Diffs and patches

qbec uses a 3-way merge patch similar to kubectl/ksonnet apply. The Kubernetes documentation describes how this works.

For existing objects, the qbec diff command produces a diff between the last applied configuration stored on the server and the current configuration of the object loaded from source. This diff is “clean” in the sense of the remote object not having additional fields, default values and so on. It faithfully represents the change between the previous and current version of the object produced from source code.

When qbec apply is run, it calculates the patch for existing objects. This calculation does have to account for the shape of the object as stored by Kubernetes.

In many if not most cases, if qbec diff does not report a diff for an object, qbec apply will also not try to update the object.

This is not always true. Among other things:

  • the local object may have fields that are never stored on the server and every run of qbec apply will attempt to update these extra fields. This is particularly noticeable for custom resources and definitions.

  • the local object may represent a value differently from how the server stores it. For example a local CPU resource of 1000m may be stored in the server as 1 instead. Every qbec apply will notice this difference and try to update the value back to 1000m.


  • Diffs and patches may not always agree on the number of objects that are different.
  • Spurious apply patches can appear in the output. These can be noisy but they’re benign. One way to fix this would be to check the YAML output from the server and try to match the source code to have the same representation of the value.