Prompt (performance level)
Prompt text
“Draft
performance and load test scenarios for \[service/endpoint].
Specify SLAs/SLOs, target RPS, concurrency patterns, payload sizes,
ramp-up, soak, spike, and metrics to collect (latency percentiles,
errors, saturation).”
How to apply
critical thinking
·
Clarify performance requirements
·
SLAs (e.g.,
p95 < 300 ms) and SLOs (e.g., 99.9%
of requests under 500 ms).
·
Expected peak vs normal
traffic, growth projections.
·
Generate performance scenarios
·
Baseline:
light load to validate harness and get reference metrics.
·
Load:
sustained target RPS with realistic traffic mix.
·
Stress/spike:
sudden surges above normal to find breaking points.
·
Soak:
long-running tests to detect leaks, slow degradation.
·
Concurrency:
many simultaneous users, with realistic think times.
·
Questions on ambiguities
·
What’s the exact definition
of “failure” for this service?
·
Which resources
are most constrained (CPU, memory, DB connections, network)?
·
Are there autoscaling
policies or rate limits that influence scenarios?
·
What test ideas might be missed
·
Multi-tenant
contention: one tenant starving others.
·
Thundering herd effects after outages or cache flushes.
·
Performance in non-happy
paths (errors, retries, timeouts).
Output template (performance/load
testing)
Context: [service/endpoint, architecture (monolith/microservice), criticality level]
Assumptions: [SLA/SLO targets, expected traffic patterns, caching, autoscaling]
Test Types: [performance, load, stress, soak]
Test Cases:
ID: [TC-PERF-001]
Type: [performance/load]
Title: [e.g., "Sustained load at target RPS"]
Preconditions/Setup:
- [Prod-like environment, data size, configs]
- [Monitoring/observability in place]
Steps:
1. [Configure load tool with target RPS, concurrency, payload mix]
2. [Run test for defined duration]
Variations:
- [Baseline low load]
- [Target steady-state load]
- [Spike from low to 2x/3x target]
- [Soak at target for N hours]
- [Different payload sizes and request mixes]
Expected Results:
- [Latency percentiles within SLO]
- [Error rate within acceptable limits]
- [Resource utilization in safe bounds, no leaks]
Cleanup:
- [Reset environment, clear queues/caches, roll back configs if changed]
Coverage notes:
- [Which scenarios/traffic patterns are validated; untested extremes]
Non-functionals:
- [Capacity headroom, resiliency under stress, autoscaling behavior]
Data/fixtures:
- [Realistic datasets, representative payload samples]
Environments:
- [Dedicated perf env, or carefully controlled prod-like setup]
Ambiguity Questions:
- [Exact SLOs, business impact of degradation, allowed mitigation strategies]
Potential Missed Ideas:
- [Downstream dependency limits, batch jobs overlapping with peak traffic, cold-start behavior]