-
Notifications
You must be signed in to change notification settings - Fork 819
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add gRPC load test for allocation service #1867
Add gRPC load test for allocation service #1867
Conversation
Build Failed 😱 Build Id: 06927514-2130-40bf-a55c-9e016d3065c7 To get permission to view the Cloud Build view, join the agones-discuss Google Group. |
Build Succeeded 👏 Build Id: 1d20b468-10e5-4162-a018-b7506d92511f The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Build Succeeded 👏 Build Id: 77dc3ef7-2f28-4be8-a24c-4c9020cccd9f The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
panic(err) | ||
} | ||
|
||
grpcClient := pb.NewAllocationServiceClient(conn) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should the gRPC client be created within the go routing for per client?
That would be a closer per client simulation.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I tried it inside the loop and performance was same. See the update
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the change. You should also add dial to the per client go routine.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That reduces the throughput. We can consider different test such as 5 MM client (e.g one per continent) with concurrent allocations using the same client
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I made the change. With 40 clients, we get few more errors (still reasonable). To allocate 5000 GSs with 40 clients, it takes ~70 secs. To increase throughput, we probably need more allocator service replicas
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think if you only use one connection, that connection goes to the same pod, therefore you are not testing multiple clients/pods.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think if you only use one connection, that connection goes to the same pod, therefore you are not testing multiple clients/pods.
This is an interesting question actually. But a couple of thoughts:
- I think you are right @pooneh-m - and we want to test across multiple pods. But I'm not sure if that's the requirement for this PR.
- Do we expect someone is going to have 50 matchmakers? Seems unlikely - so having a single client may actually make sense, since that's what most architectures will have (or at least, a very low number)
- (Separate to this PR) Should we look at how we can make it easier to better implement gRPC load balancing? Some resources:
- https://grpc.io/blog/grpc-load-balancing/
- https://blog.nobugware.com/post/2019/kubernetes_mesh_network_load_balancing_grpc_services/
- https://kubernetes.io/blog/2018/11/07/grpc-load-balancing-on-kubernetes-without-tears/
- https://itnext.io/proxyless-grpc-load-balancing-in-kubernetes-ca1a4797b742
- https://cloud.google.com/solutions/exposing-grpc-services-on-gke-using-envoy-proxy
If so, this sounds like something we should track in a separate issue for discussion.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think it is a good discussion.
The current load test client, creates one connection per client so it should leverage all the pods. The performance is more or less so keeping it as is.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Started an issue: #1872
f224beb
to
89cab54
Compare
Build Failed 😱 Build Id: c33e9c49-c98f-4e61-9ad0-57ec209c9a40 To get permission to view the Cloud Build view, join the agones-discuss Google Group. |
Linter fail:
|
Build Failed 😱 Build Id: 85e728d4-b108-4bfa-b389-c46bb3312047 To get permission to view the Cloud Build view, join the agones-discuss Google Group. |
Build Succeeded 👏 Build Id: d6fce878-c83e-45dd-954a-ab0bc7a90b7a The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Build Succeeded 👏 Build Id: 1a15fae9-f3cb-4e40-b4fa-19a820f2ceea The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Build Succeeded 👏 Build Id: c7888bc3-3fbe-4560-85de-08aac01d0618 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
77e6c68
to
9122913
Compare
Build Succeeded 👏 Build Id: 8cdcc050-5a6b-470a-914d-716eadf3bfa9 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Build Succeeded 👏 Build Id: 2204b1ae-acc5-4174-9bf0-ce634e95b0e5 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Other than that one small nit in the fleet.yaml, this is good to go for me! |
Build Succeeded 👏 Build Id: fef813d5-673d-45ff-9964-b0ad00a2a6c7 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
Build Succeeded 👏 Build Id: 78da6162-e170-4345-bfae-9d2294846234 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
LGTM! Anything left outstanding for you @pooneh-m or @roberthbailey ? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Thanks for adding the load test @ilkercelikyilmaz! Very useful.
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: ilkercelikyilmaz, pooneh-m 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:
Approvers can indicate their approval by writing |
New changes are detected. LGTM label has been removed. |
Build Succeeded 👏 Build Id: 57474ca2-9588-4533-9de3-58ea42371541 The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
What type of PR is this?
/kind documentation
What this PR does / Why we need it:
This adds a sample grpc load test client for the allocation service.
Which issue(s) this PR fixes:
Provides allocation load test client and instructions
Special notes for your reviewer: