Endurance & Soak Testing Services | Long-Run Stability Skip to main content
QAonTOP
Contact Us

Endurance Testing Services

Some problems only appear after hours of normal use: memory that never comes back, a queue that grows by a few messages a minute, a log disk that fills on day three. Endurance testing finds them before your users do.

We hold realistic load on your system for hours or days and track every resource trend while it runs. You get clear charts of what stayed flat and what crept up, the leak or build-up behind each trend, and how long you could run before it would hurt.

  • Runs of hours to days
  • Memory, connections and disk tracked
  • JMeter, k6, Gatling and more
  • Free one-week POC

[ Why endurance testing ]

Passing a one-hour test doesn’t mean you’ll last the week

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 →

01

Fewer mystery restarts

Leaks and build-ups found, so servers don’t need rebooting to stay fast.

02

Steady speed all week

Response times checked at hour 1 and hour 24, not just at the start.

03

A known runway

How long you can run before memory, disk or connections become a problem.

04

Safer gaps between releases

Confidence that what you ship today still behaves the same next month.

[ What we test ]

Eight things that quietly build up over time

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.

Memory leak detection

Heap, GC, native memory

We watch memory over hours of steady load and separate normal caching from real leaks: memory that is taken and never given back.

What we test
  • Heap and native memory trends per service
  • Garbage collection frequency and pause times
  • Heap dumps and allocation profiles as memory climbs
  • Container memory limits and out-of-memory restarts
When you need it

When servers need regular restarts, or memory grows from release to release.

The problem it preventsCrashes that arrive like clockwork every few days

Response-time drift

Hour 1 vs hour 24

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.

What we test
  • Response times per journey across the run
  • Throughput holding steady at constant load
  • Error rates creeping up over time
  • The first hours compared with the last
When you need it

For any product that runs for long periods between restarts or deploys.

The problem it preventsA product that gets slower every day until someone restarts it

Connection & thread leaks

Pools, sockets, handles

Database connections, threads, file handles and sockets that are opened and never closed. Each leak is tiny; together they bring a service down.

What we test
  • Database and HTTP connection pools over time
  • Thread counts and blocked threads
  • Open file handles and sockets
  • What happens when a pool or limit finally runs out
When you need it

After changes to data access, integrations or concurrency code.

The problem it preventsA service that stops accepting work after a few days

Long runs create real volumes of data. We check that queries, indexes and storage cope as tables grow hour by hour.

What we test
  • Query times as tables and indexes grow
  • Locks, deadlocks and long-running transactions
  • Database storage and transaction log growth
  • Archiving, clean-up and maintenance jobs
When you need it

For data-heavy products and systems with fast-growing tables.

The problem it preventsA query that’s fine on day one and times out on day thirty. See also data & database testing. For very large data sets, see volume testing.

Disk, logs & temporary files

Storage over time

Logs, uploads, temporary files and caches grow quietly. We track disk use across the run and check that rotation and clean-up actually happen.

What we test
  • Log volume per hour and log rotation
  • Temporary and upload files cleaned up
  • Local caches and session stores kept bounded
  • Alerts firing before disks run out
When you need it

For services that write logs, files or local caches.

The problem it preventsAn outage caused by a full disk nobody was watching

Scheduled jobs under steady load

Cron, batch, backups

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.

What we test
  • Batch and scheduled jobs overlapping live traffic
  • Job duration across repeated runs
  • Backups and index rebuilds during load
  • Queues and backlogs that jobs create
When you need it

When nightly or hourly jobs run on the same systems as your users.

The problem it preventsA nightly job that slowly grows until it runs into the morning rush

Queues & background workers

Kafka, RabbitMQ, SQS

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.

What we test
  • Queue depth and consumer lag over time
  • Worker memory and restarts
  • Dead-letter queues growing
  • Retries and duplicate processing
When you need it

For event-driven systems and anything with background workers.

The problem it preventsA backlog that takes hours to clear once it’s finally noticed

Over long runs, containers restart, nodes rotate and autoscaling settles into patterns. We check that your platform stays healthy and predictable.

What we test
  • Pod restarts and out-of-memory kills
  • Autoscaling that stays stable instead of flapping
  • Certificates, tokens and sessions expiring mid-run
  • Cloud cost per hour at steady load
When you need it

For Kubernetes and autoscaled platforms that run continuously.

The problem it preventsSurprise restarts, flapping capacity and bills that creep up. See also cloud & SaaS testing.

[ Endurance testing vs other performance tests ]

The other tests ask “how much?” Endurance testing asks “for how long?”

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.

AspectLoadStressSpikeEndurance
The questionDo 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 levelUp to expected peakPast peak until failureNormal to several times peak, fastTypical busy load, held steady
DurationLong enough at peak to settleUntil failure or a stop ruleShort burstsMany hours, sometimes days
What it findsBottlenecks at peakThe limit and how you failSlow scaling, queues, retriesLeaks, drift and build-up
Sample soak test finding24-hour soak
Load
70% of normal peak, held for 24 hours
Hour 1
All journeys within target
Hour 18
Search slowed from 0.6 s to 2.4 s (95th percentile)
Trend
Search service memory up about 180 MB an hour, never released
Root cause
In-memory cache for search filters with no eviction
Fix
Size limit and expiry added to the cache
Re-soak
Memory flat and search within target for 24 hours

Example only. Real reports use your journeys, targets and data.

[ How we work ]

Six steps from soak plan to flat trend lines

The length of every run is set by how your system really runs. Every script, chart and report belongs to you.

  1. Step 1 of 6

    Choose the load and the length

    A steady load based on your normal busy hours, and a run long enough to match how long you go between restarts.

    You getSoak test plan
  2. Step 2 of 6

    Agree targets and checkpoints

    What “stable” means for you: response times, memory, connections and disk, checked at agreed intervals.

    You getAgreed targets
  3. Step 3 of 6

    Script and wire up monitoring

    Journey scripts and test data that won’t run out over a long run, plus dashboards for every resource trend.

    You getTest scripts you ownTrend dashboards
  4. Step 4 of 6

    Run the soak

    Hours or days of steady load, with trends checked throughout and a stop rule if something goes wrong.

    You getRun results
  5. Step 5 of 6

    Find the leaks

    Rising trends matched to heap dumps, profiles, logs and traces, so each one points to a cause.

    You getLeak and build-up report
  6. Step 6 of 6

    Fix, re-soak, hand over

    The same soak repeated after fixes to prove the trends are flat, with scripts ready for your next release.

    You getBefore-and-after reportReusable soak suite
  • Free one-week POCWe run an overnight soak on one key journey and show you every trend that moved, so you can judge our work first.
  • Watched, not left aloneLong runs are checked throughout, with a stop rule agreed with your team and hosting provider.
  • NDA before accessWe sign your NDA first and use test accounts and test data by default.

[ Tools ]

The tools behind our endurance testing

12 tools across three jobs. We’re happy to work in the tools and monitoring you already have.

Load generation

Open-source tools we script in most often.

04 TOOLS
  • JMeterProtocol-level load for web and APIs
  • k6Load tests as code, CI-friendly
  • GatlingHigh-throughput HTTP load
  • LocustUser behaviour written in Python

Enterprise & cloud execution

When you need scale or already own a licence.

03 TOOLS
  • LoadRunnerEnterprise protocols and large suites
  • NeoLoadEnterprise load with CI integration
  • BlazeMeterCloud load generators

Monitoring & analysis

Trends tracked for the whole run.

05 TOOLS
  • GrafanaLive dashboards and trends
  • PrometheusServer and service metrics
  • New RelicAPM and traces
  • DatadogInfrastructure and APM
  • AWS CloudWatchAWS resource metrics

See every tool we use on our technology stack page →

[ Case Studies ]

Products that run around the clock

Load and performance checks were one part of the wider QA work on these products. Each links to the full write-up.

All case studies →

[ Ways to work ]

Six ways to get endurance testing done

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 ]

Endurance testing questions, answered

{{ f.a }}

[ Contact us ]

Find the problems that only show up on day three

✓ Free one-week POC✓ NDA before access✓ A QA lead replies the same day

[ Keep exploring ]

Related testing services

All services →