Reliability & Diagnostics

LabHit is built to be resilient by structure: every run is accounted for, build failures are easy to diagnose, and unreliable stages surface on their own. This page covers the three pieces.

Never lose a run#

A pipeline run is recorded the moment it starts -- before any stage executes -- not only when it finishes. So if the engine restarts or crashes mid-run, the run is not lost.

On the next startup, any run still left in a non-terminal state is reconciled to a terminal interrupted state. interrupted is honest about what happened: the run was cut short, so its true outcome is unknown. It is never reported as a misleading pass or fail -- on a GitHub Check Run it maps to a neutral conclusion.

The guarantee is simple: a run is always recoverable to a terminal state, even after an unclean shutdown. No configuration is required; this is on by default with persistent storage.

Diagnose a build failure#

When a build fails, paste the log into the diagnoser at init.labhit.dev and get the likely cause and the fix in plain language.

• It matches your log against a curated library of common CI/CD failure patterns -- out-of-memory, dependency conflicts, missing secrets, timeouts, test failures, and more.
• It is read-only: it never runs or changes anything.
• Your log never leaves your browser -- the analysis is entirely client-side, so nothing is uploaded.

For each finding it shows the exact line it matched, why it happened, and how to fix it. When it is unsure, it says so.

Pipeline health#

Once you have run history, the pipelineHealth GraphQL query summarizes it:

• Success rate over the window (interrupted runs count against it; still-running runs do not).
• Flaky stages -- stages that both passed and failed in the window.
• Slowest stages by average duration.
• Interrupted count -- a reliability signal: how many runs were cut short by a restart.

{
  pipelineHealth(pipelineName: "ci", limit: 50) {
    successRate
    flakyStages
    stages {
      name
      runs
      failures
      avgDurationSecs
      flaky
    }
  }
}

See the API Reference for the full query and every field.