Fewer mystery restarts
Leaks and build-ups found, so servers don’t need rebooting to stay fast.
[ Why endurance testing ]
Most performance tests run for under an hour. Production runs for weeks between releases. In that time, small problems add up: a cache that never evicts, a connection opened but never closed, a nightly job that runs a little slower each time, a table that grows until one query falls off a cliff. None of these show up in a short test, and all of them eventually wake someone up at 3 a.m. Endurance testing, also called soak testing, gives them the time they need to surface, in a test environment rather than in front of customers.
Need to know your peak capacity? See load testing services →
Leaks and build-ups found, so servers don’t need rebooting to stay fast.
Response times checked at hour 1 and hour 24, not just at the start.
How long you can run before memory, disk or connections become a problem.
Confidence that what you ship today still behaves the same next month.
[ What we test ]
Each one is invisible in a short test and obvious after a long one. Every card lists what we track, when you need it and the problem it prevents.
We watch memory over hours of steady load and separate normal caching from real leaks: memory that is taken and never given back.
When servers need regular restarts, or memory grows from release to release.
A system can meet every target in the first hour and miss them by the end of the day. We compare performance across the whole run.
For any product that runs for long periods between restarts or deploys.
Database connections, threads, file handles and sockets that are opened and never closed. Each leak is tiny; together they bring a service down.
After changes to data access, integrations or concurrency code.
Long runs create real volumes of data. We check that queries, indexes and storage cope as tables grow hour by hour.
For data-heavy products and systems with fast-growing tables.
Logs, uploads, temporary files and caches grow quietly. We track disk use across the run and check that rotation and clean-up actually happen.
For services that write logs, files or local caches.
Backups, reports, imports and clean-up jobs running while users are active. We check they finish on time and don’t drag everything else down.
When nightly or hourly jobs run on the same systems as your users.
A queue that takes in one message a minute more than it processes looks fine for an hour and fails after a day. We watch the trend, not the snapshot.
For event-driven systems and anything with background workers.
Over long runs, containers restart, nodes rotate and autoscaling settles into patterns. We check that your platform stays healthy and predictable.
For Kubernetes and autoscaled platforms that run continuously.
[ Endurance testing vs other performance tests ]
Load, stress and spike tests are short and intense. A soak test is long and ordinary on purpose, because that’s how production really runs. On the right is what a soak test finding looks like.
| Load | Stress | Spike | Endurance | |
|---|---|---|---|---|
| The question | Do we meet targets at expected peak? | Where do we break? | Do we cope with a sudden surge? | Do we stay stable over hours and days? |
| Load level | Up to expected peak | Past peak until failure | Normal to several times peak, fast | Typical busy load, held steady |
| Duration | Long enough at peak to settle | Until failure or a stop rule | Short bursts | Many hours, sometimes days |
| What it finds | Bottlenecks at peak | The limit and how you fail | Slow scaling, queues, retries | Leaks, drift and build-up |
Example only. Real reports use your journeys, targets and data.
[ How we work ]
The length of every run is set by how your system really runs. Every script, chart and report belongs to you.
A steady load based on your normal busy hours, and a run long enough to match how long you go between restarts.
What “stable” means for you: response times, memory, connections and disk, checked at agreed intervals.
Journey scripts and test data that won’t run out over a long run, plus dashboards for every resource trend.
Hours or days of steady load, with trends checked throughout and a stop rule if something goes wrong.
Rising trends matched to heap dumps, profiles, logs and traces, so each one points to a cause.
The same soak repeated after fixes to prove the trends are flat, with scripts ready for your next release.
[ Tools ]
12 tools across three jobs. We’re happy to work in the tools and monitoring you already have.
Open-source tools we script in most often.
When you need scale or already own a licence.
Trends tracked for the whole run.
See every tool we use on our technology stack page →
[ Case Studies ]
Load and performance checks were one part of the wider QA work on these products. Each links to the full write-up.
[ Ways to work ]
A one-off soak before a big release, a performance engineer in your team, or scheduled soaks before every release.
Not sure which fits? Find your model in four questions, compare all six or estimate your cost.
[ FAQ ]
{{ f.a }}
[ Contact us ]
✓ Free one-week POC✓ NDA before access✓ A QA lead replies the same day
[ Keep exploring ]