Comment 23 for bug 1949089

I spend a bit of quality time with this bug today and it seems there is (also?) a kernel dimension to it. I ran the following:
"""
$ git status
On branch tests-use-core18-from-gce
Your branch is up to date with 'sergiocazzolato/tests-use-core18-from-gce'.
$ git diff
diff --git a/overlord/devicestate/devicestate.go b/overlord/devicestate/devicestate.go
index d40b32bcbd..77ec3fc0a8 100644
--- a/overlord/devicestate/devicestate.go
+++ b/overlord/devicestate/devicestate.go
@@ -117,6 +117,8 @@ func canAutoRefresh(st *state.State) (bool, error) {
        if !seeded {
                return false, nil
        }
+ // HACK
+ return false, nil

        // Try to ensure we have an accurate time before doing any
        // refreshy stuff. Note that this call will not block.
diff --git a/spread.yaml b/spread.yaml
index 0aef5dea7c..68cbdc09b7 100644
--- a/spread.yaml
+++ b/spread.yaml
@@ -37,7 +37,7 @@ environment:
     MANAGED_DEVICE: "false"
     # a global setting for LXD channel to use in the tests
     LXD_SNAP_CHANNEL: "latest/edge"
- UBUNTU_IMAGE_SNAP_CHANNEL: "latest/candidate"
+ UBUNTU_IMAGE_SNAP_CHANNEL: "beta/1.11"
     CORE_CHANNEL: '$(HOST: echo "${SPREAD_CORE_CHANNEL:-edge}")'
     BASE_CHANNEL: '$(HOST: echo "${SPREAD_BASE_CHANNEL:-edge}")'
     KERNEL_CHANNEL: '$(HOST: echo "${SPREAD_KERNEL_CHANNEL:-edge}")'
diff --git a/tests/lib/prepare.sh b/tests/lib/prepare.sh
index e6b984c4d0..215d0611be 100755
--- a/tests/lib/prepare.sh
+++ b/tests/lib/prepare.sh
@@ -973,8 +973,8 @@ EOF
     fi

     if os.query is-core18; then
- curl -s -o core18.snap https://storage.googleapis.com/snapd-spread-tests/snaps/core18_20211102_amd64.snap
- EXTRA_FUNDAMENTAL="$EXTRA_FUNDAMENTAL --snap $PWD/core18.snap"
+ curl --insecure -o pc-kernel.snap https://people.canonical.com/~mvo/tmp/pYVQrBcKmBa0mZ4CCN7ExT6jH8rY1hza_831.snap
+ EXTRA_FUNDAMENTAL="$EXTRA_FUNDAMENTAL --snap $PWD/pc-kernel.snap"
     fi

     /snap/bin/ubuntu-image snap \
"""
which essentially moves the image to a 2 month older 4.14 kernel. With that I could not reproduce the error for three subsequent runs of 2 repeats.

However when I disabled this code and use the current "18/edge" pc-kernel (or stable, it's the same revision) then I can trigger the bug right away.