docs(skills): add perf-investigation skill#185
Draft
rhefner1 wants to merge 1 commit into
Draft
Conversation
A methodology skill for root-causing performance bottlenecks in a deployed GoodData.CN stack from observability data (Grafana/Prometheus/Loki/Tempo), triggered by a slow trace or a "X was slow at time T" report. Leads with the method — symptom != cause, decompose the trace, classify with a signature table, prove before fixing — and keeps the Grafana query patterns and the request-flow map in references. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds a new agent skill,
perf-investigation, for diagnosing performance bottlenecks in a deployed GoodData.CN stack from observability data — given a slow trace ("why is traceabc123taking 13s?") or a time-bounded report ("dashboards were slow ~10 min ago"). It is post-hoc diagnosis, not load-test running.Why
Distilled from a long real performance investigation. The recurring lesson: the symptom is rarely where the cause is, and most gates are small hardcoded concurrency constants throttling a service that sits at low CPU — so the reflex to "add pods/CPU" is usually wrong. The skill encodes a repeatable method so the next investigation doesn't rediscover this the hard way.
Contents
SKILL.md— the method: pin the symptom (ask the user when ambiguous) → reproduce evidence before theorizing → map the request path → localize via trace decomposition → classify with a signature table (idle CPU + deep queue = a concurrency constant; cause vs. victim; architectural ceilings; pool limits; burstable-credit cliffs) → prove (Little's law, read the image source) → name the fix and its lever class. Plus discipline (pre-register hypotheses, one variable at a time, environmental drift) and a workload-realism check before declaring an architectural wall.references/grafana-queries.md— Prometheus/Tempo/Loki query cookbook: CPU + CFS throttling, latency, Pulsar unacked backlog (a common non-obvious culprit), thejqspan-timeline decomposition, slow-trace search, log correlation, and the pull-the-image-and-read-the-source recipe.references/request-flow.md— the GoodData.CN request flow (gateway routing, AFM execution chain) annotated for where bottlenecks live, for cause-vs-victim reasoning.Public/customer-facing: generic service names only, no internal cluster/org/repo references.
Notes for reviewers
request-flow.mdare the bits most worth a second look before this is treated as public-facing docs.🤖 Generated with Claude Code