Readiness probes fail if the path contains a querystring

[paulgrav]

What version of Knative?

1.10.1

Expected Behavior

We are trying to run ApolloServer in Knative. ApolloServer recommends the following healthcheck

https://your.server/graphql?query=%7B__typename%7D

We expect that ApolloServer should start when using the recommended healthcheck.

Actual Behavior

The QueueProxy fails to establish that the ApolloServer app is healthy. We observe that the healthchecks fail and the Knative app never becomes available.

Observing the pod resource we can see that the healthcheck is the following:

- name: SERVING_READINESS_PROBE                                                                                                                                                                                                                                                   
value: '{"httpGet":{"path":"/?query={__typename}","port":4000,"host":"127.0.0.1","scheme":"HTTP","httpHeaders":[{"name":"content-type","value":"application/json"},{"name":"K-Kubelet-Probe","value":"queue"}]},"initialDelaySeconds":5,"timeoutSeconds":1,"periodSeconds":10,"successThreshold":1,"failureThreshold":3}'

When we observe the app’s logs we can see that the healthcheck that is being made is:

/%3Fquery={__typename}

The above URL 404s.

Steps to Reproduce the Problem

Rather than recreating an example with Apollo, we’ve reduced the problem.

1. Deploy the following app: kn service update hello --image docker.io/library/nginx --port 80 --probe-readiness "http:localhost:80:/?foo=bar"

Note the following log lines:

user-container 127.0.0.1 - - [16/Aug/2023:10:35:20 +0000] "GET /%3Ffoo=bar HTTP/1.1" 404 153 "-" "kube-probe//" "-"

We’d expect to see the following:

user-container 127.0.0.1 - - [16/Aug/2023:10:35:20 +0000] "GET /?foo=bar HTTP/1.1" 200 153 "-" "kube-probe//" "-"

[dprotaso]

/triage accepted

[dprotaso]

Thanks for the report

[dprotaso]

/assign @nak3