On the latest centos 7 release, we are unable to resize our instances filesystems

[J-P methot]

Since the CentOS 7 update to cloud-init 0.7.9, we've been unable to resize our instance's root drives. Of course, since root volume resize is not officially supported in openstack, what we do is a bit of hack. We basically shutdown the vm -> cinder reset-state --state available *volume id* -> cinder extend *volume id* *size* -> cinder reset-state --state in-use *volume-id* -> migrate the VM (to remake the iscsi mount) -> startup the VM. This increases the volume size and still works. The issue is when cloud-init needs to execute the growpart module.

See https://bugs.centos.org/view.php?id=13931

The only error we get is related to vendordata, which may be preventing the resize process to happen :
2017-09-28 02:01:43,641 - util.py[WARNING]: failed stage init
2017-09-28 02:01:43,641 - util.py[DEBUG]: failed stage init
Traceback (most recent call last):
  File "/usr/lib/python2.7/site-packages/cloudinit/cmd/main.py", line 513, in status_wrapper
    ret = functor(name, args)
  File "/usr/lib/python2.7/site-packages/cloudinit/cmd/main.py", line 281, in main_init
    init.update()
  File "/usr/lib/python2.7/site-packages/cloudinit/stages.py", line 357, in update
    self._store_vendordata()
  File "/usr/lib/python2.7/site-packages/cloudinit/stages.py", line 381, in _store_vendordata
    util.write_file(self._get_ipath('vendordata_raw'), raw_vd, 0o600)
  File "/usr/lib/python2.7/site-packages/cloudinit/util.py", line 1755, in write_file
    content = encode_text(content)
  File "/usr/lib/python2.7/site-packages/cloudinit/util.py", line 154, in encode_text
    return text.encode(encoding)
AttributeError: 'dict' object has no attribute 'encode'

This is on openstack Ocata though, and the vendordata has always been empty, as we use the default settings for it.

Cloud provider: we are the cloud provider, PlanetHoster, using Openstack Ocata.
version: Cloud-init 0.7.9-9.el7.centos.2

relevant cloud-init config:

cloud_init_modules:
 - migrator
 - bootcmd
 - write-files
 - growpart
 - resizefs
 - update_etc_hosts
 - rsyslog
 - users-groups
 - ssh

[Scott Moser (smoser)]

Any datasource that uses convert_vendordata is possibly affected.

This is bad in that it fails, but not as bad as it could be. Cloud-init
itself never re-constitutes the blob on disk. Its only a convenience for
the user.

What we should do:
 a.) always have vendordata_raw be bytes
 b.) write them as bytes to the file

We can still allow the datasource to use something like 'convert_vendordata'
intelligently to get types other than bytes. We just have to have
the 'get_vendordata' function access a different attribute in the datasource.

ie:
- DataSource.vendordata_raw should always be the raw bytes as found.
- get_vendordata would then use some other attribuate 'vendordata_converted' (?) to take advantage of the datasource specific knowlege.

[J-P methot (jphilippemethot)]

After reading this comment, I still have a few question:

Regarding the vendordata error, is it confirmed that it's really the issue preventing growpart from running, or are the two problems unrelated?

Additionally, for the time being, is there a way I could circumvent that bug? Like setting some random, inconsequential vendor data in openstack so that it's not empty, for example?

[J-P methot (jphilippemethot)]

Added cloud.cfg for reference

[James Falcon (falcojr)]

Tracked in Github Issues as https://github.com/canonical/cloud-init/issues/3029