Diag Search

Find what the vault already knows about a bug before diagnosing from zero. Pure ripgrep over ~/ops — frontmatter labels first, then symptom keywords, then the cluster index.

Step 1 — Form the query

From the bug at hand, extract: component(s) (scheduler, device-plugin, runtime, k8s distro, …), failure type (scheduling failure, OOM, …), vendor (gpu vendor a, gpu vendor b, …), and 2–3 distinctive symptom keywords.

Step 2 — Search by labels, then keywords

# label hits — frontmatter only
rg -l 'labels:.*<component>' ~/ops/*.md
# keyword hits — symptom terms across body
rg -il -e '<keyword1>' -e '<keyword2>' ~/ops/*.md

Union both sets. Zero hits on labels alone means broaden to keywords; zero on both means go to Step 2b — do NOT conclude the vault has nothing yet (vocabulary mismatch is the #1 false-negative here).