list pods assigned to a node by pod informer cache

[Garrybest]

What this PR does / why we need it:
Now there is a perf problem about the way we list pods that assigned to a node. When the cluster is large scale, like 3k+ nodes and 90k+ pods. The descheduler is not able to work because every interval it will request kube-apiserver for almost 30k+ times(enable 10 plugins and every plugins request for 3k+ times) with a FieldSelector. Meanwhile, it leads to a lot of pressure for kube-apiserver. So I try to make a revision by using pod informer cache.

Which issue(s) this PR fixes:
Fixes #671

Special notes for your reviewer:
I'm afraid the PR is too large to review (beacause listing pods is a really basic action and the function is called everywhere in strategy files and test files). So I separate this PR into 3 commits in order to make code review easier.😄

1. Reform ListPodsOnANode by using pod informer and indexer
   This commit is the most important one. The code is very concise. I establish an indexer to map the pods and their assigned nodes by an informer indexer. So every time we list pods we could use informer cache instead of requesting kube-apiserver by filed selectors.

2. Reform all strategies by using getPodsAssignedToNode
   This commit let all other files of descheduler adapt to listing pods by function getPodsAssignedToNode. Not so important, just some adaptions.

3. Reform all test files
   This commit is not important. Just a lot of adaptions of test files and it took me a lot of time to do the adaptions (cause the test cases and strategies are so many). 🤣

[k8s-ci-robot]

Welcome @Garrybest!

It looks like this is your first PR to kubernetes-sigs/descheduler 🎉. Please refer to our pull request process documentation to help your PR have a smooth ride to approval.

You will be prompted by a bot to use commands during the review process. Do not be afraid to follow the prompts! It is okay to experiment. Here is the bot commands documentation.

You can also check if kubernetes-sigs/descheduler has its own contribution guidelines.

You may want to refer to our testing guide if you run into trouble with your tests not passing.

If you are having difficulty getting your pull request seen, please follow the recommended escalation practices. Also, for tips and tricks in the contribution process you may want to read the Kubernetes contributor cheat sheet. We want to make sure your contribution gets all the attention it needs!

Thank you, and welcome to Kubernetes. 😃

[k8s-ci-robot]

Hi @Garrybest. 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.

[damemi]

/ok-to-test

[ingvagabund]

@Garrybest you were faster sir :) #674

[ingvagabund]

@Garrybest I need this change merged pretty soon. I have another PR on the way that relies on the pod lister and removing the field selector. What is your availability?

[Garrybest]

What is your availability?

Hi @ingvagabund, I'm totally available recently and I'm willing to make cooperations.😄

By the way, according to what @damemi said in #671 (comment), I have some experimental results and facts to show.

1. I have used an informer indexer, which makes the code simple and concise.

According to the first commit of my PR, I build a GetPodsAssignedToNodeFunc to provide a method to get pods that assigned to a node. The code is very simple, just establish a inform indexer between pods and assgined nodes.

2. Using filter function to replace the field selector.

Actually, the field selector is a special kind of pod filter. So I aggregate the origin incoming filter function and our filed selector filter function as a final pod filter in ListPodsOnANode. So actually the field selector still works but as another form.

3. Some test result in my own clusters of production environment.

Test method: Run ListPodsOnANode without any extra filter functions looped by all ready nodes and then measure the elapsed time.
Test machine: A cloud VM with 6 cpu cores and 9.4GB memory.

• A cluster with 233 nodes and 1933 assigned pods.
  • Before: 96s
  • After: 0.2s
  • Accelerate rate: 480 times
• A cluster with 5235 nodes and 42477 assigned pods.
  • Before: More than 30min
  • After: 3.36s
  • Accelerate rate: more than 535 times

I think the benefit of this PR is very obvious. Note that this test is only listing once. If we enable all plugins, the accelerate rate will be a very large number since the listing of all nodes will be called more than once.

[ingvagabund]

This PR is well done!!!

@Garrybest thank you. I like the changes you made to the code.

/approve
/lgtm

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: Garrybest, ingvagabund

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 [ingvagabund]

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