fix issue 912: do dedupe with resourceVersion

[fishautumn]

reason: on timeout-reconnect it will add new a new copy to all objects in FIFO items map.
fix: check resourceVersion when adding object copy, skip if newly added object has same resourceVersion with previous latest object.

[k8s-ci-robot]

Thanks for your pull request. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please follow instructions at https://git.k8s.io/community/CLA.md#the-contributor-license-agreement to sign the CLA.

It may take a couple minutes for the CLA signature to be fully registered; after that, please reply here with a new comment and we'll verify. Thanks.

---

• If you've already signed a CLA, it's possible we don't have your GitHub username or you're using a different email address. Check your existing CLA data and verify that your email is set on your git commits.
• If you signed the CLA as a corporation, please sign in with your organization's credentials at https://identity.linuxfoundation.org/projects/cncf to be authorized.
• If you have done the above and are still having issues with the CLA being reported as unsigned, please log a ticket with the Linux Foundation Helpdesk: https://support.linuxfoundation.org/
• Should you encounter any issues with the Linux Foundation Helpdesk, send a message to the backup e-mail support address at: [email protected]

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. I understand the commands that are listed here.

[k8s-ci-robot]

Welcome @fishautumn!

It looks like this is your first PR to kubernetes-client/java 🎉. 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-client/java 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. 😃

[fishautumn]

/assign @brendandburns

[brendandburns]

I believe that the Go implementation of the informer also returns an update on re-list. I think we'd prefer to match that implementation for the principle of least suprise.

@yue9944882 wdyt?

[yue9944882]

I believe that the Go implementation of the informer also returns an update on re-list

the Go implementation doesn't actually perform any de-dup when dealing w/ two lists w/ the same resource version --- this can happen upon some relatively inactive resources like namespaces/serviceaccounts/... i wrote a test to show the behavior how delta-fifo processes identical lists, and i did the same on golang library and the result shows that they're working exactly the same. so i wonder if the memory leakage is caused by the abortion of the processor thread consuming the DeltaFIFO. @fishautumn can you share w/ us how you reproduce/locate the leakage?

but i think the proposed change in this pull is still valuable at least before we getting know the cause

[fishautumn]

in ReflectorRunnable.run(), it calls list method without resourceVersion. golang client Reflector.Run() calls list method with resourceVersion.

I suspect this is the reason of java client memory leak.

[fishautumn]

I suspect java client has issue with watching apis http readtimeout setting. BatchV1Api only applys timeout to URL parameter, not changing okhttp client options, while default okhttp client's readtimeout is 10 seconds. this speeds up the java client OOM.

[fishautumn]

@yue9944882 here is my steps to repro it:

1. create a informer for V1Job list
2. create a thread that creates a job every 10 seconds
3. after running for 17 hours, the program consumes all memory and spends all CPU for GC activity.

I guess when the list size is big enough that object producing speed (list size every 10 seconds) is higher than consuming speed (single thread removes object one by one from list), the memory will increase forever.

[fishautumn]

in ReflectorRunnable.run(), it calls list method without resourceVersion. golang client Reflector.Run() calls list method with resourceVersion.

I suspect this is the reason of java client memory leak.

I tried adding resourceVersion to list call, but it doesn't wok. according to api reference, resourceVersion is only working in watch call. list-job-v1-batch

[fishautumn]

so the main difference with golang implementation is the timeout setting. java client is 10 seconds, while golang client is 5~10 minutes (randomly).

this makes it requires 45 times of list size to repro the issue in golang client.

[yue9944882]

I tried adding resourceVersion to list call, but it doesn't wok. according to api reference, resourceVersion is only working in watch call. list-job-v1-batch
so the main difference with golang implementation is the timeout setting. java client is 10 seconds, while golang client is 5~10 minutes (randomly).
this makes it requires 45 times of list size to repro the issue in golang client.

@fishautumn thanks, the informer explicitly requires the okhttp client's read-timeout to be infinite to align w/ the golang library, see here. there's no such thing as read-time in the golang client. if that is the case, i'm leaning on adding another defensive check (probably upon constructing the informer instance) that the given ApiClient#OkHttpClient's read-time is set to 0. the current state of the pull is good to me. @brendandburns

/lgtm

[brendandburns]

SGTM, throw IllegalStateException or IllegalArgumentException probably.

[fishautumn]

• pre-check is adding to SharedInformerFactory. it throws IllegalArgumentException if ApiClient's timeout is not zero.
• there is a piece of code that is setting timeout to zero if not, but I suspect it's not working for watch call. I refined it.
• I suspect the timeout is also not working for GenericKubernetesApi, so I added a comment. fix me if I'm wrong.

[yue9944882]

/lgtm

[brendandburns]

/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: brendandburns, fishautumn

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 [brendandburns]

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