dnsdist topSlow() does not work as expected

[paddg]

The topSlow() list often shows google.com at the top.

# dnsdist -e 'topSlow()'
   1  google.com.                                 3  2.1%
   2  c.8.f.3.5.5.e.f.f.f.7.4.4.b.0.4.8.7.5.5.4.2.0.0.e.8.0.1.6.0.8.2.ip6.arpa.    3  2.1%
   3  chat.hpeprint.com.                          3  2.1%
   4  _ldap._tcp.default-first-site-name._sites.dc._msdcs.vmv-ad.cc.private.hm.edu.    3  2.1%
   5  smart.360.cn.                               3  2.1%
   6  eidachs.duckdns.org.                        2  1.4%
   7  201.199.36.175.in-addr.arpa.                2  1.4%
   8  hpeguestlogin.hpecorp.net.                  2  1.4%
   9  eunt00100.kmt-eu.kmtg.net.                  2  1.4%
  10  sslvpn.co.san-diego.ca.us.                  2  1.4%
  11  Rest                                      121 82.9%

Even if I set limit very high:

# dnsdist -e 'topSlow(10,1000000)'
   1  google.com.                                 3 15.0%
   2  myitactionspro.glb.itcs.hpecorp.net.        2 10.0%
   3  c4.ibotv.tn.                                1  5.0%
   4  _ldap._tcp.pdc._msdcs.fk06ad1.pe.private.hm.edu.    1  5.0%
   5  _ldap._tcp.ae92f6f7-c9cb-489e-a2f3-e203d3491bf7.domains._msdcs.vmv-ad.cc.private.hm.edu.    1  5.0%
   6  _ldap._tcp.dc._msdcs.vmv-ad.cc.private.hm.edu.    1  5.0%
   7  gukieyckrmynjf.vmv-ad.cc.private.hm.edu.    1  5.0%
   8  _ldap._tcp.default-first-site-name._sites.dc._msdcs.speedport_w_724v_typ_a_05011603_06_003.vmv-ad.cc.private.hm.edu.    1  5.0%
   9  mail-northeuropeazon11021106.outbound.protection.outlook.com.daikieauxqco3jfjglnaxg3jme.zrd.dq.spamhaus.net.    1  5.0%
  10  auth.hpicorp.net.                           1  5.0%
  11  Rest                                        7 35.0%

The list is only empty from a limit of 4294968:

# dnsdist -e 'topSlow(10,4294968)'
   1  Rest                                        0 100.0%

Something seems to be wrong. Especially because I have set setUDPTimeout(5) which means 5000ms.

Is it possible that microseconds are being compared here rather than milliseconds? Although the implementation does not suggest that.

[paddg]

My assumption was wrong. On a timeout, std::numeric_limits<unsigned int>::max() is written to the ring buffer.

Some IRC conversation:

<ph1> winfried: tcpdump it and you'll see these google.com "queries" are some weird crap instead
<ph1> winfried: we see this all the time
<ph1> winfried: iirc a lot were actually IQUERY or something else 
<winfried> ph1: yes you're right! Many OpCode 2. That explains that google.com is in the list, thanks! But it doesn't explain the behavior of the "limit" setting of topSlow(). I would expect the list to be empty if limit > UDPTimeout.
<winfried> ph1: It could be that the query time stored in the ringbuffer for timeouts is 2^32-1=4.294.967.295 us. Because topSlow list is only empty from a "limit" of 4294968 ms
<winfried> cannot be a coincidence
<ph1> winfried: grepq() uses this as a timeout so I assume the other stuff would as well, if (entry.usec != std::numeric_limits<decltype(entry.usec)>::max()) {
<ph1> winfried: in which case your assumption about it being a very large number is right on

So everything is fine. You just have to know that std::numeric_limits<unsigned int>::max() (which is 4.294.967.295us or 4.294.967ms) is representing timeouts.

[paddg]

Because we get a lot of garbage that is dropped by the backend (Recursor) anyway, topSlow() is not very useful. It would be helpful if topSlow() could filter out the timeouts I think.

Or a division into topSlow() and topTimeout().

[paddg]

@rgacogne implemented a solution for this issue.