semantic-search service logged, replayed against demo org1. The record collapsed to its true fragment set on its own after the first cache refresh — no data surgery, exactly as #1167 predicted.

Why not leave the mute in. It would keep a working check from paging, and a green monitor would go on meaning "MDR/LDE is healthy" rather than "the demo works" — which is the misreading the mute's own PR body warned about.

Why this doesn't reintroduce the noise. Alerting still requires two consecutive failures before it pages (#1157), so a single flake after this lands stays silent.

Related Issues

Refs #1165

Type of Change
  • Infrastructure/deployment change
Project Area(s) Affected
  • test/ or e2e/

Checklist
  • commit message follows commit guidelines (see commitlint.config.mjs)
Testing
  • Manual testing performed

Monitor run after the demo deploy: success, with the Advisor step genuinely passing (Advisor happy path: success — not the continue-on-error artifact, which is what the mute's summary line existed to disambiguate). Workflow YAML re-validated after the revert: 10 steps, no continue-on-error, every if: condition intact.

Additional Notes

Follow-up worth its own issue, found during the promotion: .github/workflows/lif_query_cache_api.yml lists components/lif/query_cache_service/** in its path filter but not components/lif/composer/**, even though projects/lif_query_cache_api ships that brick. #1167 rebuilt only because it happened to also touch the call site — a composer-only change would merge and silently never rebuild the image.

…e its data defect is open"

This reverts commit 69d60e6.

The mute existed because the Advisor check failed 100% of the time on a known,
tracked data defect, and a check that always fails trains people to filter the
alert topic. LIF-Initiative#1167 fixed the defect and LIF-Initiative#1169 promoted it to demo, so the reason
is gone.

Verified before reverting, on demo:

- the Advisor happy path passes on its own, unmuted
- the query the Advisor actually issues (replayed from the semantic-search log)
  returns 10,386 bytes, down from 6,271,948 -- about 2,600 tokens against a
  1,047,576 limit it used to blow by 50%
- learner 100001 collapsed to its true fragment set: Name 2, Contact 2,
  Identifier 3, from 1,294 / 1,296 / 1,876
- CourseLearningExperience 5,096 -> 8 items, CredentialAward 5,733 -> 9

Leaving the mute in place after the fix is the worse failure: it would keep a
working check from paging, and a green monitor would go on meaning "MDR/LDE is
healthy" rather than "the demo works".

Alerting still requires two consecutive failures before it pages, so a single
flake after this lands stays silent -- un-muting does not reintroduce the noise
the mute was written to stop.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Backlog

Development

Successfully merging this pull request may close these issues.

1 participant