Bazel cache analytics

Know which actions missed the cache—and why.

Aggregate hit rates hide the expensive misses. Hermetiq compares action keys, inputs, command lines, platforms, and environments across invocations so teams can distinguish correct invalidation from avoidable drift.

What you can answer

Prioritize the misses that cost engineers the most time.

  • Action-level hit and miss analysis
  • Cross-invocation action-key comparison
  • Environment and toolchain drift
  • Miss cost and rebuild impact
Questions Hermetiq helps answer

Start with the expensive miss, then inspect what changed.

A cache percentage is a symptom. Hermetiq helps connect a missed action to comparable builds and the key-forming evidence available for that action.

Expensive misses

Which action types and targets account for the most rebuild time after a cache miss?

Regression point

When did cache behavior change across comparable invocations?

Key divergence

Which observed commands, inputs, environments, or execution platforms differ?

Evidence boundary

Does the data indicate first-seen work, key divergence, an instance or configuration mismatch, or a gap in the cache evidence?

Integration model

Anchor the analysis in a build, then add action-level cache evidence.

The diagnosis is only as specific as the telemetry connected for the project. Start with BEP, then add the cache and action records needed to explain individual misses.

01

Connect Bazel build events

Use BEP to establish invocation identity, configuration, targets, timing, and outcomes.

02

Add cache and action evidence

Connect the remote-cache events and action-key evidence available for your Bazel and remote-cache configuration.

03

Compare equivalent builds

Hold the workload, flags, platform, and environment as constant as possible, then prioritize misses by rebuild impact.

Product evidence

Find the regression, then inspect the invocation-level difference.

Cache behavior by action type

Compare hit-rate changes and identify high-miss targets before drilling into individual actions.

Evidence model

Separate observed cache evidence from likely explanations.

Invocation identity

BEP anchors each recorded build, configuration, target set, and outcome used in the comparison.

Cache outcomes

Connected cache events support hit and miss analysis and the classifications actually reported by the cache integration.

Observed differences

Invocation comparison shows recorded command, input, environment, platform, timing, and outcome differences when available.

Rebuild impact

Execution-time and cost impact requires the corresponding completed-action records and configured cost inputs.

Technical references

Verify the underlying telemetry and performance model.

Try it with your builds

Investigate the next cache regression in your own build.

Connect one Bazel project to replace aggregate hit-rate guesses with action-level evidence.

Technical review: August 29, 2026 · Verify configuration details against the Bazel and Buildbarn versions deployed in your environment.