Running a Test
From the web interface
Open a project, go to Run Test, choose where to run, and press Run.
There is no separate deploy step. Atlas reserves the machines, powers on any that are switched off, copies the files, and starts JMeter.
Your settings are kept
Everything you set on Run Test — where it runs, the test settings, extra JMeter args, AI Summary — is kept for next time, for you in this project, on any browser. You don't need a preset for that. If you picked a preset, it stays selected; when your changes differ from it, the page says so and offers to save them into the preset, which never changes on its own. Reset Form in the Preset section starts again from the plan's values.
Test Settings: users, ramp-up, duration and variables
Test Settings has a tab for each thread group of your plan, one for each User Defined Variables block, and a Properties tab. Change any value for this run without editing the plan: Atlas changes only its own copy on the load generators, and the file you uploaded stays as it is.
- Thread groups have Users per load generator, Ramp-up, Duration and Loop count (tick Infinite to repeat until the duration ends, or until you stop the run), filled with the values in your plan.
- Setting a Duration makes each user run for that long. If the loop count is a number, users stop at whichever comes first.
- When a setting reads a property (
${__P(threads,10)}) that other parts of the plan read too, the field shows the property's name and value. Click the name to change it under Properties, where it changes everything that uses it. To give one thread group its own value, press the pencil; Reset goes back to the property. A property only one thread group reads is a plain field, with its name under the label. - User Defined Variables blocks (for example the recorder's "Settings - Sleeps" and "Settings - Host") have an On/Off switch, starting where the plan has it, and their values below. Switching a block is a change you can Reset, and presets save it.
- Properties lists every
${__P(name,default)}or${__property(name,,default)}your plan reads: the ones thread groups and variables use come first, with what uses them, then the rest (think times, a host name). - A field you don't touch keeps the plan's value. A field you change shows You set this and the plan's value, with Reset to go back. Values can't contain spaces.
- The total beside the heading ("Total of 150 users on 3 load generators") is users per load generator × load generators: each load generator runs the whole plan.
The values are saved in presets, so a preset like "Smoke" can run 10 users and "Peak" 500. When you upload a new version of the plan, values you set are kept. If the plan's default changed, the field says so. A preset value for something the plan no longer has is not sent, and the field offers Remove from Preset.
AI Summary when the run ends
Tick AI Summary when the run ends to have a written AI Summary waiting on the results page when the test finishes, without coming back to ask for it. While the test runs, the results page says the summary is coming, with a Don't Summarize This Run link if you change your mind. The option is saved in presets, so a scheduled run that uses such a preset gets a summary every time. Each one counts toward the account's monthly AI Summary allowance, and the checkbox shows how many are left.
Choosing where it runs
Run On has two tabs. You only see the choices your account has, and no tabs when there is only one way.
| Number of Load Generators | Say how many: from your own load generators, the Atlas shared load generator (free plan), or a cluster. Atlas picks free ones; cluster units are created for the test and removed afterwards. |
| Specific Servers | Pick load generators by name. Switched-off machines can still be selected — they are powered on for you. |
The number of machines multiplies your load: a plan with 100 threads run on 5 generators applies 500 concurrent users to your system.
Thread counts are per machine
Atlas does not divide your thread count across machines — each one runs the plan as written. Five generators running a 100-thread plan is 500 users, not 100.
From the CLI
atlas run -p SHOP --jmx checkout.jmx --servers 5 --wait--servers 5 | use any 5 available load generators |
--server gen-01 --server gen-02 | pick specific ones by name |
--cluster my-cluster --units 10 | run on 10 units in a cluster |
--jmx <filename> | which uploaded plan to run (needed only if the project holds more than one) |
--args "<jmeter args>" | extra JMeter arguments |
--no-split | send whole data files to every machine |
--wait | follow the run and print the report when it finishes |
--timeout 30m | how long --wait waits before giving up (default 2h) |
While it runs
The Run Test tab becomes a live view showing each stage — powering on, deploying, running — with a Stop button.
Watch in the sidebar shows live per-machine statistics while a test is in progress, which is the quickest way to confirm every generator is actually producing load.
From the CLI:
atlas status SHOP-20260818-K3F9If a machine is busy
A load generator can only run one test at a time. If you ask for machines that are already in use, your run is queued and starts when they free up. You don't need to retry.
Stopping a test
Press Stop, or:
atlas stop SHOP-20260818-K3F9Atlas asks JMeter to shut down cleanly so the final results are recorded, and only reports the machine as stopped once the process has actually ended.
Tip
If a machine can't be reached when you stop a test, Atlas tells you rather than pretending it worked, and the run stays open. Retry once the machine is reachable — this prevents a test carrying on unnoticed in the background.
When a test ends
A test ends by itself when the plan says so — a thread group's duration or loop count running out. When it does, the machines are released and any that Atlas powered on are switched off.