Closed psheorangithub closed 5 months ago
We are experiencing the same issue with our Kubeflow deployment that uses Dex and LDAP for authentication. Upon attempting to access the Kubeflow URL, we face a recurring 403 error. The Kubeflow environment has 6 users, and they encounter this issue on the Kubeflow login page approximately every 4 to 5 days.
The following error is observed in the logs of the authservice-0
pod within the istio-system
namespace:
# kubectl logs authservice-0 -n istio-system
...
{"log":"time="..." level=error msg="Failed to save state in store: error trying to save session: input/output error" ip=... request=/jupyter/api/namespaces/.../notebooks\n","stream":"stderr","time":"..."}
We have inspected the resource usage of the authservice-0
pod and found no anomalies. CPU and memory consumption appear to be normal:
# kubectl top pod authservice-0 -n istio-system
NAME CPU(cores) MEMORY(bytes)
authservice-0 1m 8Mi
We have also checked the status of the persistent volume claim (PVC) associated with the authservice-0
pod and found that it contains the expected data:
# ls -lh /export/kubernetes/istio-system-authservice-pvc-pvc-4a8228e9-2b6d-456c-9a7a-f9fbf8d9a209/
total 112K
-rw-r--r-- 1 systemd-network tss 128K Jul 19 09:06 data.db
As a temporary workaround, we have tried deleting the authservice-0
pod, and it automatically gets recreated. After this, the Kubeflow Dex login page is accessible again.
Environment:
kfctl version
):minikube
): On-premisekubectl version
): v1.19.2/etc/os-release
): Ubuntu 20.04Any insights or suggestions to resolve this persistent 403 error on the Kubeflow login page would be highly appreciated.
Additional Content:
After setting the log level of oidc-authservice
to DEBUG, I rechecked the logs when the error occurred again. I discovered that the error is related to boltstore/reaper, which is responsible for releasing unnecessary resources, rather than using boltdb for session management.
2023/08/01 03:22:10 boltstore: remove expired sessions error: input/output error
time="2023-08-01T03:22:57Z" level=warning msg="Request doesn't have a valid session." ip=192.168.200.15 request=/logout
time="2023-08-01T03:22:57Z" level=error msg="Failed to save state in store: error trying to save session: input/output error" ip=192.168.200.15 request=/
I got the same issue.
First encounter with this problem in 70 days of creating the oidc-authservice
pod.
/close
oidc-authservice is deprecated and this discussion belongs to kubeflow/manifests
@juliusvonkohout: Closing this issue.
/kind bug
We have integrated kubeflow with OIDC flow(Heracles+LDAP)We are unable to login to kubeflow UI. GUI throws below error.
Access to kubeflow.aiwb-enc-data-cpu1.uscentral-prd-az3.k8s.int was deniedYou don't have authorization to view this page. HTTP ERROR 403
While checking the authservice pod logs, I see below error. It happens every couple of days.
2023/03/15 14:08:40 boltstore: remove expired sessions error: input/output error time="2023-03-15T14:06:29Z" level=error msg="Failed to save state in store: error trying to save session: input/output error" ip= request=/
The issue resolves after restarting authservice pod but it re-appear after every 10-15 days. We have checked the underlying PVC status, it looks healthy.
Can someone look into it and suggest what could be the cause
Environment:
kfctl version
):kubectl version
): 1.23/etc/os-release
): centos