-
Notifications
You must be signed in to change notification settings - Fork 660
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
[Docs] The GCP deployment docs should fix/remove the instructions to use the GCP Ingress controller #3730
Comments
Flyteadmin does have an HTTP health check that should work: https://github.com/flyteorg/flyteadmin/blob/3379031b05bec09df6b7bb1bd99d499cf900b268/pkg/server/service.go#L175 Unfortunately, it currently doesn't check for much: https://github.com/flyteorg/flyteadmin/blob/3379031b05bec09df6b7bb1bd99d499cf900b268/pkg/server/service.go#L158. But this should work well with the existing GCP LB health checks, in theory, and we can continue to make it better. |
Also @fg91 while it's true, the Having said so, there's work in progress to publish a reference implementation for GCP, and we'll incorporate feedback there |
It would be really great if the reference implementation would use GCP Identity Aware Proxy instead of letting Flyte serve the login screen. |
Ah interesting, I wasn't aware that this guide is not part of the latest docs anymore. |
The manual instructions were removed from In terms of the content of that section, the practical experience shows that probably The ideal approach on GCP would be to use the work you did to enable IAP support and add it to the docs. That effort will be tracked here |
Description
The SSL Certificate section in the GCP deployment guide documents both the use of a GKE
ManagedCertificate
in combination with the GKE ingress controller as well as the use of cert manager with the nginx ingress controller.However, when using the GCP ingress, the
flyteadmin
backend will always be shown as unhealthy (even though it isn't) as the GKE ingress controller cannot interpret probes of typeexec.command
which flyteadmin uses and which are recommended for gRPC services. Instead, the GKE ingress would need to either get a path + port where the app returns 200 if it is healthy or it needs to be given a TCP port where it will try to open a socket. (The way the GKE ingress checks for health can be customized with a so-called backend config. (In the Flyte Slack, users have reported that they used this mechanism to fix the broken health check.)The documentation should either provide instructions for using the GKE ingress controller that work out of the box or remove references to it and instead only recommend the nginx ingress controller.
Are you sure this issue hasn't been raised already?
Have you read the Code of Conduct?
The text was updated successfully, but these errors were encountered: