DSA: unhealthy state leads to AdmissionError

[tkatila]

Describe the bug
When a workload starts quickly after the DSA plugin has commenced, the workload can fail to start with "UnexpectedAdmissionError".

To Reproduce

while true; do make e2e-dsa; done

Expected behavior
e2e-dsa works on a good system without issues or crashes.

Screenshots

      pod "dpdk" failed with status: 
          <v1.PodStatus>: 
              message: 'Pod was rejected: Allocate failed due to no healthy devices present; cannot
                allocate unhealthy devices dsa.intel.com/wq-user-dedicated, which is unexpected'
              phase: Failed
              qosClass: Burstable
              reason: UnexpectedAdmissionError
              startTime: "2025-07-02T09:47:31Z"
      {

System (please complete the following information):

• OS version: Ubuntu 24.04
• Kernel version: 6.8
• Device plugins version: 0.32.1
• Hardware info: GNR

Additional context
Since DSA and IAA use the same base structure, I suspect IAA also suffers from this. But I haven't seen it.

I tried adding a time.sleep(time.Second * 2) here: https://github.com/intel/intel-device-plugins-for-kubernetes/blob/main/test/e2e/dsa/dsa.go#L94 which allowed me to run e2e-dsa 20 times in a row. I suspect this is a timing issue in the api server or the plugin. The plugin doesn't seem to crash but I haven't seen logs from it so far.