Internal docker registry fails to pull the docker image

I have configured Rancher cluster with 3 nodes.
I am using the internal docker registry.

After the Rancher Pipeline execution is complete, only certain node can pull the docker image at deployment time, and the remaining nodes will have ErrImagePull errors.

502 Gateway error occurs and docker image cannot be pulled.

What kind of problem is this?

https://rancher.com/docs/rancher/v2.x/en/k8s-in-rancher/pipelines/

The internal Docker Registry is only accessible from cluster nodes and cannot be directly accessed by users.

The above documents suggest that all nodes in the cluster are accessible.

Configuring Persistent Data for Pipeline Components

I have finished setting up Persistent Data by referring to the document.

10.8.181.12 server works well.
But nodes at different addresses have problems.

Below is the event logs of the node that failed image pull.

Type Reason Message Last Seen
Normal BackOff Back-off pulling image “127.0.0.1:34293/rancher-node-test2:15” a few seconds ago
Warning Failed Error: ImagePullBackOff a few seconds ago
Warning Failed Failed to pull image “127.0.0.1:34293/rancher-node-test2:15”: rpc error: code = Unknown desc = failed to pull and unpack image “127.0.0.1:34293/rancher-node-test2:15”: failed to resolve reference “127.0.0.1:34293/rancher-node-test2:15”: unexpected status code [manifests 15]: 502 Bad Gateway a few seconds ago
Warning Failed Error: ErrImagePull a few seconds ago
Normal Pulling Pulling image “127.0.0.1:34293/rancher-node-test2:15” a minute ago
Normal Scheduled Successfully assigned p-8xm4j-pipeline/rancher-node-test2-app-7c79999dd5-7jcv2 to gcloud-seoul-2ae659bde981b0ad9c0f24835a8d1232 Invalid date

Below is the event logs of the node that succeeded in image pull.

Type Reason Message Last Seen
Normal Pulled Container image “127.0.0.1:34293/rancher-node-test2:15” already present on machine a minute ago
Normal Created Created container rancher-node-test2 a minute ago
Normal Started Started container rancher-node-test2 a minute ago
Normal Scheduled Successfully assigned p-8xm4j-pipeline/rancher-node-test2-app-7c79999dd5-5m72w to gcloud-seoul-78853ecbfdcbfed58ed8483bf205990b Invalid date

I want to know how to solve this problem.

Component Version
Rancher v2.4.4
User Interface v2.4.22
Helm v2.16.5-rancher1
Machine v0.15.0-rancher36
k3s v1.17.5+k3s1