Load Testing Anti-Patterns

Purpose: Use this file to prevent unrealistic load tests, select performance budgets, and integrate performance checks earlier in delivery.

ID	Anti-pattern	Symptom	Fix
`AP-01`	Happy Path Only	only success cases, no degraded traffic	include `20-30%` 4xx/5xx/timeout/error scenarios
`AP-02`	Unrealistic Data	test data does not match production distributions	use sanitized production-like data
`AP-03`	Single Endpoint Focus	one endpoint looks fine, whole journey still fails	model end-to-end user flows
`AP-04`	Missing Think Time	artificial overload from zero pacing	add realistic `sleep(1-3s)` pacing
`AP-05`	Flat Load Profile	no insight into scale transitions or spikes	combine ramp, spike, soak, and stress patterns
`AP-06`	Testing in Isolation	staging numbers do not predict production	use production-like environments or document scaling corrections
`AP-07`	Shared Test Environment	noisy results from parallel teams	reserve isolated capacity or compare against a baseline
`AP-08`	No Warmup Phase	cold caches and JIT distort results	warm up for `5-10 min` before measurements
`AP-09`	Average-Only Metrics	tails and outliers are hidden	always report `p50/p95/p99/max` and histograms
`AP-10`	One-Shot Testing	flaky or non-reproducible conclusions	run at least `3` times and compare variance

Pattern	Why it matters to Siege	Detection hint
Busy Database	backend is saturated before app code is	DB CPU, long queries, lock wait
Busy Front End	browser or edge becomes the bottleneck	main-thread and CPU profile
Chatty I/O	too many small calls dominate latency	request count vs latency
Extraneous Fetching	payload size is larger than user value	response size vs data used
Improper Instantiation	repeated object creation adds GC pressure	allocation and GC frequency
Monolithic Persistence	one datastore cannot fit all access patterns	read/write pattern mismatch
No Caching	identical expensive work repeats	duplicate query analysis
Noisy Neighbor	one tenant distorts another	per-tenant resource dashboards
Retry Storm	errors amplify load	retry rate vs backend saturation
Synchronous I/O	threads block on remote work	thread-pool starvation and latency growth

Delivery stage	What Siege should recommend
Design	define performance budgets and capacity assumptions
Development	micro-benchmarks, N+1 checks, component-level hotspots
CI	threshold-based regression checks for impacted paths
Pre-release	production-like scenario replay and SLO validation

/siege