We value your privacy
We use necessary cookies to make Loadero work. We ask for your permission to set additional cookies to understand site usage, make site improvements and to remember your settings. We also use cookies set by other sites to help deliver content from their services. See Cookie Policy for more info.
Skip to main content

Performance metric assertion

Loadero allows setting up assertions (also referred to as asserts) for metrics gathered during the test run. Asserts are automatically checked after the Selenium script execution concludes and the result will be reflected in the success status of the participants. If any of the set up asserts fails for a participant, then that participant's status will be set to "Fail" even if script execution itself was successful. Metric assertions can be configured during test creation or when editing the test by navigating to the "Asserts" tab, or via the API.

Script failure vs. assert failure

The header of any participant's report has two cells - "Participant result" and "Selenium result". If "Selenium result" is "Pass", but "Participant result" is "Fail", then it means that the participant failed because of an unsatisfied assert, but the script execution itself was successful.

Assert structure​

This section details a full description of an assert's structure with the assumption that it is being set up via the test builder UI in the Loadero web application. Asserts can also be managed via the API.

Statistic types​

Asserts are divided into four categories based on the metric type that they assert:

Measurements​

After selecting the statistic type, the next step in assert creation is the selection of the specific measurement that will be asserted. These will depend on the statistic selected in the previous step. For example, the FPS (frames per second) measurement will be classified under webrtc/video, bitrate - under machine/network, etc. All available measurements that can be asserted for each statistic are covered in the respective statistic's section.

Direction​

Some measurements will ask to specify whether to assert incoming or outgoing metric values (toggling incoming by default). This is denoted by a toggleable switch with "In" and "Out" values.

The incoming direction refers to metrics regarding what the participant is receiving from other participants, e.g., what FPS the video stream of another participant is, how many bytes are being received from other participants, etc.

The outgoing direction refers to metrics regarding what the participant is sending to other participants, e.g., how many FPS the participant's own video stream has, how many bytes the participant itself is transmitting to others.

The direction component of an assert is specific only to machine network statistics and WebRTC statistics. Other statistics do not have an incoming/outgoing split.

info

It tends to be the case that your own video is clearer and more crisp than what the other person receives. As such, it is somewhat common for two identical participants in a test to have worse incoming metrics than outgoing metrics.

Metric aggregators​

A metric can fluctuate a lot during the test and in order to assert the general behavior of a metric throughout the test (such as incoming FPS during a call that lasts an hour), it is necessary to summarize the metric values into a single number. This is achieved through the use of metric aggregators, which are listed in the table below.

The abbreviation column displays what the aggregator is called in the app.

AggregatorAbbreviation
1st percentile1st
5th percentile5th
25th percentile25th
50th percentile (median)50th
75th percentile75th
95th percentile95th
99th percentile99th
average (arithmetic mean)avg
minimummin
maximummax
sumtotal
standard deviationstddev
relative standard deviationrstddev

The aggregators listed are not available for every single available measurement. For example, percentile metrics are not available for measurements like machine/cpu/available because this measurement is static all throughout the test and asserting a percentile for it would be pointless. For the same reason some measurements do not have aggregators available at all.

Relative standard deviation

Relative standard deviation gives more context by reflecting the offset keeping the scale of value. For example, if standard deviation is only 1 byte it does not give any indications about the significance of fluctuations. But by using relative standard deviation it would return fluctuation in percentage form relative to the average value, immediately giving perspective on the fluctuations without the need to check the average value.

An additional aggregator unique to packetsLost measurements is the percent of lost packets relative to the total amount, abbreviated as percent during assert creation. This aggregator will return a value between 0 and 100.

Negative packet loss

If packets are lost, then they are retried by WebRTC which can result in a smaller packet loss since the previously presumed-as-lost packets end up getting received after all. Packet loss can even be reported as negative because of these duplicated, retried packets (where both the original packet is received, as well as the duplicated one). This is a peculiarity of how WebRTC itself works and is not a bug. You can read more about it in the RTP standard, specifically the "cumulative number of packets lost" section.

Operators and the expected value​

After selecting an aggregator (or sometimes after selecting a measurement, as some measurements won't have aggregators at all), you will be prompted to select an operator. The table below lists all the operators available and how to identify them while creating an assert in the app's test builder.

If you would like to manage asserts via the API, the notation for that is given in the API usage section.

OperatorSymbol in the test builder
equal to==
not equal to!=
less than<
less than or equal to<=
greater than>
greater than or equal to>=
regular expressionRegex

The selected operator defines the nature of the comparison that will take place between the actual value that is measured during the test run and the value that was set up for comparison.

The final step in assert creation is defining exactly that value to compare against. If an assert's expected value type does not match the type of the actual measurement provided, then the assert will always be ignored after a run's execution. Make sure to provide a numeric expected value if the metric itself is numeric. For example, machine/cpu/used/max > ten would be skipped, as the metric is numeric, but the provided value to compare against is a string.

When using the Regex operator, the assert will check whether the specified metric matches the regular expression provided in the "Expected value" field.

Action statistics​

Action statistics measure metrics such as duration for different actions performed in the Selenium script. Such actions are defined within the Selenium script. Currently there is only one action statistic available in Loadero - timeExecution. This corresponds to measurements taken when a Time Execution command is executed in the test run.

The first thing that sets these actions/timeExecution statistics apart from all others is the fact that instead of choosing one of several pre-defined measurement types, you are instead prompted to input the name of the timecard. A timecard is what is generated as a result of the Time Execution command, and the name that needs to be provided in the assert corresponds to the name parameter that had been provided to the command when calling it in the script.

For example, if you have called the command in the test script and provided the name locate_search_bar like in the example shown below, then a timecard will be generated linking the name to a duration. By creating an actions/timeExecution/locate_search_bar assert, the timecard with that same name will be identified and the duration will be fetched for comparison.

client => {
    const reallyLongPause = 5 * 60 * 1000; // 5 minutes

    client
        .url("https://duckduckgo.com/")
        .timeExecution("locate_search_bar", () => {
            client
                .waitForElementVisible("#searchbox_input", 10 * 1000)
                .sendKeys("#searchbox_input", "QA Processes")
                .click("[aria-label='Search']")
                .waitForElementVisible("#r1-0 > div > h2", 10 * 1000)
                .pause(reallyLongPause);
        })
        .takeScreenshot("screenshot.png");
}

Another peculiarity with the actions/timeExecution statistics is that their measurements are not aggregable. Even if you call the Time Execution command with the same name multiple times, each one will generate a separate timecard and each timecard will be asserted against. E.g., if you call the Time Execution command with the name locate_search_bar five times, then each of those five will be asserted. You cannot assert the average, maximum, etc. of all five instances of this name.

info

If an assert is created for a timecard, but that timecard never gets generated during the test run, then the assert's result will be "Skipped". Skipped asserts do not contribute to the final decision of whether the participant has a PASS or FAIL status.

Machine statistics​

Machine statistics measure metrics regarding the hardware that the participants are executing the script on. There are three types of machine statistics - cpu, ram, and network.

All of the machine statistic measurements that can be asserted are provided in the table below. The machine prefix has been removed from the first column for brevity, as it is there for every single machine statistic measurement.

Measurements marked as static have an unchanging value throughout the test. Static machine statistic measurements have only one aggregator available - total - which actually aggregates just that one single static value. E.g., cpu/available/total for a G4 participant will be 200%, because this measurement is not actually monitored in time, but rather just measured once at the beginning of the test, since the amount of available CPU does not change.

MeasurementUnitDescription
cpu/availablepercentageStatic. Total CPU power available
cpu/usedpercentageTotal CPU power used
cpu/percentpercentageTotal CPU power used relative to CPU power available
ram/availablemegabytesStatic. Total RAM available
ram/usedmegabytesTotal RAM used
ram/percentpercentageTotal RAM used relative to RAM available
network/bitratekilobits per secondNetwork bitrate
network/byteskilobytesTotal amount of network traffic
network/errorscountNumber of network errors
network/packetspackets per secondNumber of network packets
network/packetsLostcountNumber of lost network packets

Network metrics also have a further split into outgoing and incoming metrics. You can read about what this split means here.

info

The used measurements for CPU and RAM detail the absolute value, whereas percent describes the usage relative to the amount that was actually available. For example, if a participant is G4, its cpu/available/total will be 200%. Let's say that the cpu/used/avg value is 175%. In this case cpu/percent/avg will be 175 / 200, or 87.5% CPU usage out of the total CPU that had been available.

The used measurement can exceed 100%, as it refers to the absolute utilization of CPU cores - if there are two CPU cores available, then used can go up to 200%. On the other hand, the percent measurement will always be within the range of 0-100%.

MOS (Mean Opinion Score) statistics​

MOS (Mean Opinion Score) statistics measure the overall quality of audio or video during a test. This can be achieved through the use of various MOS calculation algorithms

Currently Loadero supports MOS calculation only for audio, so mos/audio is the only MOS statistic type. The mos/audio statistic also has only one available measurement type - e-model, as it is currently the only audio MOS calculation algorithm available in Loadero.

E-model MOS is calculated at the end of the test from the averages of network metrics that had been gathered throughout the whole test duration, meaning the algorithm's score is a reflection of what the audio quality was on average. This is why the only aggregator available for this measurement is avg.

WebRTC statistics​

WebRTC statistics measure various aspects of audio and video streams during the test as denoted in the WebRTC protocol.

There are two types of WebRTC statistics - audio and video. All of the WebRTC statistic measurements that can be asserted are provided in the tables below. Multiple of the measurements are available for both audio and video statistics, so the available measurements have been separated into three tables:

caution

WebRTC statistics usually have many tracks being sent and received at the same time. However, Loadero currently doesn’t assert each track individually, meaning you can’t check that the average value of each individual track meets a threshold, but rather that all tracks combined meet that threshold. Let’s look at an example. Assume that there are 2 tracks, both of which receive data for the same amount of time, but the first track receives 50kbps constantly, while the second track receives 150kbps constantly. Individually the average bitrates would be 50kbps and 150kbps respectively, however after combining the average bitrate will be reported as 100kbps.

The original WebRTC dump is available for download to perform customized assertions.

info

If an assert has been created for a measurement that never retrieved any metrics during the test, that assert will be set to the "Skipped" status and it will not have any impact on whether the participant gets a "Fail" or "Pass" status. Some reasons where such assert skipping may occur are:

  • Mozilla Firefox simply does not provide metrics for some audio measurements.
  • The test run never started a WebRTC call, so there are no metrics for WebRTC measurements.

Skipped asserts do not affect the final Pass/Fail result of the participant. It is as if the assert was never created at all. If you would like to make sure asserts fail if WebRTC connections failed to be established, make sure to set up an assert for the connections measurement to be greater than 0.

Measurements for both audio and video​

MeasurementUnitDescription
bitratekilobits per secondActual bitrate of media
byteskilobytesAmount of data conveyed throughout the whole test
codecstring valueCodec of the stream
connectionscountNumber of WebRTC connections established
jittermillisecondsDuration of media jitter
jitterBuffermillisecondsSize of jitter buffer
packetspackets per secondNumber of network packets
packetsLostcountNumber of lost network packets
rttmillisecondsDuration of round-trip time
tip

The bitrate measurement tracks how the metric changes every second. If you would like to assert the total amount of data transmitted throughout the entire test duration, a bytes/total assert is what you're looking for.

The following measurements are not aggregable:

  • codec - unchanging in time, as well as the fact it's a string value
  • connections - measurement is not monitored in time, but rather the number you see is already the total amount of connections established throughout the test

Audio-specific measurements​

MeasurementDescription
levelValue in the range 0..1 (linear), where 1.0 represents 0 dBov, 0 represents silence, and 0.5 represents approximately 6 dBSPL change in the sound pressure level from 0 dBov.

Video-specific measurements​

MeasurementUnitDescription
fpscount per secondNumber of frames encoded/decoded per second
frameWidthpixelsFrame width
frameHeightpixelsFrame height

Conditional assert execution​

It is possible to define conditions that must be met in order for an assert to be executed. If this condition is not met, then the assert will be skipped for the given participant. This is useful in situations where you want something to be asserted for one participant but not for another. This is achieved via preconditions.

When a participant finishes test execution, it will automatically go through all asserts and report on whether these asserts have passed or failed according to the metrics gathered throughout the test run. If a precondition is identified for the assert, then a check is performed for whether this precondition is met. If it is not met, then the assert is skipped.

info

One example use case is a test where you have both participants with default network settings and participants with impaired network conditions. The expected network bitrate threshold is likely to be different for each type of participant, and you can define these thresholds separately by creating two identical asserts, just with different expected values and different preconditions.

Preconditions can be managed for an assert while creating or editing an assert in the test builder. Preconditions consist of three parts: the property, operator, and value. Once a participant finishes execution and begins iterating over all asserts, it will check whether its properties match what is specified in the precondition. Examples:

  • A participant using G4 compute units finishes test execution. While iterating over asserts, a machine/network/out/packetsLost/total == 0 assert is found. This assert has the precondition "Compute units == G4". This participant's "Compute units" property really is equal to G4, therefore the precondition is met and the assert is executed.
  • The same participant using G4 compute units also finds another assert - machine/network/out/packetsLost/percent < 5%. This assert has the precondition "Compute units == G1". This participant's "Compute units" property is not equal to G1, therefore the assert is skipped.
info

Skipped asserts do not contribute to the final decision of whether the participant has a PASS or FAIL status. It is as if the assert had never been set up at all.

Available properties​

If setting up preconditions via the Loadero API, the syntax for some properties is very particular. In the list above most properties link to the specific page describing what these properties are. For the audio feed, compute units, location, network, and video feed properties refer to the "API identifier" column in the table of the linked page. For the browser property, refer to the "API usage" section of the "Browsers" page. The remaining properties are far more free-form.

On group and participant numbers

When the test begins execution, each group is assigned a group number. For each group defined for the test, a set of run groups will be created, as many as the count parameter specifies. Each of these run groups has a group number which indicates the order in which these run groups began execution.

For example, let's say a test has two groups A and B - A has count=4, and B has count=2 and the test is set to use the "Linear group" strategy. Group A will spawn four run groups. The first run group will have the group number 0, the second - 1, the third - 2, the fourth - 3. Group B will spawn two run groups. The first run group will have the group number 4, the second - 5. If you were to create a precondition Group number == 0 for an assert, then this assert would be executed only for those participants who belonged to the first run group that began execution.

Participant number logic differs slightly. While groups are numbered in the order they began the test, participants are numbered in the order they began the the test within their own group. E.g., if there are two groups, the first participant of each group will have the participant number 0.

Precondition chaining​

If multiple preconditions are set for a single assert, then they are chained using the logical AND operator. In other words, that assert will be executed only if all of its set preconditions were fulfilled. For example, if the assert should be executed only when the participant's location is not equal to "AP East - Hong Kong" AND when the browser is "Latest Mozilla Firefox", your assert would look like the image below.

An assert with two preconditions

To implement chaining with the logical OR operator instead (execute the assert if AT LEAST one precondition is met), it is necessary to create duplicate asserts - one for each precondition. For example, if the assert should be executed only when the participant's location is not equal to "AP East - Hong Kong" OR the browser is "Latest Mozilla Firefox", you would have to create two asserts like the image below.

Two identical asserts with a different precondition

info

If a participant meets both preconditions, then each of the two asserts will be executed, regardless of the fact that they are both identical. Each of these asserts' result will be visible in the result report, despite that result being the same for all duplicates.

caution

Pay attention when chaining multiple preconditions with the AND operator from the same property field, e.g., Browser. In many cases such preconditions can be contradicting.

For example, chaining two preconditions via the AND operator that the browser must be a version of Chrome and that the browser must be a version of Firefox will always result in a skipped assert.

API usage​

When using the Loadero API to manage asserts, the assert is composed of the metric path, operator, and expected value. The operator and expected value fields mirror their field counterparts in test builder assert creation, however, the metric path is a composite string, consisting of the statistic, measurement, direction, and aggregator.

The operators are also defined differently when using the API, these definitions are provided in the table in this section.

Metric path​

As mentioned above, the metric path consists of the statistic, measurement, direction, and aggregator. If one of these components does not exist for an assert (most commonly - direction, occasionally - aggregator), it is simply skipped in the metric path.

Metric path construction generally follows the template of {statistic}/{measurement}/{direction}/{aggregator}. Below are listed some examples of how a metric path is constructed.

Metric path examples
  • Statistic: webrtc/audio
  • Measurement: level
  • Direction: in
  • Aggregator: 95th

These will form the metric path webrtc/audio/level/in/95th.


  • Statistic: machine/cpu
  • Measurement: used
  • Direction: this statistic type does not have a direction component
  • Aggregator: max

These will form the metric path machine/cpu/used/max.


  • Statistic: actions/timeExecution
  • Measurement: locate_search_bar timecard
  • Direction: this statistic type does not have a direction component
  • Aggregator: this statistic type does not have an aggregator component

These will form the metric path actions/timeExecution/locate_search_bar.

Operators in the API​

When defining an operator via the API, they should be referenced like in the table below. This is relevant both for the operator of an assert and the operator of a precondition.

OperatorHow it is defined in the API
equal toeq
not equal toneq
less thanlt
less than or equal tolte
greater thangt
greater than or equal togte
regular expressionregex