What prints the summary + and summary = lines during a JMeter CLI run?

## Where the lines come from They are printed by a **Summariser** test element - the same class the GUI tree calls **Generate Summary Results**. You almost never add it yourself, because `bin/jmeter.properties` ships ``` summariser.name=summary ``` **uncommented**, and `JMeter.runNonGui` reads that property: if it is non-empty it logs and prints `Creating summariser <summary>`, constructs a `Summariser` with that name. Where it goes depends on whether you asked for a result file: with `-l` it is handed to the `ResultCollector` JMeter creates for that file (`new ResultCollector(summariser)`), and the collector — attached to the root of the cloned test plan — forwards every result to it; without `-l` the Summariser is attached to the root directly. Either way the lines print, which is why they appear whether or not you asked for a `.jtl`. Clear the property and a CLI run prints no summary lines at all; rename it and the label at the start of each line changes with it. ## Reading one line ``` summary + 82 in 0:00:30 = 2.7/s Avg: 1518 Min: 1003 Max: 2020 Err: 0 (0.00%) Active: 5 Started: 5 Finished: 0 ``` - `summary` is the element's name, which is also the accumulator key. - `+` marks a **delta** line: everything since the previous report. `=` marks the **running total**. - `82 in 0:00:30 = 2.7/s` is samples, elapsed and the rate that works out to. - `Avg`, `Min` and `Max` are response times in milliseconds; `Err` is the error count and percentage. - `Active`, `Started` and `Finished` are thread counts, and they are appended **only to `+` lines** - a `=` line never carries them. There is no percentile anywhere in that format. The console summary is a liveness and error signal, not a distribution. ## The cadence, and why it can slip `summariser.interval` (default `30`, in seconds) sets the reporting period, and reports land on wall-clock boundaries so that two runs started at different moments still line up. But the check that decides whether to report lives inside `sampleOccurred`: the element only looks at the clock when a sample result is handed to it. The manual states the consequence directly - a summary line is written only if samples are being emitted, so the interval is not respected when nothing completes. That makes silence meaningful. During a long soak watched on a dashboard, a console that has printed nothing for two minutes is not a broken summariser; it means no sample has finished in that time, which is exactly the moment you want to look at the run. ## Three details that bite 1. **Adding your own Generate Summary Results named `summary` merges into the automatic one.** The element groups by name into a shared accumulator, so the totals become the sum of both scopes. The manual calls this a design choice, not a bug - name yours something else. 2. **`summariser.log` and `summariser.out` both default to `true`.** The line goes to `jmeter.log` and to standard output. Turn `summariser.out` off and a CI job's captured console loses it. 3. **`summariser.ignore_transaction_controller_sample_result` defaults to `true`.** Samples generated by a Transaction Controller are skipped, so the counts follow real samplers rather than doubling. The first and last delta lines are usually short: the first because JMeter is synchronising to the interval boundary, the last because a test rarely ends on one. The final `+` line is printed only if the delta holds samples and something was reported earlier.