Assertions ¶

Assertions support profile comparison as well to assert the performance evolution of your code:

1
2
3
4
5
6

tests:
    "Pages should not become slower":
        path: "/.*"
        assertions:
            - "percent(main.wall_time) < 10%"       # time does not increase by more than 10%
            - "diff(metrics.sql.queries.count) < 2" # less than 2 additional SQL statements

Blackfire exposes metrics that are associated with the current profile. The value of one dimension of a metric can be used in assertions; for instance, the count value of the sql.queries metric is stored in metrics.sql.queries.count.

Learn more about all built-in metrics exposed in assertions by default and how to create your own.

The available dimensions for metrics are the following ones:

• count
• wall_time
• cpu_time
• memory
• peak_memory
• network_in
• network_out
• io

For each metric (count, time, or memory), a default unit is defined:

• The default time unit is the millisecond;
• The default memory unit is the byte.

When using 10 in an assertion for a time value, that evaluates to 10 milliseconds. It is possible to add an explicit unit to a metric, which will override the default unit.

The following time units are supported: ms, s.

The following memory units are supported: kb, kib, mb, mib, gb, gib.

The following generic units are supported: k, ki, m, mi, g, gi.

The following operators are supported in assertions:

Variables are useful when you configure several Blackfire environments to run performance tests on various machines hosting the same application. Variables make it possible to use the same .blackfire.yaml configuration on several servers that have different purposes (e.g. development, staging, production).

Variables can be defined in the environment configuration and used in an expression by passing the metric key to the var() function. Variable values may also use units.

Example: You have configured two different environments in Blackfire:

• Integration hits your integration server, where your application runs with debug mode on;
• Production hits your production servers, where your application runs with debug mode off.

The same application runs on all servers, but the debug mode increases memory usage a lot. As such, you cannot set the same maximum value for your assertions on memory.

Your .blackfire.yaml would look like this when using variables:

1
2
3
4
5

tests:
    "Pages shouldn't use too much memory":
        path: "/.*"
        assertions:
            - "main.peak_memory < 10mb * var('memory_coeff')"

Now in both your Integration and Production Blackfire environments, you have to create the memory_coeff variable:

• For Integration, memory_coeff == 2
• For Production, memory_coeff == 1

The var() function may also receive a default value as a second argument. This default value is used whenever the variable is not defined in one of your environments.

1
2
3
4
5

tests:
    "Pages shouldn't use too much memory":
        path: "/.*"
        assertions:
            - "main.peak_memory < 10mb * var('memory_coeff', 1)"

An optional description can be added to an assertion. This could provide a context or an educational note left for the other developers of your team.

This note can help understand the performance challenges of some parts of the application.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

tests:
    "The autoloader classmap should be dumped":
        path: "/.*"
        assertions:
            - "metrics.composer.autoload.find_file.count == 0"
        description: |
            By default, the Composer autoloader runs relatively fast. However,
            due to the way PSR-4 and PSR-0 autoloading rules are set up, it
            needs to check the filesystem before resolving a classname
            conclusively.

            This slows things down quite a bit, but it is convenient in
            development environments because when you add a new class it can
            immediately be discovered/used without having to rebuild the
            autoloader configuration.

            https://getcomposer.org/doc/articles/autoloader-optimization.md

The description is displayed on the Assertions tab of a Profile:

And within Build reports:

When an assertion is run from an environment, the is_dev() function returns false when the environment is configured for production usage.

When using builds, it is possible to compare one build to another. This is useful when you want to validate a code merge (e.g. a pull-request) by triggering a build webhook or a Blackfire Player collection of scenarios. It can also ensure that your code doesn't have performance regression over time using periodic builds.

Comparison can be made within assertion expressions thanks to the percent() and diff() functions:

If you want to assert that a metric value does not increase by more than a given percentage between two builds, use the percent() function:

1

"percent(main.wall_time) < 10%"

You can also test the evolution of a metric value in absolute terms with the diff() function:

1

"diff(metrics.sql.queries.count) < 2"

In this example, the assertion checks that the profile from the newer build has less than 2 additional SQL statements compared to the previous one.