Stress testingAll features.
Everything a Stress test does today: the shape of a run, each figure it measures, how the rating is decided, and what is in the report.
31 features in 7 groups, checked against the product on 14 November 2026.
Setting up a run
One verified site
You choose the site from a list of your own verified sites.
30, 60 or 90 seconds
Three lengths; it starts at 60 seconds.
A fixed peak of six connections
The load is not a setting: every run peaks at six connections.
Four phases
Baseline, about the first sixth of the run: two connections with a pause between requests. Ramp, to the half-way point: connections are added in steps up to six and the pause shrinks to nothing. Peak, to about five sixths: six connections with no pause. Recovery, the last sixth: back to two connections.
The home page, over HTTPS
Every request is a GET of the site's home page from Cloudflare's network, with redirects followed and a 15-second limit.
A recognisable user agent
Requests are sent as BCDRPulse-StressTest/1.0.
What it measures
Figures for each phase
Requests, errors, the error rate, the average response time and the 50th, 95th and 99th percentiles.
Degradation factor
The 95th percentile at the peak divided by the 95th percentile at the baseline. 1.0 means the peak was as fast as the baseline.
Error onset
How many seconds into the run the first request failed, if any did.
Recovery
The first 5-second window after the peak in which the median response time is back within 1.25 times the baseline's. If none is, the report says recovery was not complete.
A reading every 5 seconds
For each 5 seconds: the phase, the requests, the errors, the average and the 50th and 95th percentiles.
Totals
Requests, errors, the error rate, requests per second, and a count for each HTTP status.
A chart in the dashboard
The 95th percentile over the run, coloured by phase.
The rating and the summary
A risk rating by fixed rules
Elevated when 5% of requests or more failed, the peak was 3 times slower than the baseline, or the site did not recover. Moderate when any request failed, the peak was 1.5 times slower, or a 95th or 99th percentile reached 1 second. Otherwise Low. AI does not decide it.
A named result
Robust, Graceful degradation, Strained, Requests failed, Did not recover or Inconclusive, set by the same rules.
It says when the test was refused
When the failures are mostly HTTP 429 or 403, the report says a rate limit or a firewall rule refused the traffic.
An AI summary
Three short parts: the failure mode, what happened and what to check. It is written from the run's own figures, and is told not to state how many users the server can handle.
The report
A PDF for every finished run
With a report ID, the time of the test and the time the report was made.
Its sections
Executive summary, risk assessment, scope and method, a table of the four phases, the degradation analysis, the 5-second timeline as a table, trend, recommendations, and the raw figures with the status codes.
Trend
Beside the previous finished stress test on the same site.
Combined report
On the Reports page up to 50 reports, from this product and the others, can be merged into one PDF with a cover and a contents page.
History
The page shows your 10 most recent runs. The Reports page lists every finished run.
Queue and controls
One run at a time
A second stress test cannot be queued while one of yours is queued or running.
Starts within about a minute
Queued runs are picked up once a minute, oldest first. When the platform is at capacity you are asked to try again in a minute.
A run that hangs is closed
A run with no result after 4 minutes is marked as failed.
Ownership and safety
Verified sites only
A run is accepted only for a site of yours whose DNS record has been confirmed, and the site is checked again when the run starts.
Public addresses only
A host that does not resolve to a public address is refused when you ask for the run and again before it starts.
Plans and limits
Pro
50 stress tests a month.
Business
300 stress tests a month. The size of a run is the same on both plans.
Resets on the 1st
Allowances count over the calendar month, in UTC, and reset on the 1st.
A run that never started is not counted
For example one refused because the host was not a public address.
What it does not do.
- It does not search for a breaking point. The peak is a fixed six connections; Cloud stress testing lets you choose a peak of up to 1,000 users.
- It does not stop early when a site struggles. A run is 90 seconds at most, and it cannot be cancelled.
- It requests the home page only. Behind a CDN the result can be the CDN's cache and not your server.
- It measures the whole response time only: no time to first byte and no bytes.
- It is not a penetration test: nothing is exploited and your application's logic is not tested. We do not offer one.