quickstart delployments broken in 1.24
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| | juju-core |
Critical
|
Unassigned | ||
| | 1.24 |
Critical
|
Eric Snow | ||
Bug Description
All quickstart jobs deploying HA Proxy broke.
http://
http://
http://
http://
All due to:
haproxy:
charm: cs:trusty/haproxy-1
exposed: false
service-status:
current: error
message: 'hook failed: "config-changed"'
From the unit log:
... INFO config-changed Traceback (most recent call last):
... INFO config-changed File "/var/lib/
... INFO config-changed main(hook_name)
... INFO config-changed File "/var/lib/
... INFO config-changed config_changed()
... INFO config-changed File "/var/lib/
... INFO config-changed if not create_services():
... INFO config-changed File "/var/lib/
... INFO config-changed services_dict = get_config_
... INFO config-changed File "/var/lib/
... INFO config-changed return parse_services_
... INFO config-changed KeyError: 'services'
... ERROR juju.worker.
| description: | updated |
| Changed in juju-core: | |
| milestone: | 1.24.6 → none |
| status: | Triaged → Invalid |
| Curtis Hovey (sinzui) wrote : | #1 |
| Eric Snow (ericsnowcurrently) wrote : | #2 |
The logs lead to the config-changed hook. In the hook code [1] it looks like one of the expected config options is not set. [2] This implies that one of those 3 changes is causing settings to not be set.
Thus, I expect that https:/
Offhand, I wonder if it's related to how default setting values are handled...
[1] "return parse_services_
[2] "services" in https:/
| Eric Snow (ericsnowcurrently) wrote : | #3 |
FTR, the bundle looks like this one:
https:/
Related resources:
* http://
* http://
* https:/
* https:/
* http://
* http://
* http://
* https:/
| Eric Snow (ericsnowcurrently) wrote : | #4 |
| Eric Snow (ericsnowcurrently) wrote : | #5 |
| Eric Snow (ericsnowcurrently) wrote : | #6 |
natefinch was able to reproduce the failure by manually running the quickstart_
Also, I've been unable to identify any semantic difference before and after 629fc0507c4b. I've added a test to check this in my fix branch. [1] The results of Nate's follow-up will help settle whether or not the suspect merge is at fault.
[1] https:/
| Eric Snow (ericsnowcurrently) wrote : | #7 |
mgz verified that the tests pass if 629fc0507c4b is reverted. So something is subtly different. We still need to determine what that is. It certainly isn't obvious to me at this point, so I'll keep at it.
| Eric Snow (ericsnowcurrently) wrote : | #8 |
Hmm. The API converts empty string config setting values into nil values. [1]
[1] https:/
| Eric Snow (ericsnowcurrently) wrote : | #9 |
Yep, that's it! I should have a fix up shortly.
Juju-CI verified that this issue is Fix Released in juju-core 1.24:
http://


The three suspect commits are /github. com/juju/ juju/commit/ bb38bf13c00283f 572f05139428961 b259275079 /github. com/juju/ juju/commit/ 01bab327cfc3d53 df5626da07c4729 8d350718a5 /github. com/juju/ juju/commit/ 629fc0507c4b61b e93fd79453244cb 8fc853f413
https:/
https:/
https:/