Gather just the necessary facts

[vrlo]

What type of PR is this?
/kind cleanup

What this PR does / why we need it:
Ansible is slow when there are many facts. This PR reduces the number of facts to necessary minimum by only gathering the ones that are needed.

Which issue(s) this PR fixes:

Fixes #5954
Also fixes #5927 as a side effect

Special notes for your reviewer:

Does this PR introduce a user-facing change?:

NONE

[k8s-ci-robot]

Hi @vrlo. Thanks for your PR.

I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[floryut]

@LuckySB please check if there isn't any conflict with your #5936

[LuckySB]

Can ansible combine everything into one task ?

- name: Gather facts
  hosts: all
  gather_facts: False

    - name: Gather facts
      setup:
        gather_subset: '!all,network,hardware'
        filter: "{{ item }}"
      loop:
        - ansible_distribution_major_version
        - ansible_default_ipv4
        - ansible_all_ipv4_addresses
        - ansible_memtotal_mb
        - ansible_swaptotal_mb

[LuckySB]

/ok-to-test

[vrlo]

Can ansible combine everything into one task ?

This would work partially, because the task with subset !all without filter would still need to be separate (to gather the minimal facts). However, combining the other two tasks wouldn't reduce the number of passes, and probably wouldn't be faster.

[LuckySB]

Great, so you can combine the other two tasks into one, which would slightly reduce the size of the playbook?

[LuckySB]

/approve

[LuckySB]

/assign @mattymo

[mattymo]

This looping will take 8x longer than just gathering all facts

[LuckySB]

why ?
the problem, as I understand it, is that on a node with running kubernetes and lots of pods, the size of facts is 30 megabytes, and the ansible loads this cache very slowly for each play

so with this task fact cache size is 10kb

[vrlo]

Ansible is slow when handling the large number of facts. This is a one time operation that ensures that the hostvars are kept small, speeding up all the other tasks.

[vrlo]

and it was 3MB -> 30kB in our case :-)

[chadswen]

The number of loop iterations here is constant and the gathers will occur across all available forks in parallel so I don't have a problem with it.

A previous method of looping through hosts to gather facts with delegation had n^2 iterations which were not ran in parallel. That method could easily take 8 times longer than a gather all so I understand the apprehension with a loop. However, this patch does not have that problem since each setup iteration works on as many nodes as allowed by forks.

[floryut]

Sorry @vrlo cancelled your pipeline because you need to rebase as tests name changed on master

[Miouge1]

I'm not an Ansible facts expert, but this looks reasonable. Since the same task is copied in a couple of playbooks, would it be possible to move that to a role or a task file?

[vrlo]

Yes, I wholeheartedly agree. The logical place would be something that's called only once, eg. bootstrap-os. However it's not called for all hosts in scale.yml (maybe that could be changed?).
The other logical place is kubespray-defaults, but it's called in every play (why?).
Maybe the best option would be the new playbook file that would be included in other playbooks, or run by itself (useful when someone wanted to use --limit, to gather facts on all hosts).

[chadswen]

/approve
/lgtm

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: chadswen, LuckySB, vrlo

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [LuckySB,chadswen]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[chadswen]

I had been testing something extremely similar to this while trying to scale to 200 node clusters in a single graceful upgrade build, so I can vouch for the approach. With this in place along with removing a few other bottlenecks, most of which coincidentally also got upstream variants merged this week, build time is greatly reduced at scale.

Thanks for the patch!