install hooks fail because enable-os-upgrade/refresh-update ignored
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Canonical Juju |
Invalid
|
Undecided
|
Unassigned | ||
cloud-images |
Invalid
|
Undecided
|
Unassigned | ||
cloud-init |
Expired
|
Undecided
|
Unassigned |
Bug Description
As seen at
http://
There is a dramatic rise in hook failures since March 24. This is because juju is not running apt-get update or upgrade.
Joyent dominates, but rackspace and GCE have had spots of failures. The unit logs show apt fails to find packages. Joyent tends to have stale images so Juju CI tests always run with os-enable-
Today I deployed the landscape bundle in joyent and repeated the postgresql install hook failures. The log confirm an apt issue
E: Failed to fetch http://
juju model-config shows apt should have been updated and the host upgraded
enable-
enable-
But /var/log/
I left the model untouched for 3 hours. It remained in error. I ran
juju ssh 2 -- sudo apt-get update
3 minutes later the unit was installed.
This bug is in 2.1 and 2.2
Changed in juju: | |
status: | Triaged → Incomplete |
milestone: | 2.2-beta3 → none |
tags: | added: id-58ee260e4c1fcd6e9adb77fe |
Changed in cloud-images: | |
status: | New → Invalid |
Both enable- os-refresh- update and enable-os-upgrade are true by default.
As a first step, I tested bootstrapping on LXD and adding a machine. For both machines, I added extra tracing to Juju to show that the cloud init script rendering was being passed true for both config values as expected. We are correctly setting the "package_update" and "package_upgrade" attributes in cloud init.
I can see in cloud init output that apt-get update and upgrade appear to be run.
Next step to test on a public cloud.