Reviewing

• Checklist
• Expectations
• Backward Compatibility
• Code Of Conduct
• Dependencies
• Documentation
• Performance Testing
• Single Concern
• Readability
• Safe Code
• Security
• Testing

Checklist

Pull request reviews are required before merging code into Vector. This document will outline Vector's pull request review requirements. The following checklist should be used for all pull requests:

• Is the code addressing a single purpose? If not, the pull request should be broken up. (see Single Concern)
• Is the code readable and maintainable? If not, suggest ways to improve this. (see Readability)
• Is the code reasonably tested? If not, tests should be improved. (see Testing)
• Is code marked as unsafe? If so, verify that this is necessary. (see Safe Code)
• Is backward compatibility broken? If so, can it be avoided or deprecated? (see Backward compatibility)
• Have dependencies changed? (see Dependencies)
• Has the code been explicitly reviewed for security issues? Dependencies included. (see Security)
• Is there a risk of performance regressions? (see Performance Testing)
• Should documentation be adjusted to reflect any of these changes? (see Documentation)

For component changes, especially pull requests introducing new components, the following items should also be checked:

• Does it comply with the configuration spec?
• Does it comply with component spec?
• Does it comply with the instrumentation spec?

Checklist - new source

This checklist is specific for Vector's sources.

• Does the source handle metrics? If it does, the Datadog Origin Metadata function (sinks::datadog::metrics::encoder::source_type_to_service), which maps the source to the correct Service value, needs to be updated. If this source is an Agent role and thus is the true origin of it's metrics, this will need to be a follow-up PR by a member of the Vector team.

Checklist - new sink

This checklist is specific for Vector's sinks.

Logic

• Does it work? Do you understand what it is supposed to be doing?
• Does the retry logic make sense?
• Are the tests testing that the sink is emitting the correct metrics?
• Are there integration tests?

Code structure

• Is it using the sink prelude (use crate::sinks::prelude::*)?
• Is the sink a stream based sink? Check that the return value from SinkConfig::build is the return from VectorSink::from_event_streamsink.
• Is it gated by sensible feature flags?
• Is the code modularized into mod.rs, config.rs, sink.rs, request_builder.rs, service.rs
• Does the code follow our style guidelines.

Documentation

• Generate a local website: cd website & make serve. Does it look good?
• Is there a cue file linking to base?
• Is there a markdown file under /website/content/en/docs/reference/configuration/sinks/?
• Are module comments included in mod.rs linking to any relevant areas in the external services documentation?

Configuration

• Are TLS settings configurable?
• Are the Request settings configurable?
• Should it have proxy settings? If so, are they in place?
• Does it need batch settings? If so, are they used?

Expectations

We endeavour to review all PRs within 2 working days (Monday to Friday) of submission.

Backward Compatibility

All changes should strive to retain backward compatibility. If a change breaks backward compatibility, it is much less likely to be approved. It is highly recommended you discuss this change with a Vector team member before investing development time.

Any deprecations should follow our deprecation policy.

Code Of Conduct

If you have not, please review Vector's Code of Conduct to ensure reviews are welcoming, open, and respectful.

Dependencies

Dependencies should be carefully selected. Before adding a dependency, we should ask the following questions:

1. Is the dependency worth the cost?
2. Is the dependency actively and professionally maintained?
3. Is the dependency experimental or in the development phase?
4. How large is the community?
5. Does this dependency have a history of security vulnerabilities?
6. Will this affect the portability of Vector?
7. Does the dependency have a compatible license?

Documentation

Documentation is incredibly important to Vector; it is a feature and differentiator for Vector. Pull requests should not be merged without adequate documentation, nor should they be merged with "TODOs" opened for documentation.

Ideally all modules should have module level documentation. Module level documentation can be omitted for modules where the purpose is obvious and covered by a general pattern. With the sinks there is typically a standard number of modules included (config.rs, request_builder.rs, service.rs, sink.rs, tests.rs), these modules don't need documentation as they will be covered with higher level documentation.

All pub and pub(crate) functions, structs and macros must have documentation.

Consider including examples for modules, structs, functions or macros that will be well used throughout Vector.

See the rustdoc book for more details on writing documentation.

Performance Testing

Vector currently offers 2 methods for performance testing:

1. Internal benchmarks located in the /benches folder.
2. A full end-to-end soak test suite for complex integration and performance testing.

For new integrations, consider whether a new soak test should be added.

Single Concern

Changes in a pull request should address a single concern. This promotes quality reviews through focus. If a pull request addresses multiple concerns, it should be closed and followed up with multiple pull requests addresses each concern separately. If you are unsure about your change, please open an issue and the Vector maintainers will help guide you through the scope of the change.

Readability

Code is read more than it is written. Code must be documented and readable.

Safe Code

Unsafe code should be reviewed carefully and avoided if possible. If code is marked as unsafe, a detailed comment should be added explaining why.

Security

Security is incredibly important to Vector. Users rely on Vector ship mission-critical and sensitive data. Please review the code explicitly for security issues. See Vector's Security guide for more info.

Testing

Code should be reasonably tested. Vector does not require 100% test coverage. We believe this level of coverage is unnecessary. As a general rule of thumb, we strive for 80% coverage, beyond this returns are diminishing. Please use your best judgment, some code requires more testing than others depending on its importance.

For integrations, consider whether the code could be integration tested.