User-configured cloud-config datasource is not used

[janeczku]

RancherOS Version: (ros os version)
v1.1.2
Where are you running RancherOS? (docker-machine, AWS, GCE, baremetal, etc.)
Openstack, Virtualbox

Steps to reproduce

1. Install RancherOS v1.1.2
2. Configure cloud-init datasource: sudo ros config set rancher.cloud_init.datasources [ec2]"
3. Reboot
4. Check cloud-init-save.log

Expected

cloud-config is fetched from ec2 metadata source. This was working at least up to v1.0.3.

Actual

ec2 datasource is not evaluated by cloud-init-save. It instead tries to use the default configdrive source.

init.log

time="2018-01-02T14:32:26Z" level=debug msg="[8/21] Done mount and bootstrap"
time="2018-01-02T14:32:26Z" level=info msg="[9/21] Starting cloud-init"
time="2018-01-02T14:32:26Z" level=info msg="ros init: Detected Hypervisor: xenhvm"
time="2018-01-02T14:32:26Z" level=info msg="init, runCloudInitServices([ec2])"

cloud-init-save.log

time="2018-01-02T14:32:33Z" level=info msg="Apply Network Config SyncHostname"
time="2018-01-02T14:32:33Z" level=info msg="datasources that will be consided: []string{"configdrive:/media/config-2"}"
time="2018-01-02T14:32:33Z" level=info msg="cloud-init: Checking availability of "cloud-drive""
time="2018-01-02T14:32:34Z" level=error msg="no such file or directory"
time="2018-01-02T14:32:34Z" level=info msg="cloud-init: Datasource unavailable, skipping: cloud-drive: /media/config-2 (lastError: no such file or directory)"
time="2018-01-02T14:32:34Z" level=info msg="Cloud-init datasource that was used: %!s(<nil>)"

[janeczku]

/cc @r0b2g1t

[niusmallnan]

Yeah, it's a bug.

We use config-drive datasource as default, but the user has set ec2 and config-drive should be overwritten.

When system-docker starts, it performs a series of tasks, something like:

preparefs
load modules
....
cloud-init
....
write cfg and log files
.....

Current, cloud-init performed prior to write cfg and log files. This causes cloud-init to execute while still unable to read the user's settings(like rancher.cloud_init.datasources).

We need to consider adjusting the startup sequence.

[kingsd041]

Tested with RancherOS v1.2.0-rc2.
After configuring the [ec2] datasource, I can get the following log in cloud-init-save.log, I think rancheros is using ec2 data sources.

time="2018-02-01T12:29:21Z" level=info msg="Apply Network Config SyncHostname"
time="2018-02-01T12:29:21Z" level=info msg="datasources that will be consided: []string{"ec2"}"
time="2018-02-01T12:29:21Z" level=info msg="cloud-init: Checking availability of "ec2-metadata-service""
time="2018-02-01T12:29:21Z" level=error msg="IsAvailable: http://169.254.169.254/:latest/user-data (lastError: Unable to fetch data: Get http://169.254
time="2018-02-01T12:29:21Z" level=info msg="cloud-init: Datasource unavailable, skipping: metadata: http://169.254.169.254/latest/user-data (lastError:
time="2018-02-01T12:29:21Z" level=info msg="Cloud-init datasource that was used: %!s(<nil>)"
time="2018-02-01T12:29:21Z" level=debug msg="init: SaveCloudConfig(post ApplyNetworkConfig): netconf.NetworkConfig{PreCmds:[]string(nil), DNS:netconf.D
time="2018-02-01T12:29:21Z" level=info msg="Apply Network Config"

Close it.