Document soak results and categorize degradation

Record what changed, when it started, and how fast it progressed so you can compare this run with the next one. Open the finished run in Grafana Cloud k6 (Testing & synthetics > Performance, then select the run), read Performance Overview (including the timeline), then use the HTTP and Threshold tabs. Refer to Inspect test results for how those sections fit together. These steps use the Grafana Cloud k6 Test result view; optional in-product guided steps may be added later without changing the learning goals below.

To document your soak results and categorize degradation, complete the following steps:

1. Open your completed soak test result in Grafana Cloud k6.

   From the main menu, go to Testing & synthetics > Performance, open your project and test if prompted, then select the completed run from the Test runs table, or use the results URL from your terminal if you still have it.

   Using the Performance Overview timeline, record the following values at three points in time in a spreadsheet or shared doc:

   Expand table

   Metric	At 15 minutes	At degradation onset	At test end
   Response time (ms)
   Failure rate (%)
   Request rate (req/s)

   Hover over the timeline at each phase to read the values. The 15-minute mark is a baseline after warm-up on this 30-minute learning run. If no degradation occurred, the “at degradation onset” and “at test end” columns should match the 15-minute values.

2. Click the HTTP tab to see aggregated p95 and p99 latency percentiles for the full test run.

   These values blend all phases together, so they won’t show per-phase p95. Comparing the overall p95 against your baseline test’s p95 shows whether sustained load caused aggregate latency to drift.

3. Click the Threshold tab to confirm whether your p95 threshold (p(95)<500) passed or failed.

   A threshold that passed for the full soak duration means the system stayed within the allowed latency envelope under sustained load.

4. Record the degradation onset time.

   This is the elapsed time from test start to the first sustained change in metrics. “Sustained” means the metric did not return to baseline within five minutes, which distinguishes real degradation from transient fluctuations.

   If the test ran for the full duration with no observable change, record “no degradation within [duration].”

5. Calculate the degradation rate if metrics changed during the test.

   Divide the total change in a metric by the elapsed time over which it changed:

   • Latency drift rate: (p95 at test end − p95 at 15 min) / test duration in hours. On a 30-minute learning run, a rise from 200 ms to 350 ms is 150 ms over 0.5 hours, or 300 ms/hour. On a longer production soak, the same math applies with the full plateau duration (for example 75 ms/hour if p95 rose from 200 ms to 350 ms over 2 hours).
   • Error accumulation rate: Total errors divided by test duration. A rate that increases over time (more errors per hour in the second hour than the first) points to progressive resource exhaustion.

   The drift rate helps prioritize work. At 10 ms/hour, a 500 ms threshold may stay green for many hours; at 100 ms/hour, the same threshold may fail much sooner.

6. Categorize the degradation type based on the pattern you observed and the infrastructure correlation from the previous milestone:

   • Memory-related: Latency rises gradually; application memory or GC metrics climb on the system under test.
   • Connection-related: Latency stable, then jumps when the pool is exhausted; errors look like timeouts or connection refused.
   • Disk-related: Performance changes when disk space or log volume approaches limits.
   • No degradation: Metrics stable for the full run. Extend duration or load in a future soak to probe further.

7. Save your documented results alongside baseline and other advanced test records.

   Label each entry with test date, VU count, and duration so you can compare runs over time.

In the next milestone, you map your findings to follow-up actions based on the degradation category.