During precaching of images on the spoke, the job uses opm to render the catalog then selects the last entry in the olm.channel entry for the required package. However, this may not be the latest operator version and instead may be an arbitrary release in the channel.

When in a disconnected environment, this becomes an issue since old versions may not be mirrored, causing a failure.

    based on the git history, this likely affects all TALM versions currently supported

    always

    1. Edit a file-based olm catalog to list the entries for a package such that the latest version is not last. Package sriov-fec in registry.redhat.io/redhat/certified-operator-index:v4.14 is an example of this appearing in a prod catalog. Additionally, ensure the last bundle includes relatedImages that would fail to pull (not mirrored, don't exist, etc.)
    2. Create a CGU with precaching enabled that manages a policy containing a subscription from the catalog in step 1
    3. Wait until the CGU enters an UnrecoverableError and the Job pod on the spoke fails

    UnrecoverableError during precaching

    Successful precache of latest version