Test me if you can. Do YML developers Dream of testing Ansible?

Date: 2019-01-22

It is text version of presentation 2018-04-25 from Saint-Petersburg Linux User Group.

• Slides
• Russian version
• English version
• Crosspost Russian version
• Crosspost English version
• Configuration example: github.com/ultral/ansible-role-testing

Introduction

I suppose that that you make configuration management, not bash. It means that you have to test it some how. Have you ever tested ansible roles? How do you do it?

In my case, we have:

• A lot of different ansible roles.
• Hyper-V hosts as hypervisor.
• A private cloud with limited possibility to create VMs on demand.
• A proxy for internet access.
• Inability test ansible roles inside docker, because of a role = whole VM configuration.
• Decision to implement green build policy for git repository with ansible roles.

How to do it?

Let us compare solutions for testing.

We decided not to reinvent the wheel & get production ready solution. Our infrastructure team had strong ruby skills & great experience with ruby, as a result we chose Test Kitchen & inspec

Kitchen-CI

Green Build Policy

Nested virtualization

As you remember, we had a private cloud with limited possibility to create VMs on demand. We decided to create VMs inside VMs.

First of all we tried to run Virtualbox x32 without nested. It was bad idea because of kernel panic. Also vast majority of our VMs in our infrastructure are x86_64, so we decided to continue research. As a result we decided to use nested virtualization. Hopefully it was supported by our host servers.

Faced issues

I was implementing testkitchen and faced with some issues.

Pass proxy settings from host into testkitchen guest VM

In some test suits, we configured proxy client settings inside VM created by testkitchen. However, proxy was not configured on testkitchen host & ansible cannot use extra variables with empty values

Solution: create erb template for setting default proxy if ENV variables is not set

<%= ENV['http_proxy'].to_s.empty? ? 'http://proxy.example.com:3128' : ENV['http_proxy'] %>

Manage network settings via playbook

Some roles configure network interfaces. Test suit was looked like:

• Deploy network settings to VMs
• Reload network
• It is failed

Solution: add interfaces to VMs

Fail if suit cases contains “-“ in name

Virtualbox can’t use “_” in a vm name

Solution: rename suitcases “vm_” => “vm-“

Oracle test fails without “.” at the end of VM name

We use role in production, however when we decided to test it, it was failed. We reproduced it.

I would like to show clue.

[root@vm-oracle vagrant]# getent ahosts vm-oracle
127.0.0.1 STREAM vm-oracle
127.0.0.1 DGRAM
127.0.0.1 RAW
[root@vm-oracle vagrant]# getent ahosts vm-oracle.
fe80::a00:27ff:febd:bd6a STREAM vm-oracle
fe80::a00:27ff:febd:bd6a DGRAM
fe80::a00:27ff:febd:bd6a RAW
10.0.2.15 STREAM
10.0.2.15 DGRAM
10.0.2.15 RAW
[root@oracle vagrant]# getent ahosts oracle.example.com.
192.168.128.182 STREAM oracle.example.local
192.168.128.182 DGRAM
192.168.128.182 RAW

Do you have any ideas what is happening?

It is tricky bug:

1. we enabled listen IPv4 only in oracle listener settings
2. oracle used FQDN
3. linux contains special database “myhostname” for resolving hostname, it is uses after /etc/hosts & dns resolving
4. Vagrant created VM & updated /etc/hosts

I’d like to clarify it a bit more: What has happened in case vm-oracle?

1. vagrant created vm
2. vagrant updated /etc/hosts(vm-oracle x2)
3. oracle listener listened IPv4
4. oracle listeners resolved vm-oracle. & geted IPv6
5. FAILED

What has happened in case vm-oracle.?

1. vagrant created vm 
2. vagrant updated /etc/hosts ( vm-oracle &  vm-oracle.)
3. oracle listener listened IPv4
4. oracle listeners resolved vm-oracle. & geted IPv4
5. OK

OOM is coming

OOM randomly was killing VMs. Testkitchen was failed with strange errors.

Solution: increase RAM

Slow builds

It was working slowly

Solutions:

• Packer. Prebuilded vagrant box with common tasks
• Concurrency

Conclusion

On one hand, current implementation works, but on the other hand, there are some issues

• is not user friendly.
• we mix ruby & python.
• there is no indepotence check.
• it works slow.
• it is hard to trace logs at single job.

As a result, molecule & docker might be pretty interesting solution.

Links

• presentation
• simple example
• http://kitchen.ci/
• https://t.me/pro_ansible
• https://github.com/chef/kitchen-inspec
• https://docs.chef.io/config_yml_kitchen.html
• https://docs.chef.io/ctl_kitchen.html
• https://github.com/neillturner/kitchen-ansible/blob/master/provisioner_options.md