Work items and work item types


Issues have the potential to be a centralized hub for collaboration. We need to accept the fact that different issue types require different fields and different context, depending on what job they are being used to accomplish. For example:

  • A bug needs to list steps to reproduce.
  • An incident needs references to stack traces and other contextual information relevant only to that incident.

Instead of each object type diverging into a separate model, we can standardize on an underlying common model that we can customize with the widgets (one or more attributes) it contains.

Here are some problems with current issues usage and why we are looking into work items:

  • Using labels to show issue types is cumbersome and makes reporting views more complex.
  • Issue types are one of the top two use cases of labels, so it makes sense to provide first class support for them.
  • Issues are starting to become cluttered as we add more capabilities to them, and they are not perfect:

    • There is no consistent pattern for how to surface relationships to other objects.
    • There is not a coherent interaction model across different types of issues because we use labels for this.
    • The various implementations of issue types lack flexibility and extensibility.
  • Epics, issues, requirements, and others all have similar but just subtle enough differences in common interactions that the user needs to hold a complicated mental model of how they each behave.
  • Issues are not extensible enough to support all of the emerging jobs they need to facilitate.
  • Codebase maintainability and feature development becomes a bigger challenge as we grow the Issue type beyond its core role of issue tracking into supporting the different work item types and handling logic and structure differences.
  • New functionality is typically implemented with first class objects that import behavior from issues via shared concerns. This leads to duplicated effort and ultimately small differences between common interactions. This leads to inconsistent UX.

Work item terminology

To avoid confusion and ensure communication is efficient, we will use the following terms exclusively when discussing work items.

Term Description Example of misuse Should be
work item type Classes of work item; for example: issue, requirement, test case, incident, or task Epics will eventually become issues Epics will eventually become a work item type
work item An instance of a work item type    
work item view The new frontend view that renders work items of any type    
legacy issue view The existing view used to render issues and incidents    
issue The existing issue model    
issuable Any model currently using the issuable module (issues, epics and MRs) Incidents are an issuable Incidents are a work item type
widget A UI element to present or allow interaction with specific work item data    

Some terms have been used in the past but have since become confusing and are now discouraged.

Term Description Example of misuse Should be
issue type A former way to refer to classes of work item Tasks are an issue type Tasks are a work item type

Migration strategy

WI model will be built on top of the existing Issue model and we’ll gradually migrate Issue model code to the WI model.

One way to approach it is:

class WorkItems::WorkItem < ApplicationRecord
  self.table_name = 'issues'

  # ... all the current issue.rb code

class Issue < WorkItems::WorkItem
  # Do not add code to this class add to WorkItems:WorkItem

We already use the concept of WITs within issues table through issue_type column. There are issue, incident, and test_case issue types. To extend this so that in future we can allow users to define custom WITs, we will move the issue_type to a separate table: work_item_types. The migration process of issue_type to work_item_types will involve creating the set of WITs for all root-level groups.

At first, defining a WIT will only be possible at the root-level group, which would then be inherited by subgroups. We will investigate the possibility of defining new WITs at subgroup levels at a later iteration.

Introducing work_item_types table

For example, suppose there are three root-level groups with IDs: 11, 12, and 13. Also, assume the following base types: issue: 0, incident: 1, test_case: 2.

The respective work_item_types records:

group_id base_type title
11 0 Issue
11 1 Incident
11 2 Test Case
12 0 Issue
12 1 Incident
12 2 Test Case
13 0 Issue
13 1 Incident
13 2 Test Case

What we will do to achieve this:

  1. Add a work_item_type_id column to the issues table.
  2. Ensure we write to both issues#issue_type and issues#work_item_type_id columns for new or updated issues.
  3. Backfill the work_item_type_id column to point to the work_item_types#id corresponding to issue’s project root groups. For example:

    issue.project.root_group.work_item_types.where(base_type: issue.issue_type)
  4. After issues#work_item_type_id is populated, we can switch our queries from using issue_type to using work_item_type_id.

To introduce a new WIT there are two options:

  • Follow the first step of the above process. We will still need to run a migration that adds a new WIT for all root-level groups to make the WIT available to all users. Besides a long-running migration, we’ll need to insert several million records to work_item_types. This might be unwanted for users that do not want or need additional WITs in their workflow.
  • Create an opt-in flow, so that the record in work_item_types for specific root-level group is created only when a customer opts in. However, this implies a lower discoverability of the newly introduced work item type.

Work item type widgets

A widget is a single component that can exist on a work item. This component can be used on one or many work item types and can be lightly customized at the point of implementation.

A widget contains both the frontend UI (if present) and the associated logic for presenting and managing any data used by the widget. There can be a one-to-many connection between the data model and widgets. It means there can be multiple widgets that use or manage the same data, and they could be present at the same time (for example, a read-only summary widget and an editable detail widget, or two widgets showing two different filtered views of the same model).

Widgets should be differentiated by their purpose. When possible, this purpose should be abstracted to the highest reasonable level to maximize reusability. For example, the widget for managing “tasks” was built as “child items”. Rather than managing one type of child, it’s abstracted up to managing any children.

All WITs will share the same pool of predefined widgets and will be customized by which widgets are active on a specific WIT. Every attribute (column or association) will become a widget with self-encapsulated functionality regardless of the WIT it belongs to. Because any WIT can have any widget, we only need to define which widget is active for a specific WIT. So, after switching the type of a specific work item, we display a different set of widgets.

Widgets metadata

In order to customize each WIT with corresponding active widgets we will need a data structure to map each WIT to specific widgets.

The intent is for work item types to be highly configurable, both by GitLab for implementing various work item schemes for customers (an opinionated GitLab workflow, or SAFe 5, etc), and eventually for customers to customize their own workflows.

In this case, a work item scheme would be defined as a set of types with certain characteristics (some widgets enabled, others not), such as an Epic, Story, Bug, and Task, etc.

As we’re building a new work item architecture, we want to build the ability to define these various types in a very flexible manner. Having GitLab use this system first (without introducing customer customization) allows us to better build out the initial system.

Work item’s base_type is used to define static mapping of what widgets are available for each type (current status), this definition should be rather stored in a database table. The exact structure of the WIT widgets metadata is still to be defined. base_type was added to help convert other types of resources (requirements and incidents) into work items. Eventually (when these resources become regular work items), base_type will be removed.

Until the architecture of WIT widgets is finalized, we are holding off on the creation of new work item types. If a new work item type is absolutely necessary, please reach out to a member of the Project Management Engineering Team.

Custom work item types

With the WIT widget metadata and the workflow around mapping WIT to specific widgets, we will be able to expose custom WITs to the users. Users will be able to create their own WITs and customize them with widgets from the predefined pool.

Custom widgets

The end goal is to allow users to define custom widgets and use these custom widgets on any WIT. But this is a much further iteration and requires additional investigation to determine both data and application architecture to be used.

Migrate requirements and epics to work item types

We’ll migrate requirements and epics into work item types, with their own set of widgets. To achieve that, we’ll migrate data to the issues table, and we’ll keep current requirements and epics tables to be used as proxies for old references to ensure backward compatibility with already existing references.

Migrate requirements to work item types

Currently Requirement attributes are a subset of Issue attributes, so the migration consists mainly of:

  • Data migration.
  • Keeping backwards compatibility at API levels.
  • Ensuring that old references continue to work.

The migration to a different underlying data structure should be seamless to the end user.

Migrate epics to work item types

Epic has some extra functionality that the Issue WIT does not currently have. So, migrating epics to a work item type requires providing feature parity between the current Epic object and WITs.

The main missing features are:

  • Get work items to the group level. This is dependent on Consolidate Groups and Projects initiative.
  • A hierarchy widget: the ability to structure work items into hierarchies.
  • Inherited date widget.

To avoid disrupting workflows for users who are already using epics, we will introduce a new WIT called Feature that will provide feature parity with epics at the project-level. Having that combined with progress on Consolidate Groups and Projects front will help us provide a smooth migration path of epics to WIT with minimal disruption to user workflow.

Work item, work item type, and widgets roadmap

We will move towards work items, work item types, and custom widgets (CW) in an iterative process. For a rough outline of the work ahead of us, see epic 6033.

Redis HLL Counter Schema

We need a more scalable Redis counter schema for work items that is inclusive of Plan xMAU, Project Management xMAU, Certify xMAU, and Product Planning xMAU. We cannot aggregate and dedupe events across features within a group or at the stage level with our current Redis slot schema.

All three Plan product groups will be using the same base object (work item). Each product group still needs to track MAU.

Proposed aggregate counter schema

graph TD Event[Specific Interaction Counter] --> AC[Aggregate Counters] AC --> Plan[Plan xMAU] AC --> PM[Project Management xMAU] AC --> PP[Product Planning xMAU] AC --> Cer[Certify xMAU] AC --> WI[Work Items Users]


The new aggregate schema is already implemented and we are already tracking work item unique actions in

For implementation details, this MR can be used as a reference. The MR covers the definition of new unique actions, event tracking in the code and also adding the new unique actions to the required aggregate counters.