- Data Collector
- Database query
- High-level overview
Value stream analytics calculates the time between two arbitrary events recorded on domain objects and provides aggregated statistics about the duration.
For information on how to configure Value Stream Analytics in GitLab, see our analytics documentation.
During development, events occur that move issues and merge requests through different stages of progress until they are considered finished. These stages can be expressed with the
- Name: Development
- Start event: Issue created
- End event: Issue first mentioned in commit
Events are the smallest building blocks of the value stream analytics feature. A stage consists of two events:
These events play a key role in the duration calculation.
duration = end_event_time - start_event_time
To make the duration calculation flexible, each
Event is implemented as a separate class. They’re responsible for defining a timestamp expression that will be used in the calculation query.
There are a few methods that are required to be implemented, the
StageEvent base class describes them in great detail. The most important ones are:
object_type method defines which domain object will be queried for the calculation. Currently two models are allowed:
For the duration calculation the
timestamp_projection method will be used.
def timestamp_projection # your timestamp expression comes here end # event will use the issue creation time in the duration calculation def timestamp_projection Issue.arel_table[:created_at] end
COALESCE). Look at the existing event classes for examples.
In some cases, defining the
timestamp_projection method is not enough. The calculation query should know which table contains the timestamp expression. Each
Event class is responsible for making modifications to the calculation query to make the
timestamp_projection work. This usually means joining an additional table.
Example for joining the
issue_metrics table and using the
first_mentioned_in_commit_at column as the timestamp expression:
def object_type Issue end def timestamp_projection IssueMetrics.arel_table[:first_mentioned_in_commit_at] end def apply_query_customization(query) # in this case the query attribute will be based on the Issue model: `Issue.where(...)` query.joins(:metrics) end
Some start/end event pairs are not “compatible” with each other. For example:
- “Issue created” to “Merge Request created”: The event classes are defined on different domain models, the
object_typemethod is different.
- “Issue closed” to “Issue created”: Issue must be created first before it can be closed.
- “Issue closed” to “Issue closed”: Duration is always 0.
StageEvents module describes the allowed
end_event pairings (
PAIRING_RULES constant). If a new event is added, it needs to be registered in this module.
To add a new event:
- Add an entry in
ENUM_MAPPINGwith a unique number, it’ll be used in the
- Define which events are compatible with the event in the
Supported start/end event pairings:
Teams and organizations might define their own way of building software, thus stages can be completely different. For each stage, a parent object needs to be defined.
Currently supported parents:
- User navigates to the value stream analytics page.
- User selects a group.
- Backend loads the defined stages for the selected group.
- Additions and modifications to the stages will be persisted within the selected group only.
The original implementation of value stream analytics defined 7 stages. These stages are always available for each parent, however altering these stages is not possible. To make things efficient and reduce the number of records created, the default stages are expressed as in-memory objects (not persisted). When the user creates a custom stage for the first time, all the stages will be persisted. This behavior is implemented in the value stream analytics service objects. The reason for this was that we’d like to add the abilities to hide and order stages later on.
DataCollector is the central point where the data will be queried from the database. The class always operates on a single stage and consists of the following components:
- Responsible for composing the initial query.
- Deals with
Stagespecific configuration: events and their query customizations.
- Parameters coming from the UI: date ranges.
Median: Calculates the median duration for a stage using the query from
RecordsFetcher: Loads relevant records for a stage using the query from
Finderclasses to apply visibility rules.
DataForDurationChart: Loads calculated durations with the finish time (end event timestamp) for the scatterplot chart.
For a new calculation or a query, implement it as a new method call in the
Structure of the database query:
SELECT (customized by: Median or RecordsFetcher or DataForDurationChart) FROM OBJECT_TYPE (Issue or MergeRequest) INNER JOIN (several JOIN statements, depending on the events) WHERE (Filter by the PARENT model, example: filter Issues from Project A) (Date range filter based on the OBJECT_TYPE.created_at) (Check if the START_EVENT is earlier than END_EVENT, preventing negative duration)
Structure of the
SELECT statement for
SELECT (calculate median from START_EVENT_TIME-END_EVENT_TIME)
Structure of the
SELECT statement for
SELECT (START_EVENT_TIME-END_EVENT_TIME) as duration, END_EVENT.timestamp
- Rails Controller (
Analytics::CycleAnalyticsmodule): Value stream analytics exposes its data via JSON endpoints, implemented within the
analyticsworkspace. Configuring the stages are also implements JSON endpoints (CRUD).
- Services (
Stagerelated actions will be delegated to respective service objects.
- Models (
Analytics::CycleAnalyticsmodule): Models are used to persist the
- Feature classes (
- Responsible for composing queries and define feature specific business logic.
Since we have a lots of events and possible pairings, testing each pairing is not possible. The rule is to have at least one test case using an
Writing a test case for a stage using a new
Event can be challenging since data must be created for both events. To make this a bit simpler, each test case must be implemented in the
data_collector_spec.rb where the stage is tested through the
DataCollector. Each test case will be turned into multiple tests, covering the following cases:
- Different parents:
- Different calculations: