Load Testing
| Testing tests | purpose | duration | typically load |
|---|---|---|---|
| Smoke test | verify system works | 1-5 min | very low |
| Load | verify expected traffic | 5-30 min | expected production load. |
| Stress | Find breaking point | 10-30 min | Beyond expected load! |
| Spike | Test sudden traffic surges | Few mins | Instant large jump |
| Soak | Find long term issues. | Hours | Normal load |
| Scalability | verify horizontal/vertical scaling | load. | |
| Breakpoint | Find exact failure point | ||
| Failure | verify recovery from failures | ||
| Chaos | Intentionally inject failures | ||
| Capacity | Determine max sustainable throughput. | ||
| Resilience test | |||
| Performance regression test |
Load Testing proper table (generally used ones)
- Test type: Smoke test
- Purpose :- Verify test setup & basic functionality
- Typical load vs normal :- 1-5% (10-15 users)
- Duration :- seconds to 5 min
- Load Test (must have)
- Validate performance under expected load
- load :- 100% (1000 users)
- Duration :- 15 min to 1 hr
- Avg. load test
- Simulate normal business day
- load :- 50-80% (500-800 users)
- duration :- 30 min to several hrs
- Stress test (must have)
- Find breaking point
- load :- 120-500% (1200-5000+ users)
- duration :- minutes
- Soak/endurance test (sustained test) (must have)
- Long period check to detect memory leak, connection leak, eta degradation, etc.
- load :- 70-100% (700-1000 users sustained)
- duration :- 4-72 hrs (3 days)
- Capacity test (sometimes )
- Find max load, while still meeting SLA.
- SLA :- Service level agreement.
- A document agreement
- defines what level of service must be defined & how it is measured.
- typical metrics include uptime, response time, resolution time, availability.
- If provide fails to meet targets, SLA may specify penalties, service credits or remedies.
- eg: if an application’s SLA is
- availability = 99.9%
- P95 latency < 300 ms
- error rate < 0.1%
- then during load tests these must be maintained.
-
Related terms
- SLI (Service level indicator)
- The measurement values
- Eg: availability, P95 latency, P95 etc
- SLO (Service level objective)
- A target for an SLI / engineerng goal
- Eg: P95 latency < 300ms - over a rolling 30-day period
- 99% successful requests over 30 days
- SLA (Service level agreement)
- customer facing promise
- involves business/legal consequences
- P95 latency (Percentile 95 latency)
- A method to display response times/latency
- monitoring:
- Eg: P95 = 300 ms means 95% of requests completed in 300ms or less - 5% faile.
- SLI (Service level indicator)
- load :- Gradually increase from 100% upward.
- duration :- 30 min - 2 hr
- Classically stress test but to find near point in compliance of SLA).
- Scalability test
- verify scaling behaviour when adding resources.
- load :- incrementally increasing load.
- duration :- Hours
- Volume test :-
- Test large data size (not user count)
- load :- normal users, huge data volume.
- duration :- series
- Concurrency test :-
- Many users hitting same resource simultaneously.
- load :- 100-300% on specific endpoint
- durations :- minutes
- Resilience test
- Behaviour during failure.
- load :- normal load + injected failures.
- duration :- 30 min to hours.
- Recovery test (outage)
- Recovery after outage/crash.
- load :- usually after stress/spike.
- duration :- mins to hours.
- Peak test (traffic)
- Validate operation during known peak traffic.
- load :- 100-150% of highest expected peak.
- duration :- 30 min to 2 hrs.
- diff from concurrency testing in a way that enteric apps heavy load is tested.
- Testing expected real life, high traffic like black friday sale.
Common load testing tools
- From local m/c to deployed appk6, Apache JMeter (UI based testing - JSlocust (Python scripts based)artillery (yaml based ver &)
- k6
- JS based
- CLI/UI friendly
- used heavily w/ nestJS + K8s
- Artillery
- node native alternative
- yaml or JS based
- Apache JMeter
- GUI based.
- Manual entry of endpoints.
- used in banks / large enterprises.
- legacy
- locust
- python based.
- distributed load testing support.
- For backend HTTP tools (for quick checks only).
- wrk, autocannon, hey
- used for raw benchmarks & not full scenarios.