Cette page n'est pas encore disponible en français, sa traduction est en cours.
Si vous avez des questions ou des retours sur notre projet de traduction actuel, n'hésitez pas à nous contacter.

CI Visibility is not available in the selected site ().


This guide describes how to programmatically set up pipeline executions in CI Visibility and defines the types of pipeline execution that CI Visibility supports.

This guide applies to pipelines created using the public CI Visibility Pipelines API. Integrations with other CI providers may vary.

Data model

Pipeline executions are modeled as traces, similar to an APM distributed trace, where spans represent the execution of different parts of the pipeline. The CI Visibility data model for representing pipeline executions consists of four levels:

Level NameDescription
Pipeline (required)The top-level root span that contains all other levels as children. It represents the overall execution of a pipeline from start to finish. This level is sometimes called build or workflow in some CI providers.
StageServes as a grouping of jobs under a user-defined name. Some CI providers do not have this level.
JobThe smallest unit of work where commands are executed. All tasks at this level should be performed on a single node.
StepIn some CI providers, this level represents a shell script or an action executed within a job.

When a pipeline, stage, job, or step finishes, execution data is sent to Datadog. To set up Pipeline Visibility, see the list of supported CI providers. If your CI provider or workflow is not supported, you can use the public API endpoint to send your pipeline executions to CI Visibility.

Example of a pipeline execution trace

Stages, jobs, and steps are expected to have the exact same pipeline name as their parent pipeline. In the case of a mismatch, some pipelines may be missing stage, job, and step information. For example, missing jobs in the job summary tables.

Pipeline unique IDs

All pipeline executions within a level must have an unique identifier. For example, a pipeline and a job may have the same unique ID, but not two pipelines.

When sending repeated IDs with different timestamps, the user interface may exhibit undesirable behavior. For example, flame graphs may display span tags from a different pipeline execution. If duplicate IDs with the same timestamps are sent, only the values of the last pipeline execution received are stored.

Pipeline execution types

Normal run

The normal run of a pipeline execution follows the flow depicted below:

Depiction of a normal pipeline execution

Depending on the provider, some levels may be missing. For example, stages may not exist, and jobs may run in parallel or sequence, or a combination of both.

After the completion of each component, a payload must be sent to Datadog with all the necessary data to represent the execution. Datadog processes this data, stores it as a pipeline event, and displays it in CI Visibility. Pipeline executions must end before sending them to Datadog.

Full retries

Full retries of a pipeline must have different pipeline unique IDs.

In the public API endpoint, you can populate the previous_attempt field to link to previous retries. Retries are treated as separate pipeline executions in Datadog, and the start and end time should only encompass that retry.

Partial retries

When retrying a subset of jobs within a pipeline, you must send a new pipeline event with a new pipeline unique ID. The payload for any new jobs must be linked to the new pipeline unique ID. To link them to the previous retry, add the previous_attempt field.

Partial retries are treated as separate pipelines as well. The start and end time must not include the time of the original retry. For a partial retry, do not send payloads for jobs that ran in the previous attempt. Also, set the partial_retry field to true on partial retries to exclude them from aggregation when calculating run times.

For example, a pipeline named P has three different jobs, namely J1, J2, and J3, executed sequentially. In the first run of P, only J1 and J2 are executed, and J2 fails.

Therefore, you need to send a total of three payloads:

  1. Job payload for J1, with ID J1_1 and pipeline ID P_1.
  2. Job payload for J2, with ID J2_1 and pipeline ID P_1.
  3. Pipeline payload for P, with ID P_1.

Suppose there is a partial retry of the pipeline starting from J2, where all the remaining jobs succeed.

You need to send three additional payloads:

  1. Job payload for J2, with ID J2_2 and pipeline ID P_2.
  2. Job payload for J3, with ID J3_1 and pipeline ID P_2.
  3. Pipeline payload for P, with ID P_2.

The actual values of the IDs are not important. What matters is that they are correctly modified based on the pipeline run as specified above.

Blocked pipelines

If a pipeline is indefinitely blocked due to requiring manual intervention, a pipeline event payload must be sent as soon as the pipeline reaches the blocked state. The pipeline status must be set to blocked.

Flow of a blocked pipeline execution

The remaining pipeline data must be sent in separate payloads with a different pipeline unique ID. In the second pipeline, you can set is_resumed to true to signal that the execution was resumed from a blocked pipeline.

Downstream pipelines

If a pipeline is triggered as a child of another pipeline, the parent_pipeline field should be populated with details of the parent pipeline.

Manual pipelines

If a pipeline is triggered manually, the is_manual field must be set to true.

Git information

Providing Git information of the commit that triggered the pipeline execution is strongly encouraged. Pipeline executions without Git information don’t appear in the My Recent Commits page. At minimum, the repository URL, commit SHA, and author email are required. For more information see the public API endpoint specification.

Further reading

Documentation, liens et articles supplémentaires utiles: