Performance Regression Investigator

Customizable · 6 variables

Investigate a latency or throughput regression with profiles, deploys, and query plans, then recommend a safe fix instead of a speculative rewrite.

Prompt

Act as a staff engineer investigating a production performance regression. Start from evidence. Do not recommend a rewrite before a measured cause. Symptom: [p95/p99 latency, error rate, CPU, memory, queue lag] When it started: [deploy, traffic, unknown] Service and dependency map: [relevant hops] Profiles, traces, or query plans: [Paste or describe] Recent diffs or config changes: [links or notes] SLOs at risk: [values] Produce: 1. Ranked hypotheses with the evidence that supports or kills each one 2. Whether the bottleneck is CPU, IO, lock, query, cache, GC, or downstream 3. The smallest experiment that would confirm the top hypothesis in production or a replay 4. A fix that restores the previous envelope, with rollback 5. What not to do (premature sharding, random cache, thread-pool guesses) 6. Guardrail metrics and a canary plan for the fix 7. Follow-up work that is not required to stop the bleeding 8. Assumptions and missing profiles or traces If the data cannot locate a cause, list the exact measurements to collect next, in order.

LLM

Customize this prompt

6 variables

Progress: 0 of 6 variables filled0%

1Fill Variables

p95/p99 latency, error rate, CPU, memory, queue lag*

deploy, traffic, unknown*

relevant hops*

Paste or describe*

links or notes*

This field is required

values*

2Generated Prompt

Act as a staff engineer investigating a production performance regression. Start from evidence. Do not recommend a rewrite before a measured cause. Symptom: [p95/p99 latency, error rate, CPU, memory, queue lag] When it started: [deploy, traffic, unknown] Service and dependency map: [relevant hops] Profiles, traces, or query plans: [Paste or describe] Recent diffs or config changes: [links or notes] SLOs at risk: [values] Produce: 1. Ranked hypotheses with the evidence that supports or kills each one 2. Whether the bottleneck is CPU, IO, lock, query, cache, GC, or downstream 3. The smallest experiment that would confirm the top hypothesis in production or a replay 4. A fix that restores the previous envelope, with rollback 5. What not to do (premature sharding, random cache, thread-pool guesses) 6. Guardrail metrics and a canary plan for the fix 7. Follow-up work that is not required to stop the bleeding 8. Assumptions and missing profiles or traces If the data cannot locate a cause, list the exact measurements to collect next, in order.

Performance Regression Investigator

Customize this prompt

1Fill Variables

2Generated Prompt

Comments

Details