[CRCR] Add additional info that helps triage the issue when health is degraded in PR and Nightly

[jewelkm89]

[CRCR] Add additional info that helps triage the issue when health is degraded in PR and Nightly

Current Behavior

The health cards show only a summary:

• PR tab: "Degraded - 4/5 recent PRs passed · pytorch/crcr-test"
• Nightly tab: "Degraded - 3/5 recent nightlies passed"

No details about which PR/SHA failed or which jobs caused the failure.

Data Currently Available

PR Health (crcr_health_last_prs):

SELECT pr_number, last_run, successes, total, pass_rate
-- Returns aggregate per PR, no job-level details

Nightly Health (crcr_nightly_dashboard):

SELECT pytorch_head_sha, job_name, status, conclusion, started_at
-- Has job-level details but not surfaced in UI

Information Needed for Triage

Field	PR Tab	Nightly Tab
Which PR/SHA failed?	Need pr_number	Need pytorch_head_sha
Which job(s) failed?	Not available	Available but not shown
Job conclusion	Not available	Available but not shown
Link to failing run	Not available	Not available
Failure timestamp	Not available	Available but not shown

Proposed UI Changes

Option A: Expandable Card

┌─────────────────────────────────────────────┐
│ CRCR Relay Health                           │
│ Degraded ▼                                  │
│ 4/5 recent PRs passed                       │
├─────────────────────────────────────────────┤
│ ❌ PR #12345 - 3/4 jobs passed              │
│    └─ l2-critical: failure (link)          │
│ ✅ PR #12344 - 4/4 jobs passed              │
│ ✅ PR #12343 - 4/4 jobs passed              │
│ ...                                         │
└─────────────────────────────────────────────┘

Option B: Popup/Modal on Click

Click "Degraded" → opens modal with:

• List of recent PRs/SHAs with pass/fail status
• Expandable rows showing failed jobs
• Direct links to GitHub Actions runs

Technical Changes Needed

1. New/Modified Query for PR Job Details

-- crcr_health_pr_job_details
SELECT 
  pr_number,
  job_name,
  conclusion,
  workflow_run_url,
  started_at
FROM default.crcr_workflow_job
WHERE downstream_repo = 'pytorch/crcr-test'
  AND pr_number IN (SELECT pr_number FROM recent_failing_prs)
  AND status = 'completed'
ORDER BY pr_number DESC, started_at DESC

2. Frontend Components

File	Changes
torchci/pages/crcr/index.tsx	Add state for expanded view, fetch job details
CrcrTestHealthCard	Add expandable/clickable UI
CrcrNightlyHealthCard	Surface existing job details from nightlyJobs

3. API Endpoint

New endpoint: /api/clickhouse/crcr_health_pr_job_details

Acceptance Criteria

• Clicking health card shows which PRs/SHAs failed
• Failed PRs/SHAs show which jobs caused failure
• Each failed job has a link to the GitHub Actions run
• Works for both PR and Nightly tabs
• No ClickHouse query required for basic triage

Related

• The nightly card already fetches job-level data via crcr_nightly_dashboard but only uses it for aggregate pass/fail calculation, not display

[subinz1]

Related pull requests

• [CRCR] Detect overdue in-progress health probes #8801 detects overdue in-progress CRCR health probes, so a broken zombie sweeper degrades health rather than hiding the stuck job.
• [CRCR] Add health card triage details #8834 adds clickable pull-request and nightly health-card details, including the affected PR or SHA, failing job, timestamp, and GitHub Actions link.

These are stacked: merge #8801 first, then retarget #8834 to main.