[freezer] get_client_instance() got an unexpected keyword argument 'api_version'

[Damian Dąbrowski]

Hi,

I'm trying to deploy openstack rocky using latest version of kolla-ansible, generally all components works fine but freezer-scheduler container returns an error:

2019-02-17 15:06:10.997 5 ERROR freezer.scheduler.freezer_scheduler [-] get_client_instance() got an unexpected keyword argument 'api_version'

[alpha23 (alpha23)]

Same issue.

[Radosław Piliszek (yoctozepto)]

Looks like programming error in freezer.

[alpha23 (alpha23)]

How is this a programming error in freezer when the kolla/centos-source-freezer-scheduler:rocky container will not start? This container was pulled from docker hub.

+ . kolla_extend_start
++ LOG_DIR=/var/log/kolla/freezer
++ [[ ! -d /var/log/kolla/freezer ]]
+++ stat -c %U:%G /var/log/kolla/freezer
++ [[ freezer:freezer != \f\r\e\e\z\e\r\:\k\o\l\l\a ]]
++ chown freezer:kolla /var/log/kolla/freezer
+++ stat -c %a /var/log/kolla/freezer
++ [[ 2755 != \7\5\5 ]]
++ chmod 755 /var/log/kolla/freezer
++ . /usr/local/bin/kolla_freezer_extend_start
+ echo 'Running command: '\''freezer-scheduler --config-file /etc/freezer/freezer.conf start'\'''
Running command: 'freezer-scheduler --config-file /etc/freezer/freezer.conf start'
+ exec freezer-scheduler --config-file /etc/freezer/freezer.conf start
get_client_instance() got an unexpected keyword argument 'api_version'

[Mark Goddard (mgoddard)]

How is it not a programming error? Even if Kolla Ansible has provided incorrect arguments or config, freezer should not fail in that way.

[alpha23 (alpha23)]

I just saw your response. For whatever reason, I did not receive and email notification...

How did the kolla-ansible tests pass given that freezer_scheduler is being run in a kolla-ansible container and not a container derived from the freezer project?

Given that freezer_scheduler is run in is a kolla-ansible container, the issue should either be a configuration issue, the incorrect parameters are being passed, or the container has been incorrectly built.

I agree that it should not fail but then the kolla-ansible team should see the same failures in the tests. Given that I've configured freezer without customized configuration files, this should not be a configuration issue. As a note, I'm running Ubuntu 18.04.4 and Docker 18.06.1-ce. The freezer containers deployed are directly from the docker repository:

kolla/centos-source-freezer-api rocky b97739f5fd3b 11 months ago 1.04GB
kolla/centos-source-freezer-scheduler rocky 5b822a78a8d9 11 months ago 952MB

[Mark Goddard (mgoddard)]

Simple - we don't test freezer deployment in CI. See the support matrix for the list of services that we test: https://docs.openstack.org/kolla/latest/support_matrix.html

CI has much improved since Rocky, but we still don't test freezer.

[alpha23 (alpha23)]

I did not know that this matrix existed. Images that are not tested really deserve a different image tag or ansible tag When deploying containers using the kolla-ansible "stable/xxx" branch, the untested images that are deployed cannot be classified as "stable." Having known this could have saved hours of frustration.

I appreciate you getting back to me quickly.

[alpha23 (alpha23)]

Which branch of freezer scheduler did the kolla-ansible team use in the kolla/centos-source-freezer-scheduler rocky image given that freezer did not have a stable/rocky branch? See also, https://github.com/openstack/freezer.

[Radosław Piliszek (yoctozepto)]

The latest Freezer release at the time - 6.0.0 from Queens

https://opendev.org/openstack/kolla/src/commit/dd0069addbfba430d1a8cf15e91005d0f0279c11/kolla/common/config.py#L339-L346

https://releases.openstack.org/teams/freezer.html#queens