| Age | Commit message (Collapse) | Author |
|---|
|
Azure's ephemeral disks are not guaranteed to be assigned the same name by
the kernel every boot. This causes problems on ~2% of Azure instances, and
can be fixed by using udev rules to give us a deterministic path to mount;
this patch introduces those udev rules and modifies the Azure data source
to use them.
Changes to a couple of config modules were also required. In some places,
they just needed to learn to dereference symlinks. In cc_mounts this
wasn't sufficient because the dereferenced device would have been put in
/etc/fstab (rather defeating the point of using the udev rules in the
first place). A fairly hefty refactor was required to separate "is this a
valid block device?" from "what shall I put in fstab?".
LP: #1411582
|
|
|
|
The ephemeral disk will not necessarily be assigned the same name at
each boot (LP: #1411582), so we use some udev rules to ensure we always
get the right one.
|
|
The fabric provides the user password in plain text via the CDROM,
and cloud-init has previously wrote the ovf-env.xml in /var/lib/waagent
with the password in plain text. This change redacts the password.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Including minor refactoring to make mocking considerably easier.
|
|
This should fix the Azure data source on Python 3, and is appropriate as
XML shouldn't really be read as a string.
|
|
Specifically, this is to support Azure's G-series VMs (which come with
disks up to 6500GB).
|
|
|
|
|
|
|
|
Fixed all complaints from running "make pep8". Also version locked
pep8 in test-requirements.txt to ensure that pep8 requirements don't
change without an explicit commit.
|
|
|
|
|
|
|
|
per instance. Under a variety of circumstances, the ephemeral device may
be presented as a default device. This patch detects when that situation
happens and triggers CC modules disk-setup and mounts to run again.
Details of changes for cloudinit/sources/DataSourceAzure.py:
- auto-detect the location of ephemeral0
- check each boot if ephemeral0 is new
- done via NTFS w/ label of "Temporary Storage" w/ no files on it
- if device is mounted, datasource will unmount it
- if is new, change mounts and disk-setup to always for that boot only
|
|
|
|
|
|
|
|
as the previous instance (LP: #1269626).
|
|
We were passing a unicode string to 'runcmd' in the path to the .crt file.
That is because the keyname was coming from ovf file as unicode.
Ie:
u'/var/lib/waagent/6BE7A7C3C8A8F4B123CCA5D0C2F1BE4CA7B63ED7.crt'
Then, logging was extending not appending errors.
|
|
LP: #1232175
|
|
makes it cloud agnostic.
|
|
Modified cc_mounts to identify whether ephermalX is partitioned.
Changed datasources for Azure and SmartOS to use 'ephemeralX.Y' format.
Added disk remove functionally
|
|
in general block device mappings should be to block devices, not
partitoins.
|
|
Also
* cloudinit/sources/DataSourceAzure.py: invalid xml in a file called
'ovfenv.xml' should raise BrokenAzureDatasource rather than
NonAzureDataSource
* cloudinit/sources/DataSourceSmartOS.py:
cloudinit/sources/DataSourceAzure.py
use 'ephemeral0' as the device name in builtin fs_setup
* tests/unittests/test_datasource/test_azure.py:
* always patch 'list_possible_azure_ds_devs' as it calls find_devs_with
which calls blkid, and dramatically was slowing down tests on my system.
* test_user_cfg_set_agent_command_plain:
fix this test to not depend on specific format of yaml.dumps().
* test_userdata_arrives: add a test that user-data makes it through
|
|
Previously we had this 'ephemeral_disk' entry in the datasource config
for Azure, and then we also copied some entries into the .cfg
for that datasource from the datasource config.
Ie, datasource['Azure']['disk_setup'] would be oddly copied
into the .cfg object that was returned by 'get_config_obj'
Now, instead, we have a BUILTIN_CLOUD_CONFIG, which has those same
values in it.
The other change here is that 'ephemeral_disk' now has no meaning.
Instead, we add a populated-by-default entry 'disk_aliases' to the
BUILTIN_DS_CFG, and then just return entries in it for
'device_name_to_device'
|
|
|
|
|
|
Azure provides a random bit of data at '/sys/firmware/acpi/tables/OEM0'.
The walinux calls this "Entropy in ACPI table provided by Hyper-V".
|
|
|
|
This simply correctly invokes subp through util.log_time.
The arguments to subp is named 'args' not 'command'.
LP: #1214541
|
|
'password' was the wrong key. It should have been setting the default
user's "plain_text_password".
Instead of doing that, though, we're encrypting the value and putting it in
'passwd', which will then be passed on to useradd. The key value in doing
this is that the plain text password will not be stored in obj.pkl.
(admittedly it is still in plain text in the ovf-env.xml file).
|
|
reading /proc/uptime is going to be slower, and no reason to do it on most
things. Better to only do it when you suspect maybe a need for it.
|
|
The reason for this is that more and more things I was wanting to be able to
see how long they took. This puts that time logic into a single place. It
also supports (by default) reading from /proc/uptime as the timing mechanism.
While that is almost certainly slower than time.time(), it does give
millisecond granularity and is not affected by 'ntpdate' having
run in between the two events.
|
|
As shown in comments of bug 1202758 and filing of ntp bug 1206164, waiting
for the output of this command causes us to wait for ntpdate to fully
finish.
Ideally I think we'd disable ntpdate running on this run, but
that is not trivially possible.
|
|
the environment that was set up to include 'interface' was not actually
being passed on to 'subp', so when the command ran it wasn't available.
|
|
|
|
See the added doc/sources/azure/README.rst for why this is necessary.
Essentially, we now are doing the following in the get_data() method
of azure datasource to publish this NewHostname:
hostname NewHostName
ifdown eth0;
ifup eth0
LP: #1202758
|
|
LP: #1204190
|
