krb5 should not try to access unavailable kdc too often

Curently, AS-REQ acts like this:

  try {
     send AS-REQ and wait for AS-REP
  } catch (KRB-ERROR needs PREAUTH) {
     send AS-REQ with PREAUTH and wait for AS-REP
  }

and the send-wait process looks like:

  for (all KDCs configed) {
     for (try 3 times) {
        try {
           send AS-REQ and wait for AS-REP
           return
        } catch (IOError) {
           continue
        }
     }
  }

The two processes are independent, which means when the first configured KDC is not accessible anymore, something like this is performed:

  1. Send AS-REQ to KDC1
  2. Send AS-REQ to KDC1
  3. Send AS-REQ to KDC1
  4. Send AS-REQ to KDC2
     KDC2 replies: KRB-ERROR needs PREAUTH
  5. Send AS-REQ with PREAUTH to KDC1
  6. Send AS-REQ with PREAUTH to KDC1
  7. Send AS-REQ with PREAUTH to KDC1
  8. Send AS-REQ with PREAUTH to KDC2
     KDC2 replies: AS-REP

Here, request #5-#7 is a waste of time.

Suggestion: maintaining a list of KDCs with the initial order according to config. When any of them is not accessible at run time, move it to the last place.