[dnsdist] dnsdist Drops, revisited

Fredrik Pettai pettai at sunet.se
Fri Mar 6 07:09:44 UTC 2020


Hi Michael,

Thanks for your comments, inline answers below.

> On 6 Mar 2020, at 05:42, Michael Van Der Beek <michael.van at antlabs.com> wrote:
> 
> Hi Fredrik,
> 
> Have you noticed this setting on dnsdist.
> setUDPTimeout(num)

Yes, I did, but I didn’t play around with that before I sent the email to the mailing list

> Set the maximum time dnsdist will wait for a response from a backend over UDP, in seconds. Defaults to 2
> I'm not sure if timeouts are classified as drops. My guess probably, because it didn't get a response in time.

Yes they are.

> Since your backend is a recursor. There are times that the recursor cannot reach or encounters a non-responsive authoritative server.  Unbound has an exponential backoff when querying such servers. I think it starts with 10s.
> https://nlnetlabs.nl/documentation/unbound/info-timeout/
> 
> I would suggest you set the dnsdist setUDPTImeout(10), frankly, if Unbound cannot respond to you in < 10 seconds, most likely the target authoritative server is not responding.

Good point, while I didn’t turn to the unbound documentation (thanks for the pointer) I played around with the UDPTimeout setting yesterday, 
first increasing to setUDPTImeout(5), which yielded better results in terms of Drops (and increased the latency) and then later to 15, just to be sure that unbound really should be done with queries, and noticed that the Drops became a lot less (and latency increase again). But as you suggest, setUDPTImeout(10) is probably the ultimate setting.  

> showServers()
#   Name                 Address                       State     Qps    Qlim Ord Wt    Queries   Drops Drate   Lat Outstanding Pools
0   worker1              127.0.0.1:53                     up    20.9       0   1  1    1527539    6814   0.0  65.4           2
1   worker2              127.0.0.1:53                     up    21.9       0   1  1    1553971    6910   2.0  59.4           3
2   worker3              [::1]:53                         up    15.9       0   1  1    1528862    6793   1.0  51.1           3
3   worker4              [::1]:53                         up    54.7       0   1  1    1523692    6880   1.0  53.3           3

Drops are now well below 1% 
So, I think we have the main answer there, UDP timeout discrepancy (vs resolver workloads)

> As to why one server has more drops then others.. 
> Assuming both servers have approximately the same number of queries/s
> So if the two servers have the same config (for unbound) and hardware.
> Note if the two servers are going via different ISPs then, their relative network speed can cause difference in response times.
> Then I would suggest, look at the some of these settings to see if they are the same.
> Note these are centos 7 settings. I'm not sure what the Debian equivalents are.
> net.core.rmem_default
> net.core.wmem_default
> net.core.rmem_max
> net.core.wmem_max

Right, those where already increased on all the systems prior to dnsdist 
(as per https://nlnetlabs.nl/documentation/unbound/howto-optimise/)

> net.netfilter.nf_conntrack_udp_timeout
> 
> Also generally, turn off connection tracking for udp/tcp packets via your firewall rules.
> https://kb.isc.org/docs/aa-01183

Yes, It’s already turned off via rules.
But I also turned off the FWs completely in my tests too, with no noticeable difference in Drops, (but I didn’t measure if there was any difference in latency though)

> Regards,
> 
> Michael
> 
> -----Original Message-----
> From: dnsdist <dnsdist-bounces at mailman.powerdns.com> On Behalf Of Fredrik Pettai via dnsdist
> Sent: Thursday, March 5, 2020 6:14 PM
> To: dnsdist at mailman.powerdns.com
> Subject: [dnsdist] dnsdist Drops, revisited
> 
> Hi list,
> 
> I’m curious on the “high" amount of Drops I see on one dnsdist 1.4.0 (debian derived packages) frontend compared to other(s) And I’m guessing the main reason is workload, which is different (services/servers use this resolver that Drops more).
> 
> I don’t find the “high” Drops numbers satisfying, but perhaps these numbers are about normal average? 
> Anyway, I'd would like to improve those numbers if possible. Here are some stats from two dnsdist frontends:
> 
>> showServers()
> #   Name                 Address                       State     Qps    Qlim Ord Wt    Queries   Drops Drate   Lat Outstanding Pools
> 0   worker1              127.0.0.1:53                     up    73.7       0   1  1     565950     278   0.0   0.5           0
> 1   worker2              [::1]:53                         up    55.7       0   1  1     584273     294   0.0   1.1           0
> 
> While one of our bigger servers doesn’t perform as well (in terms of Drops ratio):
> 
>> showServers()
> #   Name                 Address                       State     Qps    Qlim Ord Wt    Queries   Drops Drate   Lat Outstanding Pools
> 0   worker1              127.0.0.1:53                     up    43.8       0   1  1    1054047   12728   0.0  31.1           4
> 1   worker2              127.0.0.1:53                     up    43.8       0   1  1    1064823   12823   0.0  17.5           4
> 2   worker3              [::1]:53                         up    20.9       0   1  1    1054548   12773   0.0  38.5           2
> 3   worker4              [::1]:53                         up    35.8       0   1  1    1081502   12854   0.0  48.9           3
> 
> FW & DNSdist rules are almost none, and the same configuration on both the above systems (actually more active rules and even Lua-code on the “fast” dnsdist-system)
> 
> I just found one earlier thread on the topic, and it didn’t describe a way to improve the situation, just how to possibly look to see what the underlying issues might be...
> 
> http://powerdns.13854.n7.nabble.com/dnsdist-drops-packet-td11974.html
> (https://mailman.powerdns.com/pipermail/dnsdist/2016-January/000052.html)
> 
> dumpStats from the above server
> 
>> dumpStats()
> acl-drops              	          0    latency0-1             	    3620405
> cache-hits             	          0    latency1-10            	      59808
> cache-misses           	          0    latency10-50           	     132513
> cpu-sys-msec           	     749565    latency100-1000        	     386909
> cpu-user-msec          	     470696    latency50-100          	     101861
> downstream-send-errors 	          0    no-policy              	          0
> downstream-timeouts    	      52571    noncompliant-queries   	          0
> dyn-block-nmg-size     	          0    noncompliant-responses 	          0
> dyn-blocked            	          0    queries                	    4382032
> empty-queries          	          0    rdqueries              	    4382007
> fd-usage               	         42    real-memory-usage      	  315129856
> frontend-noerror       	    3254422    responses              	    4329454
> frontend-nxdomain      	     902996    rule-drop              	          0
> frontend-servfail      	     172012    rule-nxdomain          	          0
> latency-avg100         	      41936.3  rule-refused           	          0
> latency-avg1000        	      44165.7  rule-servfail          	          0
> latency-avg10000       	      43366.6  security-status        	          0
> latency-avg1000000     	      41994.4  self-answered          	          1
> latency-count          	    4329455    servfail-responses     	     172012
> latency-slow           	      27681    special-memory-usage   	   95940608
> latency-sum            	  172860695    trunc-failures         	          0
> 
>> topSlow(10, 1000)
>   1  uyrg.com.                                  69 46.9%
>   2  115.61.96.156.in-addr.arpa.                19 12.9%
>   3  nhu.edu.tw.                                 9  6.1%
>   4  nbkailan.com.                               8  5.4%
>   5  aikesi.com.                                 8  5.4%
>   6  168.122.238.45.in-addr.arpa.                6  4.1%
>   7  45-179-252-62-dynamic.proxyar.com.          4  2.7%
>   8  callforarticle.com.                         3  2.0%
>   9  default._domainkey.nhu.edu.tw.              3  2.0%
>  10  205.78.127.180.in-addr.arpa.                3  2.0%
>  11  Rest                                       15 10.2%
> 
> (Many are probably spammy relay IPs, sending domains, etc) 
> 
> Is there a way to optimise the dnsdist configuration, for instance making a slow path?
> either for the slow queries, or possibly the clients that ask those queries?
> 
> (Also, It’s unbound in the backend of all dnsdist frontend, and it’s caching heavily, also expired answers).
> 
> Re,
> /P
> 
> 
> 
> 
> 
> _______________________________________________
> dnsdist mailing list
> dnsdist at mailman.powerdns.com
> https://mailman.powerdns.com/mailman/listinfo/dnsdist



More information about the dnsdist mailing list