The Benthos service offers a command
benthos test for running unit tests on sections of a configuration file. This makes it easy to protect your config files from regressions over time.
Let's imagine we have a configuration file
foo.yaml containing some processors:
One way to write our unit tests for this config is to accompany it with a file of the same name and extension but suffixed with
_benthos_test, which in this case would be
foo_benthos_test.yaml. We can generate an example definition for this config with
benthos test --generate ./foo.yaml which gives:
parallel instructs as to whether the tests listed in this definition should be executed in parallel. Under
tests we then have a list of any number of unit tests to execute for the config file.
Each test is run in complete isolation, including any resources defined by the config file. Tests should be allocated a unique
name that identifies the feature being tested.
target_processors is a JSON Pointer that identifies the specific processors within the file which should be executed by the test. This allows you to target a specific processor (
/pipeline/processors/0), or processors within a different section on your config (
/input/broker/inputs/0/processors) if required.
environment allows you to define an object of key/value pairs that set environment variables to be evaluated during the parsing of the target config file. These are unique to each test, allowing you to test different environment variable interpolation combinations.
When tests are run in parallel they will NOT retain their environment variables during execution. In order to retain custom environment variables ensure that
parallelis set to
input_batch lists one or more messages to be fed into the targeted processors as a batch. Each message of the batch may have its raw content defined as well as metadata key/value pairs.
For the common case where the messages are in JSON format, you can use
json_content instead of
content to specify the message structurally rather than verbatim.
output_batches lists any number of batches of messages which are expected to result from the target processors. Each batch lists any number of messages, each one defining
conditions to describe the expected contents of the message.
If the number of batches defined does not match the resulting number of batches the test will fail. If the number of messages defined in each batch does not match the number in the resulting batches the test will fail. If any condition of a message fails then the test fails.
Sometimes it's more convenient to define your tests within the config being tested. This is fine, simply add the
tests field to the end of the config being tested. When defining inline tests the field
parallel is not supported.
Sometimes when working with large Bloblang mappings it's preferred to have the full mapping in a separate file to your Benthos configuration. In this case it's possible to write unit tests that target and execute the mapping directly with the field
target_mapping, which when specified is interpreted as either an absolute path or a path relative to the test definition file that points to a file containing only a Bloblang mapping.
For example, if we were to have a file
cities.blobl containing a mapping:
We can accompany it with a test file
cities_test.yaml containing a regular test definition:
And execute this test the same way we execute other Benthos tests (
benthos test ./dir/cities_test.yaml,
benthos test ./dir/..., etc).
Sometimes the number of tests you need to define in order to cover a config file is so vast that it's necessary to split them across multiple test definition files. This is possible but Benthos still requires a way to detect the configuration file being targeted by these fragmented test definition files, which we can do by prefixing our
target_processors field with the path of the target relative to the definition file.
The syntax of
target_processors in this case is a full JSON Pointer that should look something like
target.yaml#/pipeline/processors. For example, if we saved our test definition above in an arbitrary location like
./tests/first.yaml and wanted to target our original
foo.yaml config file, we could do that with the following:
Executes a Bloblang expression on a message, if the result is anything other than a boolean equalling
true the test fails.
Checks the full raw contents of a message against a value.
Checks whether the full raw contents of a message matches a regular expression (re2).
Checks a map of metadata keys to values against the metadata stored in the message. If there is a value mismatch between a key of the condition versus the message metadata this condition will fail.
Checks that both the message and the condition are valid JSON documents, and that they are structurally equivalent. Will ignore formatting and ordering differences.
You can also structure the condition content as YAML and it will be converted to the equivalent JSON document for testing:
Checks that both the message and the condition are valid JSON documents, and that the message is a superset of the condition.
Executing tests for a specific config can be done by pointing the subcommand
test at either the config to be tested or its test definition, e.g.
benthos test ./config.yaml and
benthos test ./config_benthos_test.yaml are equivalent.
In order to execute all tests of a directory simply point
test to that directory, e.g.
benthos test ./foo will execute all tests found in the directory
foo. In order to walk a directory tree and execute all tests found you can use the shortcut
benthos test ./... will execute all tests found in the current directory, any child directories, and so on.