"padding" system test issues

Some of the checks in the "padding" system test appear to be very fragile:

Both of these job failures were triggered by the following check:

echo_i "checking that a padding server config should enforce TCP ($n)"
ret=0
n=`expr $n + 1`
$RNDCCMD 10.53.0.2 stats
opad=`grep  "EDNS padding option received" ns2/named.stats | \
    tail -1 | awk '{ print $1}'`
$DIG $DIGOPTS foo.example @10.53.0.4 > dig.out.test$n
$RNDCCMD 10.53.0.2 stats
npad=`grep  "EDNS padding option received" ns2/named.stats | \
    tail -1 | awk '{ print $1}'`
if [ "$opad" -ne "$npad" ]; then ret=1; fi
if [ $ret != 0 ]; then echo_i "failed"; fi
status=`expr $status + $ret`

What happens in the above two failed jobs is that (I believe due to QNAME minimization on ns3, which is queried shortly before the check quoted above), ns3 may send a TCP query for ns2.example/AAAA to ns2 after the first rndc stats invocation but before ns4 is queried for foo.example. This causes the "EDNS padding option received" counter to be different between the two rndc stats invocations in the check quoted above, triggering a false positive. This seems to be a very fragile method of checking whether ns4 sent a query with an EDNS padding option present.

The test also fails to preserve diagnostic information for later inspection.

Finally, I find the test descriptions quite confusing - EDNS padding does not enforce TCP, it requires TCP.

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information