Test Writer use cases

Writing a LAVA job submission

Numerous changes have been made compared with the previous JSON submission format. There is no compatibility between the old JSON files and the new pipeline job submissions. There is no conversion tool and none will be supported. Each test job needs to be understood and redesigned. Compatibility has only been preserved inside the Lava Test Shell Definitions.

General principles

API

  1. The API here is still in development and changes may still be required. Sample jobs are available from the LAVA team and are updated regularly.

  2. Only certain deployment types, boot types and device types are currently supportable. These guidelines will be enlarged as support grows.

  3. The pipeline does not make assumptions and the only defaults are constrictive and only provided for admin reasons.

Validity checks

  1. Decisions about the validity of a test job submission are made as early as possible.

  2. Some validity checks will be done before a job submission is accepted

  3. Most will be completed before the job submission is scheduled.

  4. An invalid job submission will result in an Incomplete test job.

  5. Some validity checks can only be done after the test job has started, e.g. checks relating to downloaded files. These checks will result in a JobError.

Results

  1. All pipeline test jobs report results, whether a test shell is used or not.

  2. Visibility of test job results is determined solely by the job submission.

  3. Results are part of the test job and cannot be manually created.

  4. Result analysis is primarily a task for other engines, results can be exported in full but the principle emphasis is on data generation.

  5. Results are posted in real time, while the job is still running. This means that a later failure in the test job cannot cause a loss of results.

Job submission data

  1. There are three actions for all test jobs: deploy, boot and test.

  2. All scheduled submissions may only specify a device type, not a target. The device type is only for use by the scheduler and is ignored by the dispatcher. Locally, dispatchers only have configuration for the devices currently running test jobs.

  3. Default timeouts can be specified at the top of the file.

    See also

    Timeouts

  4. priority can be specified, the default is medium.

  5. Always check your YAML syntax

  6. The actions element in a pipeline job submission is a list of dictionaries - ensure that the line ends with a colon :, the next line needs to be indented and needs to start with a hyphen -.

  7. YAML supports comments using #, please use them liberally. Comments are not preserved in the database after submission.

  8. The new result views know about the deployment type and boot type, so the job_name can concentrate on the objective of the test, not the methods used in it. The job name will still need to exist as a file in the test shell and as a URL in the results, so underscores and hyphens need to be used in place of spaces.

Note

Timeouts are specified in human readable units, days, hours, minutes or seconds. Avoid specifying timeouts in smaller units when a larger unit is available: i.e. you should never use 120 seconds, always 2 minutes. Schema rules may be introduced to enforce this and your jobs could be rejected. The requested timeout and the actual duration of each action class within a test job is logged and excessive timeouts can be identified.

Writing a new TestJob

See Dispatcher Action Reference for details of the available actions and use the sample jobs as examples.

YAML syntax

Caution

Indenting is critically important to YAML. A valid YAML document can still render an object which lacks the structure required for a valid submission. The parser errors do tend to be cryptic but will at generally indicate the last tag encountered.

Always use an editor which shows the actual whitespace. Many text editors have syntax highlighting for YAML. However, syntax highlighting may not be sufficient to identify common YAML syntax errors.

Common YAML errors

- boot:
  method: u-boot

Using the Online YAML parser, this results in:

[
  {
    "boot": null,
    "method": "u-boot"
  }
]

Note how the entire boot block is loaded as a null. method is now out of place. It has been made into a new entry in the list of actions. The submission is trying to create a test job which does:

  1. deploy

  2. boot

  3. method

  4. test

The correct syntax is:

- boot:
    method: u-boot

Note how method is indented beneath boot instead of at the same level.

Using the parser, this results in:

[
  {
    "boot": {
      "method": "u-boot"
    }
  }
]

This now creates a submission which is trying to do:

  1. deploy

  2. boot

    • method

  3. test

Understanding available support

Devices to run pipeline jobs must be set as a pipeline device by the admin of the LAVA instance. Check for a tick mark in the Pipeline Device column of the device type overview. The instance itself must be enabled for pipeline usage - one indicator is that an updated instance has a Results section in the top level menu.

Understanding a TestJob

To convert an existing job to the pipeline, there are steps to be covered:

  1. Be explicit about the type of deployment and the type of boot

  2. Be explicit about the operating system inside any rootfs

  3. Start with an already working device type or job submission.

  4. Start with singlenode jobs, use of the multinode protocol can follow later.

  5. Drop details of submitting results

Instead of calling a “deploy_kernel” or “deploy_image” command and passing parameters, a pipeline job submission requires that the type of deployment and the type of boot is specified as part of a single deploy action which covers all devices and all jobs.

Equally, a pipeline job submission requires that assumptions are removed in favor of explicit settings. Just because a URL ends in .gz does not mean that the dispatcher will assume that gz decompression can be used - this must be specified or no decompression is done at all.

The pipeline will not assume anything about the operating system of a rootfs specified in a URL - the job submission will simply fail to validate and will be rejected.

Booting beaglebone-black with an nfsrootfs requires knowledge of how that device can use NFS - in this case, using tftp.

actions:
 - deploy:
     to: tftp
     kernel:
       url: https://images.validation.linaro.org/functional-test-images/bbb/zImage
     # nfsrootfs: file:///home/linaro/lava/nfsrootfs/jessie-rootfs.tar.gz
     nfsrootfs:
       url: https://images.validation.linaro.org/pipeline/debian-jessie-rootfs.tar.gz
       compression: gz
     dtb:
       url: https://images.validation.linaro.org/functional-test-images/bbb/am335x-bone.dtb

Note

the use of comments here allows the writer to flip between a remote image and a local test version of that image - this would be suitable for running directly on a local dispatcher.

The same deployment stanza can be used for any device which supports NFS using tftp, just changing the URLs.

To change this deployment to a ramdisk without NFS, still using TFTP, simply provide a ramdisk instead of an nfsrootfs:

actions:

 - deploy:
    to: tftp
    kernel:
      url: https://images.validation.linaro.org/functional-test-images/bbb/zImage
    ramdisk:
      url: https://images.validation.linaro.org/functional-test-images/common/linaro-image-minimal-initramfs-genericarmv7a.cpio.gz.u-boot
      compression: gz
    dtb:
      url: https://images.validation.linaro.org/functional-test-images/bbb/am335x-bone.dtb

Note

ramdisk-type must be explicitly set, despite the URL in this case happening to have a u-boot extension. This is not assumed. Without the ramdisk-type being set to u-boot in the job submission, the U-Boot header on the ramdisk would be mangled when the test definitions are applied, resulting in an invalid ramdisk.

Submissions using advanced features

Templating

Not all test jobs are written by hand and many Continuous Integration systems will generate test jobs for submission to LAVA using a templating system. For example, KernelCI.org uses jinja2 to convert the data used to build the kernel artifacts into test job submissions for LAVA V2.

Including YAML

It is also possible to include YAML directly into a V2 test job submission using the include: dictionary which takes3 a single string as the URL of a remote YAML file. This file will be downloaded and inserted into the test job YAML. Any existing values at this point of the file will be updated. Any new values from the included file will be added.

This feature can be used to include generic boilerplate into YAML files or to help insert metadata or other elements. The URL provided must be publicly accessible to the master during submission.