How to cleanup old router uuid definitions?

[#1]

Recently our deployments have started failing on our internal OPDK cluster with 6 routers/message processors. The deployment API is returning:

...    
}, {
      "status" : "undeployed",
      "type" : [ "router" ],
      "uUID" : "3322d9ce-66cd-4c84-9e1f-236518062ed4"
    }, {
      "status" : "undeployed",
      "type" : [ "router" ],
      "uUID" : "8ca2c2bf-982c-4ae1-8b8a-6a9003b1f041"
    }, {
      "status" : "undeployed",
      "type" : [ "router" ],
      "uUID" : "2f63ab90-221d-4054-a96a-66c3c79bdb64"
    }, {
      "error" : "com.apigee.zookeeper.ZooKeeperException{ code = zookeeper.ErrorSettingData, message = Error while setting data for path : /organizations/MYORG/environments/prod/apiproxies/MYPROXY/revisions/3/statuses/78ebd00d-2663-4596-a2de-27df6c47e93a/status, associated contexts = []}",
      "status" : "error",
      "type" : [ "router" ],
      "uUID" : "78ebd00d-2663-4596-a2de-27df6c47e93a"
    }, {
      "status" : "undeployed",
      "type" : [ "router" ],
      "uUID" : "a1802e7d-8c23-47e7-8abb-e32cb97c13f1"
    }, {
      "status" : "undeployed",
      "type" : [ "router" ],
      "uUID" : "4a4b2e88-c486-4381-8f40-ae627e197205"
    }, {
      "error" : "com.apigee.kernel.exceptions.spi.UncheckedException{ code = application.bootstrap.FailedToConfigure, message = Configuration failed, associated contexts = []}",
      "status" : "error",
      "type" : [ "router" ],
      "uUID" : "fe55b507-9399-40bf-9b2d-bd99ff088a25"
    } ],
    "state" : "error"
  } ],

There's deployment errors on two routers. But, it's listing 7 routers when we only have 6. The management server API GET /v1/o/MYORG/e/prod/servers returns a list of servers, where only 6 are routers.

We've tried manually deleting the extraneous router guid, restarting zookeeper, restarting routers/message processors, but the deploys still fail with the same error. It's trying to deploy to routers which don't exist.

How can we properly clean up these extraneous routers? Is there some cache in front of zookeeper that we need to clear?

This is blocking our ability to perform deployments.