Stale-Read load is unbalance in TiKV

[crazycs520]

TiDB Cluster Topology

tidb_servers:
  - host: 192.168.78.2
      labels: { zone: "z1", host: "tidb0" }
tikv_servers:
  - host: 192.168.80.2
    config:
      server.labels: { zone: "z1", host: "tikv0" }
  - host: 192.168.80.3
    config:
      server.labels: { zone: "z1", host: "tikv1" }
  - host: 192.168.80.4
    config:
      server.labels: { zone: "z2", host: "tikv2" }

Then send stale-read point-select SQL request to tidb0, then the stale-read get request in TiKV should be balanced in tikv0 and tikv1, and doesn't have stale-read get request in tikv2. But in TiDB v6.5.3, the stale-read request in tikv0 and tikv1 is unbalanced. Following is the TiKV metrics, as you can see, the stale-read get request on tikv0 is twice as high as on tikv1, which is not as expected, and there is no stale-read get request on tikv2, which is expected.