Client Uses Wrong Scheduler IP's

[kalininsn]

Bug report:

Hello, we have a test k8s cluster with Dragonfly OSS there, we are seing behavior when client pod tries to connect to wrong IP's of scheduler on each image request from containerd.

As can be seen from screenshot, at first list of schedulers are actual, but then client tries to connect to some addresses that are not in that list. This might be an old scheduler addresses, but they are even from previous install of dragonfly. Even in database in table scheduler there are no such addresses.

Configurations for client and all the rest are almost default, that comes with latest helm chart, only thing that are updated are dynconfig refreshinterval(configured to 5s instead of 5m for client and 1m for seedClient)

Expected behavior:

Client uses actual IP addresses of Scheduler

How to reproduce it:

This happens for us after restart of all dragonfly pods(without restart we are not able to test, because it requires containerd config change and restart)

Environment:

• Client version: 1.2.0
• OS: Ubuntu 22.04
• Kernel: Linux cl17ep5p71980prfj71l-abus 5.15.0-163-generic chore: add clang to Dockerfile #173-Ubuntu SMP Tue Oct 14 17:51:00 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
• Others: K8s v1.32.1

[chlins]

@kalininsn Please try to specify the schedulerClusterID in the config, refer to https://d7y.io/docs/next/getting-started/quick-start/multi-cluster-kubernetes/.