instance_dd_test.sh runs out of disk space without noticing

[Dr. Jens Harbott]

When doing some testing with samples/tasks/scenarios/vm/boot-runcommand-delete, I noticed that the I/O times that are measured are way too low when comparing them with some dd runs performed manually on an instance.

The reason is that with the current setup, there are only a couple of MB of free space in the cirros root filesystem, so the dd stops with an error almost immediately, but the script ignores the error and happily returns the short runtime.

Maybe it is not useful to make the script very complicated by adding sophisticated error handling, but instead it would suffice to at least have some warning comment about this.

As to why there is not enough space in the root fs:

1. Cirros does not resize its root fs by default, one can tune an option in /etc/default/resizefs to enable it, but that would require users to create and upload a patched cirros image. Also note that setting RESIZE_MODE='foreground' is currently broken (https://bugs.launchpad.net/cirros/+bug/1431320).

2. Even if cirros does the resize, with m1.nano the root fs would still have much less then 1GB of space, and even with the proposed change to use m1.tiny instead, the total size will be 1GB and writing a new file of 1GB size still will fail prematurely.

[Boris Pavlovic (boris-42)]

So maybe we should just use normal images?