Ryan

hello. what operation do you take before this error happened? could you please provide more information?

> I have cherry pick @RyanZhaoXB commit to fix the failed device e2e and the e2e is green now, cc @fisherxu @gy95 @vincentgoat @Shelley-BaoYue PTAL OK. i will close https://github.com/kubeedge/kubeedge/pull/4213

> 设备管理是基于 K8S的CRD的，K8S的apiserver，没有批量创建CRD的接口，如果我要建几万个设备的话，就会很慢，我们的场景是这样的，我们做了一个创建设备模板和实例的界面，界面上一个批量创建设备的功能，是根据导入excel解析设备，来创建的设备实例，所以要循环调用创建CRD的接口，这个过程非常的长。 批量创建设备这个能力确实是需要的，我们后续方案设计中会做考虑，目前想到的方法是做一个单独的批量设备的CRD，用户只要预先准备好一个CR的yaml，执行一次kubectl create -f就可以实现批量创建设备的操作，但如果设备数量过多，在kube-apiserver和etcd这里仍然会有时间瓶颈。可能还有需要考虑的一点就是，批量创建的设备跟每个单独的设备之间是什么关系，是否批量设备的CR创建完其生命周期就结束了，每个设备后续可以单独管理。

the implementation of DMI for KubeEdge is proposed in https://github.com/kubeedge/kubeedge/pull/4013. the implementation of DMI for a demo mapper of modbus is proposed in https://github.com/kubeedge/mappers-go/pull/79

could you please provide more logs about this problem?

as far as i can see it, every time the metrics url is called, a new connection to the edge node will be added in the session. and when the...

i have tested it locally for dozens of times and they are all passed.

> yes. more discussion is needed. we can make it in https://github.com/kubeedge/kubeedge/issues/4126 . i will close this pr.

thanks for reporting this bug. the error is "failed to access database:sql: transaction has already been committed or rolled back " we have found this bug and we are fixing...

> > why not use daemonset to deploy the device mapper? If we have many device and edge node, then there maybe many deployment i think we should not use...