Understanding results

Learn how to read load test metrics and make data-driven decisions about your infrastructure.

Key metrics

Response time

http_req_duration

How long your server takes to respond. Look at the median (p50) for typical performance and p95/p99 for tail latency.

Good
Within the latency budget agreed for this workload
Investigate
p95 over 1s for APIs, or rising steadily over time

Throughput

http_reqs

Requests per second your server handles. In an arrival-rate run the plan fixes the rate sent, so throughput should track the plan.

Good
Throughput matches the rate the plan asked for
Investigate
Throughput falls short of the plan: k6 ran out of VUs and dropped iterations, because responses slowed down

Error rate

http_req_failed

Share of requests that returned an HTTP error or failed to connect. loadtest.dev splits it into 5xx, 4xx, 429, timeout and transport.

Good
Within the error-rate threshold agreed for this workload
Investigate
Over 1%: look at which class grew before fixing anything

Rate limited

429

Responses refused by a rate limiter. The service reports 429s separately; investigate the target and any upstream limiter before increasing the rate.

Good
Within the target’s agreed rate-limit policy
Investigate
Unexpected 429s: inspect the target and upstream limits

Check pass rate

checks

Standalone k6 scripts can validate response bodies, headers and status codes with checks. The loadtest.dev plan does not accept custom checks, and this metric is not part of its run summary.

Good
Over 99% pass rate
Investigate
Under 95%: your API returns unexpected responses under load

Data transfer

data_received / data_sent

Standalone k6 reports transferred bytes. This can help identify payload changes, but loadtest.dev does not expose this metric in its run summary.

Good
Consistent with expected response sizes
Investigate
Much larger than expected: check for verbose logging or debug responses

Understanding percentiles

An average can hide slow responses. If p99 is 5 seconds, roughly one percent of measured requests took at least that long. loadtest.dev merges scenario histograms within a run before reading percentiles; it does not average scenario percentiles.

PercentileWhat it means
p50 (median)Half of requests are faster than this. Your typical response time.
p9090% of requests are faster. A good SLA target for most APIs.
p9595% of requests are faster. The usual standard for API performance SLAs.
p99Only 1 in 100 requests is slower. Critical for high-traffic apps.

Red flags to watch for

Response time rises with the rate
Investigate resource saturation, dependency latency and request queues before deciding whether to scale.
Throughput falls below the planned rate
Check generator capacity, request timeouts, database connection pools and target rate limits.
Error rate spikes suddenly
Inspect the error classes, target logs, memory, CPU and connection pools to identify the cause.
Wide gap between p50 and p99
Inconsistent performance. Look for cold caches, garbage collection pauses, or noisy neighbours.
loadtest.dev — Part of the Narduk Enterprises network