testland/slo-load-test-plan
Use when SLOs and an endpoint list exist but nobody has decided which load tests to run, what load each carries or what counts as failure. Turns each SLO into scenarios with a load profile, workload model, pass/fail threshold and error-budget allowance, before any tool is configured.
91%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
runners.mdreferences/
Runner executors and a worked plan
Reference material for slo-load-test-plan: how the open/closed distinction
appears in the two most common runners, and a full worked example turning a set
of SLOs into a scenario matrix. The plan itself stays runner-agnostic; this file
is where the runner-specific names and the end-to-end example live.
Open vs closed in the two most common runners
| Runner | Open model | Closed model |
|---|---|---|
| k6 | constant-arrival-rate, ramping-arrival-rate executors (k6 open and closed models) | constant-vus and the other non-arrival-rate executors (k6 open and closed models) |
| Gatling | injectOpen with atOnceUsers, rampUsers, constantUsersPerSec, rampUsersPerSec, stressPeakUsers (Gatling injection) | injectClosed with constantConcurrentUsers, rampConcurrentUsers, incrementConcurrentUsers (Gatling injection) |
Note the trap in the Gatling column: rampUsers and atOnceUsers are
open-model blocks that inject a number of users over a window, while the
closed-model blocks are the *ConcurrentUsers family, which hold a level of
concurrency in the system (Gatling injection).
"Users" in a profile name does not mean closed.
Worked example
Input: a checkout service. SLOs are p95 < 300 ms on POST /api/checkout at
peak, p95 < 100 ms on GET /api/orders at average, error rate < 0.1% under a 3x
spike, and 99.9% availability over four weeks. Traffic mix is
GET /api/orders 60%, POST /api/checkout 20%, everything else 20%. Peak is
300 RPS, average 100 RPS, with a 3x spike on sale days.
Resulting scenario matrix:
| Scenario | SLO governed | Profile | Injection | Target | Duration |
|---|---|---|---|---|---|
smoke-mix | none (gate) | Smoke | Open | 1 arrival/sec | 60 s |
orders-latency-average | p95 < 100 ms on orders | Average-load | Open | 100 arrivals/sec | 30 min |
checkout-latency-peak | p95 < 300 ms on checkout | Stress | Open | 300 arrivals/sec | 30 min |
spike-error-budget | error rate < 0.1% | Spike | Open | 900 arrivals/sec | 3 min |
soak-availability | 99.9% over four weeks | Soak | Open | 300 arrivals/sec | 4 h |
capacity-ceiling | none (capacity finding) | Breakpoint | Open | ramp to failure | until failure |
Thresholds:
| Scenario | Expression | Scope | Abort | Traceable to |
|---|---|---|---|---|
orders-latency-average | p(95)<100 | orders tag | no | orders latency SLO |
checkout-latency-peak | p(95)<300 | checkout tag | no | checkout latency SLO |
spike-error-budget | rate<0.001 | all | no (recovery is the measurement) | spike error-rate SLO |
soak-availability | rate<0.001 | all | yes, delayAbortEval: '10s' | 99.9% availability, pro-rata over 4.32M requests = 4,320 failures |
capacity-ceiling | none | n/a | n/a | capacity finding, not an SLO |
Every load number is open-model arrivals per second, because every SLO here is stated per request rather than per session.