kome

kome

Modern CI/CD builds are nowadays parallelized, by running independent steps on parallel jobs. As a result, test outcomes are scattered between multiple jobs. It becomes hard to aggregate test results from all these jobs to provide consolidated and actionable feedback, especially given that not every job will finish cleanly and make take different amount of time.

Idea

1. Build steps write files to $KOME_PATH/*.json.

2. Run:

   npx kome

3. kome writes the files found in $KOME_PATH to a centralized database, and creates or updates a GitHub comment based on the data collected so far.

Data model

All data is stored on Firebase Realtime Database. kome accesses the database using a service account, so the database can be set up in locked mode (disallowing all external reads and writes).

• <baseRef>
  To allow the same database to be used with multiple projects, kome can be configured to used as a subtree inside the database as the base path. This is configured via firebase.baseRef settings inside kome.config.js.

  Metadata can be attached to a commit (applies to a single comment) or to a pull request.

  • commits/$sha
    Holds the commit metadata. This is arbitrary data, generated by your build, collected and uploaded by the kome command.

  • pulls/$number
    Holds the pull request metadata. This is arbitrary data, generated by your build, collected and uploaded by the kome command.

  • comments/$prNumber
    Holds the state for pull request comments on GitHub.

    • commentId
      The comment ID. Existing comment will be edited if it exists. Otherwise, a new comment will be created, an its ID will be written here.

    • hash
      The hash of the comment contents. Can be used to prevent unnecessary edits to the comment.

    • lock
      Since multiple builds may run this command simultaneously (in rare cases), this key is used to make sure the comment is edited by only one build at a time.

      If this is null, then there is no active lock. If this is not null, then a build is working on editing this comment.

      • owner
        The process UUID of the process that locks this comment. When the CLI is run, a random identifier is generated as the process UUID. When unlocking, the unlock transaction makes sure not to unlock if it’s not owned by the current process.

      • acquiredAt
        The timestamp at which the lock has been acquired. If the lock has not been released within 10 seconds, we assume that the owner has gone, and will treat the lock as nonexistent.