ISSUE_TRIAGE_PROCESS.rst

Issue reporting and resolution process

This document explains the issue tracking and triage process within Apache Airflow including labels, milestones, and priorities as well as the process of resolving issues.

An unusual element of the Apache Airflow project is that you can open a PR to fix an issue or make an enhancement, without needing to open an issue first. This is intended to make it as easy as possible to contribute to the project.

Labels

Since Apache Airflow uses GitHub Issues as the issue tracking system, the use of labels is extensive. Though issue labels tend to change over time based on components within the project, the majority of the ones listed below should stand the test of time.

The intention with the use of labels with the Apache Airflow project is that they should ideally be non-temporal in nature and primarily used to indicate the following elements:

Kind

The "kind" labels indicate "what kind of issue it is". The most commonly used "kind" labels are: bug, feature, documentation, or task.

Therefore, when reporting an issue, the label of kind:bug is to indicate a problem with the functionality, whereas the label of kind:feature is a desire to extend the functionality.

There has been discussion within the project about whether to separate the desire for "new features" from "enhancements to existing features", but in practice most "feature requests" are actually enhancement requests, so we decided to combine them both into kind:feature.

The kind:task is used to categorize issues which are identified elements of work to be done, primarily as part of a larger change to be done as part of an AIP or something which needs to be cleaned up in the project.

Issues of kind:documentation are for changes which need to be made to the documentation within the project.

Area

The "area" set of labels should indicate the component of the code referenced by the issue. At a high level, the biggest areas of the project are: Airflow Core and Airflow Providers, which are referenced by area:core and area:providers. This is especially important since these are now being released and versioned independently.

There are more detailed areas of the Core Airflow project such as Scheduler, Webserver, API, UI, Logging, and Kubernetes, which are all conceptually under the "Airflow Core" area of the project.

Similarly within Airflow Providers, the larger providers such as Apache, AWS, Azure, and Google who have many hooks and operators within them, have labels directly associated with them such as provider/Apache, provider/AWS, provider/Azure, and provider/Google. These make it easier for developers working on a single provider to track issues for that provider.

Most issues need a combination of "kind" and "area" labels to be actionable. For example:

• Feature request for an additional API would have kind:feature and area:API
• Bug report on the User Interface would have kind:bug and area:UI
• Documentation request on the Kubernetes Executor, would have kind:documentation and area:kubernetes

Response to issues

Once an issue has been created on the Airflow project, someone from the Airflow team or the Airflow community typically responds to this issue. This response can have multiple elements.

Priority

After significant discussion about the different priority schemes currently being used across various projects, we decided to use a priority scheme based on the Kubernetes project, since the team felt it was easier for people to understand.

Therefore, the priority labels used are:

• priority:critical: Showstopper bug that should be resolved immediately and a patch issued as soon as possible. Typically, this is because it affects most users and would take down production systems.
• priority:high: A high priority bug that affects many users and should be resolved quickly, but can wait for the next scheduled patch release.
• priority:medium: A bug that should be fixed before the next release, but would not block a release if found during the release process.
• priority:low: A bug with a simple workaround or a nuisance that does not stop mainstream functionality.

It's important to use priority labels effectively so we can triage incoming issues appropriately and make sure that when we release a new version of Airflow, we can ship a release confident that there are no "production blocker" issues in it.

This applies to both Core Airflow as well as the Airflow Providers. With the separation of the Providers release from Core Airflow, a priority:critical bug in a single provider could trigger an unplanned patch release of the Airflow Providers.

Milestones

The key temporal element in the issue triage process is the concept of milestones. This is critical for release management purposes and will be used represent upcoming release targets.

Issues currently being resolved will get assigned to one of the upcoming releases. For example a feature request may be targeted for the next feature release milestone such as 2.x, where a bug may be targeted for the next patch release milestone such as 2.x.y.

In the interest of being precise, when an issue is tagged with a milestone, it represents that it will be considered for that release, not that it is committed to a release. Once a PR is created to fix that issue and when that PR is tagged with a milestone, it implies that the PR is intended to released in that milestone.

Please note that Airflow Core and Airflow Providers are now released and versioned separately. The use of milestones as described above is directed towards Airflow Core releases.

Transient Labels

Sometimes, there is more information needed to either understand the issue or to be able to reproduce the issue. Typically, this may require a response to the issue creator asking for more information, with the issue then being tagged with the label pending-response. Also, during this stage, additional labels may be added to the issue to help classification and triage, such as reported_version and area.

Occasionally an issue may require a larger discussion among the Airflow PMC or the developer mailing list. This issue may then be tagged with the needs:discussion label.

Some issues may need a detailed review by one of the core committers of the project and this could be tagged with a needs:triage label.

Good First Issue

Issues which are relatively straight forward to solve, will be tagged with the GoodFirstIssue label.

The intention here is to galvanize contributions from new and inexperienced contributors who are looking to contribute to the project. This has been successful in other open source projects and early signs are that this has been helpful in the Airflow project as well.

Ideally, these issues only require one or two files to be changed. The intention here is that incremental changes to existing files are a lot easier for a new contributor as compared to adding something completely new.

Another possibility here is to add "how to fix" in the comments of such issues, so that new contributors have a running start when then pick up these issues.

Timeliness

For the sake of quick responses, the general "soft" rule within the Airflow project is that if there is no assignee, anyone can take an issue to solve.

However, this depends on timely resolution of the issue by the assignee. The expectation is as follows:

• If there is no activity on the issue for 2 weeks, the assignee will be reminded about the issue and asked if they are still working on it.
• If there is no activity even after 1 more week, the issue will be unassigned, so that someone else can pick it up and work on it.

There is a similar process when additional information is requested from the issue creator. After the pending-response label has been assigned, if there is no further information for a period of 1 month, the issue will be automatically closed.

Invalidity

At times issues are marked as invalid and later closed because of one of the following situations:

• The issue is a duplicate of an already reported issue. In such cases, the latter issue is marked as duplicate.
• Despite attempts to reproduce the issue to resolve it, the issue cannot be reproduced by the Airflow team based on the given information. In such cases, the issue is marked as Can't Reproduce.
• In some cases, the original creator realizes that the issue was incorrectly reported and then marks it as invalid. Also, a committer could mark it as invalid if the issue being reported is for an unsupported operation or environment.
• In some cases, the issue may be legitimate, but may not be addressed in the short to medium term based on current project priorities or because this will be irrelevant because of an upcoming change. The committer could mark this as wontfix to set expectations that it won't be directly addressed in the near term.