We're seeing occassional flakes on the Chromium waterfall in test steps with packages with large #s of blobs, due to Amber crashing:

[00087.081] pkgsvr: asking amber to fetch 5534 needed blobs
[00087.793] fatal error: concurrent map iteration and map write
[00087.799]
[00087.799] goroutine 6 [running]:
[00087.799] runtime.throw(0xa6300f049ce, 0x26)
...

Note that we are still SCP'ing a far to the guest, and then 'pm install'ing the package, rather than letting amber pull packages from a configured repository.