Add backward_sorted and backward_reverse_sorted ordering strategies

[segal90]

SUMMARY

Add backward_sorted and backward_reverse_sorted ordering strategies.

ISSUE TYPE

• Feature Pull Request

COMPONENT NAME

Orders

ADDITIONAL INFORMATION

This ordering can be useful if you have hostnames like fw-dc1-01, fw-dc1-02, fw-dc2-01, fw-dc2-02.
With these orders you can run plays against 02 (backup) nods first then the 01 (primary) nodes.

[samdoran]

We're on the fence about accepting this feature. What is the use case for this? The test case gives the same result as reverse_sorted. At a minimum, the name needs to be changed to clarify exactly what it is doing because it's rather confusing as is. Should 'reverted name' be 'reversed name'?

needs_info

[segal90]

Test cases give the same result because you need hostnames like fw-dc1-01, fw-dc1-02, fw-dc2-01, fw-dc2-02 for seeing the difference.
Let's see what will be the result with the different sorting:
sorted: fw-dc1-01, fw-dc1-02, fw-dc2-01, fw-dc2-02
reverse_sorted: fw-dc2-02, fw-dc2-01, fw-dc1-02, fw-dc1-01
backward_sorted: fw-dc1-01, fw-dc2-01, fw-dc1-02, fw-dc2-02
backward_reverse_sorted: fw-dc2-02, fw-dc1-02, fw-dc2-01, fw-dc1-01

As you can see with the 2 new algorithms you can run a play on primary or secondary nodes first. This can be handy if you have an active-backup HA solution in multiple data centers.

Putting DC's name before numbering in hostnames is pretty common in larger companies that have multiple DCs.

Should 'reverted name' be 'reversed name'?
Yes, you are right. Fixed.

[samdoran]

Ok. Can you add a changelog fragment and update the test data? See this fragment as an example. This would be classified as minor_changes.

[acozine]

@segal90. This isn't intuitive and can lead to unexpected results.

For example, if I had servers ending in -01 through -12, with backward_sorted I will get: -10, -01, -11, -02, -12, -03, etc.

Could you get the ordering you want using inventory or reverse_inventory instead?

[segal90]

@acozine,
Inventory is not sorted and the order of the hosts can change on every run. This is a valid use case with any inventory plugin which pulls hosts from external sources (eg: databases).

This algorithm could be useful with active-backup HA clusters where you have only 2-3 hosts (eg: lots of network device pairs).

Probably you are right and it can lead to unexpected results but comparing to the benefit mentioned above, I think it worth.

[acozine]

@segal90 thanks for the response. If this functionality gets merged, the docs team can add a sample inventory to the playbooks_strategies page to illustrate the results of each ordering strategy.

[samdoran]

@segal90 Would you be able to attend an IRC meeting to bring this up for discussion? ansible/community#541

[bcoca]

Hi!

Thanks very much for your submission to Ansible. It sincerely means a lot to us that you've taken time to contribute.

Unfortunately, we're not sure if we want this feature in the program, and I don't want this to seem confrontational. Our reasons for this are:

• This is not something we wish to add to Ansible at this point in time
• Historically we've added other sorting strategies but we do not wish for this to become a catch all for any future sorting algorithms people wish for in the future
• It is probably better to create a dynamic inventory to achieve the sorting mechanism you want https://docs.ansible.com/ansible/latest/dev_guide/developing_inventory.html

However, we're absolutely always up for discussion. Since this is a really busy project, we don't always see comments on closed tickets, but want to encourage
open dialog. You can stop by the development list, and we'd be glad to talk about it - and we might even be persuaded otherwise!

• https://groups.google.com/forum/#!forum/ansible-devel

In the future, sometimes starting a discussion on the development list prior to implementing a feature can make getting things included a little easier, but it's not always necessary.

Thank you once again for this and your interest in Ansible!