Skip to content

Latest commit

 

History

History
130 lines (90 loc) · 11.5 KB

activity-linking.md

File metadata and controls

130 lines (90 loc) · 11.5 KB

Activity Linking

Activity linking is the act of connecting activities within a pipeline to each other. It includes not just linking pipeline steps to each other, but also linking hierarchically, such as an overall pipeline activity, a set of sub activities for a build or a test, or linking multiple complete pipelines to each other.

Activity links are important both for tracing what activity/activities preceded a certain activity or what activity/activities followed after a certain activity, but also for tracing a full chain of parallel/serial activities within a pipeline. Tracing the activities in pipelines is the base for various kinds of pipeline visualizations and also for metrics and KPI measurements.

Event Driven vs Orchestrated Pipeline

Pipeline executors can have different mechanisms for choosing when its activities should be triggered. We find it useful to define the following extremes:

A fully event driven pipeline in Eiffel terminology is a pipeline where all activities in it are triggered explicitly by Eiffel events. Those events could either come from earlier pipeline steps, or from other pipelines producing for example artifacts that this pipeline would be configured to trigger on.

A fully orchestrated pipeline is completely controlled by a dedicated pipeline orchestrator, such as Jenkins Pipeline or Argo Workflows, and has no activities triggered by Eiffel events. A fully orchestrated pipeline is probably often initiated by a source change in some SCM system, and that source change is then propagated to the pipeline orchestrator through some non-Eiffel-event channel (e.g. a Github web hook or a Gerrit stream-event)

None of the scenarios above is probably relevant for most of the Eiffel event users, but rather a combination of the two where a pipeline is often triggered by an Eiffel event, and an orchestrator deals with controlling (at least parts of) the pipeline. Such triggers could for example be SCM events (SCC/SCS) or artifact events (e.g. ArtC/ArtP/CLM).

To handle the different possible scenarios for pipeline execution, multiple link types are defined in the Eiffel protocol. Those link types are used for linking between activity events.

Link Types Involved

This section describes the main link types involved in linking activities in pipelines

CAUSE

Identifies a cause of the event occurring, in the situations where the cause is an Eiffel event. This link type is not relevant only for EiffelActivity*Events, but for any Eiffel event that represent an occurrence that was caused by an earlier Eiffel event.

Required: No
Legal sources: Any
Legal targets: Any
Multiple allowed: Yes

PRECURSOR

This link type is used to declare temporal relationships between activities in a pipeline, i.e. what other activity/activities preceded this activity.

The image below shows how an activity that precedes another activity is denoted by a PRECURSOR link from the second ActT event back to the first one.

A Simple PRECURSOR Example

An activity could have multiple PRECURSOR activities (also known as "fan-in"), which means that the activity was triggered in serial after a certain group of parallel activities were triggered. This can be used to declare merge points of a group of parallel activities into a succeeding activity.

The image below shows how three parallel activities are followed by one activity in serial. It is denoted by PRECURSOR links from the ActT event of the last activity to the ActT event of all preceding parallel activities.

A Parallel PRECURSOR Example

The fact that a certain activity is triggered, having some preceding activity/activities as PRECURSOR(s) does not in itself mean that the preceding activity/activities is/are finished before this activity was triggered. It merely means that it was triggered after that/those other activity/activities were triggered.

PRECURSOR links are only valid to be used on the triggering event of an activity, i.e. on EiffelActivityTriggeredEvent or on EiffelTestSuiteStartedEvent, and the target of this link type shall be an event of the same type.

This link type is relevant mostly to non event-triggered activities. It is though recommended to also use it for event-triggered activities, as it helps visualizing the full chain of activities in a pipeline in a common way regardless of how each activity was triggered. By always providing PRECURSOR links between activities, a visualization tool does not need to follow any CAUSE links in order to visualize the activity relationships.

For event links between two or more complete pipelines, e.g. when a source change in a repository is automatically created by an update to an upstream dependency for that repository, there would be a CAUSE link to the event notifying that updated dependency. The protocol currently does not recommend to use a PRECURSOR link in that scenario, as the PRECURSOR link type is intended to be used between events of the same type.

Required: No
Legal sources: EiffelActivityTriggeredEvent, EiffelTestCaseTriggeredEvent, EiffelTestSuiteStartedEvent
Legal targets: EiffelActivityTriggeredEvent, EiffelTestCaseTriggeredEvent, EiffelTestSuiteStartedEvent
Multiple allowed: Yes

ORIGINAL_TRIGGER

This link type is used for sequences of (usually) failing activities where the first activity declares the "root" trigger and the subsequent activities are retries of the first activity.

The image below shows how a source code change triggers an activity. That activity fails and is retried, and the second activity (which might be manually triggered by a user) has an ORIGINAL_TRIGGER link to the first activity. If the second activity also fails and a third attempt takes place, that activity's ORIGINAL_TRIGGER link would also link to the first activity.

An example of using ORIGINAL_TRIGGER

To enhance clarity, an event could combine the ORIGINAL_TRIGGER with a CAUSE link. For example, an activity that retries failing builds based on events might use a CAUSE link to the ActF that announces the failure. To increase the fidelity of the model the event could also use an ORIGINAL_TRIGGER to the ActT to signal which event was the original trigger of the activity sequence.

Required: No
Legal sources: EiffelActivityTriggeredEvent, EiffelTestCaseTriggeredEvent, EiffelTestSuiteStartedEvent
Legal targets: EiffelActivityTriggeredEvent, EiffelTestCaseTriggeredEvent, EiffelTestSuiteStartedEvent
Multiple allowed: No

CONTEXT

This link type is used to declare hierarchies of activities within a pipeline. The CONTEXT identifies the activity or test suite of which this event constitutes a part. For example:

  • This pipeline step is executed in the CONTEXT of that pipeline, i.e. the pipeline step is part of a certain pipeline, and the ActT event of the pipeline step SHOULD have a CONTEXT link to the ActT event of the pipeline itself.
  • This test suite is executed in the CONTEXT of that pipeline step, i.e. the test suite is executed within a certain pipeline step, and the TSS event for the test suite SHOULD then have a CONTEXT link to the ActT event of the pipeline step.
  • This artifact was built within the CONTEXT of that pipeline step, i.e. the ArtC event SHOULD have a CONTEXT link to the ActT event of the pipeline step.

It is probably not relevant to provide a CONTEXT link to a test case execution, so therefore the Eiffel protocol does not allow EiffelTestCaseTriggeredEvent to be a legal target for this link type.

Required: No
Legal sources: Any
Legal targets: EiffelActivityTriggeredEvent, EiffelTestSuiteStartedEvent
Multiple allowed: No

FLOW_CONTEXT

This link type is not related to activity linking, but is mentioned here as it could be confused with the CONTEXT link type. A FLOW_CONTEXT link points at an EiffelFlowContextDefinedEvent which provides additional metadata about for example the program or track that the event is emitted for and thus doesn't describe relationships between activities.

Required: No
Legal sources: Any
Legal targets: EiffelFlowContextDefinedEvent
Multiple allowed: Yes

Why not use CAUSE instead of PRECURSOR?

The link type PRECURSOR is more recently added to the protocol than the CAUSE type, and the scenario described for the PRECURSOR type above was previously often described using CAUSE links instead. It is perfectly possible to create a fully traceable event graph using CAUSE links and not involving PRECURSOR links, but it has a number of drawbacks. For example:

  • The semantics of a CAUSE link declare why a certain event/occurrence took place. That is not really applicable to for example an orchestrated pipeline driven by for example Jenkins Pipeline or Argo Workflows, where the orchestrator determines why to start a certain activity rather than an activity being triggered by an earlier event. In other words, there is no direct causal relationship between an activity triggered by an orchestrator and a previous activity in the same pipeline triggered by the same orchestrator.
  • If a CAUSE link should be used to explicitly connect pipeline step activities, it would often be used between the ActT of a step to ActF or a previous step, but an activity could also be CAUSEd by a previous activity being triggered or started, or even by an internal occurrence within an activity. To deal with all scenarios an event consumer would need to consider all such combinations instead of just relying on links between ActT (or TSS) events.
  • When parallel pipeline steps are followed by a single pipeline step a CAUSE link could be used to all those parallel steps activity events, but it might be more natural to only have the CAUSE link to one of them, depending on the scenario. Adding a new link type (PRECURSOR) provides the possibility to connect these activity events without needing to use CAUSE links to tie activities together which do not actually have a causal relationship to each other.