[Pdns-users] dnsdist marking downstream servers down: tuning issue?

Dez C dez at otenet.gr
Mon Jul 26 16:32:35 UTC 2021


Hello again,

just talking to myself here :) . After making sure that there was no 
issue with the backend resolvers, I noticed a lot of TCP resets from the 
testing subnet towards the DoH endpoint, so I went ahead and increased 
setMaxTCPClientThreads from 128 to 4096, I also increased file 
descriptors to 2^18 and finally upgraded to dnsdist 1.6.0 . Everything 
is OK now, I can easily reach 70k QpS (probably more)

cheers,
Yannis

On 7/25/21 7:41 PM, Dez C via Pdns-users wrote:
> Hi,
>
> I'm also controlling the backend servers and no such rate limit 
> exists. As I mentioned, each backend server can easily handle 60k QpS 
> and the numbers of queries that should be reaching the backend servers 
> from the performance testing, should not be more than 100-200 QpS per 
> backend server.
>
> The reason I sent my dnsdist configuration is so that people -more 
> experienced than me- could spot obvious omissions or errors.
>
> cheers
>
> On 7/24/21 11:41 AM, Winfried Angele via Pdns-users wrote:
>> Maybe there is some kind of ratelimt in the backend. Keep in mind 
>> that from the backend point of view all queries come from the same source
>>
>> Winfried
>>
>> Am 24. Juli 2021 10:28:49 MESZ schrieb Winfried Angele via Pdns-users 
>> <pdns-users at mailman.powerdns.com>:
>>
>>     Right. Therefore I would recommend rather to troubleshoot the backend
>>
>>     Winfried
>>
>>
>>     Am 23. Juli 2021 20:56:55 MESZ schrieb Dez C via Pdns-users
>>     <pdns-users at mailman.powerdns.com>:
>>
>>         Hi,
>>
>>         On 7/23/21 4:14 PM, Winfried Angele via Pdns-users wrote:
>>>         Does this only happen with DoH frontends? Did you try with
>>>         UDP frontends as well? Sounds like a bottleneck on your
>>>         backends imo.
>>
>>         I'm only using dnsdist for DoH so I haven't tried with
>>         UDP/53. I doubt it'll make a difference because the problem
>>         manifests itself when dnsdist is trying to query the backends
>>         (eventhough most replies already exist in dnsdist's cache)
>>
>>         cheers
>>
>>>
>>>         Winfried
>>>
>>>
>>>         Am 23. Juli 2021 13:32:39 MESZ schrieb Yannis via Pdns-users
>>>         <pdns-users at mailman.powerdns.com>:
>>>
>>>             hello,
>>>
>>>             We're using dnsdist (1.5.1 on Ubuntu 20.04, 16 cores, 32GB RAM) as a DoH
>>>             proxy/LB with normal DNS/53 resolvers as backend. This is a test
>>>             installation and we're trying to figure out the performance. It can
>>>             barely handle 1.5k QpS, which I consider pretty low (each backend
>>>             resolver can easily handle >60k QpS). It seems that each time the
>>>             queries rate is higher than ~1.5k, all backend servers are marked "DOWN"
>>>             until the rate goes below 1k. I understand that dnsdist marks the
>>>             servers down because it's not receiving a response on its healthcheck
>>>             query and I wonder why.
>>>
>>>             Should I increase "checkTimeout" and "checkInterval"? Should I use a
>>>             large number for "sockets"? Am I missing other tuning options or maybe
>>>             something more important?
>>>
>>>             Here's the relevant config (addresses, etc changed)
>>>
>>>             setLocal('0.0.0.0:5300')
>>>             addLocal('[::1]:5300')
>>>             controlSocket('local_public_address:xxxx')
>>>             setKey("XXX")
>>>             setConsoleACL('x.x.x.x/24')
>>>             NotRule(MaxQPSRule(50000))
>>>             setMaxUDPOutstanding(65535)
>>>             setMaxTCPClientThreads(128)
>>>             setMaxTCPQueuedConnections(10000)
>>>             setMaxTCPConnectionDuration(600)
>>>             PrimaryCache = newPacketCache(30000000, { keepStaleData=true,
>>>             maxTTL=86400, minTTL=0, numberOfShards=8, maxNegativeTTL=600, staleTTL=60 })
>>>             getPool(""):setCache(PrimaryCache)
>>>             addDOHLocal('10.2.3.4', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('10.2.3.4', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('10.2.3.4', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('10.2.3.4', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('2001:DB8::443', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('2001:DB8::443', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('2001:DB8::443', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             addDOHLocal('2001:DB8::443', 'cert.pem', 'key.key', "/dns-query", {
>>>             reusePort=true, minTLSVersion='tls1.2' })
>>>             newServer({address="2001:DB8::62", qps=10000})
>>>             newServer({address="2001:DB8::61", qps=10000})
>>>             newServer({address="2001:DB8::60", qps=10000})
>>>             newServer({address="2001:DB8::59", qps=10000})
>>>             newServer({address="2001:DB8::58", qps=10000})
>>>             newServer({address="2001:DB8::57", qps=10000})
>>>             newServer({address="2001:DB8::56", qps=10000})
>>>             newServer({address="2001:DB8::55", qps=10000})
>>>             newServer({address="2001:DB8::48", qps=10000})
>>>             newServer({address="2001:DB8::47", qps=10000})
>>>             newServer({address="10.10.10.62", qps=10000})
>>>             newServer({address="10.10.10.61", qps=10000})
>>>             newServer({address="10.10.10.60", qps=10000})
>>>             newServer({address="10.10.10.59", qps=10000})
>>>             newServer({address="10.10.10.58", qps=10000})
>>>             newServer({address="10.10.10.57", qps=10000})
>>>             newServer({address="10.10.10.56", qps=10000})
>>>             newServer({address="10.10.10.55", qps=10000})
>>>             newServer({address="10.10.10.48", qps=10000})
>>>             newServer({address="10.10.10.47", qps=10000})
>>>             setServerPolicy(roundrobin)
>>>
>>>             thanks in advance, I'd appreciate any input :)
>>>             ------------------------------------------------------------------------
>>>             Pdns-users mailing list
>>>             Pdns-users at mailman.powerdns.com
>>>             https://mailman.powerdns.com/mailman/listinfo/pdns-users  <https://mailman.powerdns.com/mailman/listinfo/pdns-users>
>>>
>>>
>>>         _______________________________________________
>>>         Pdns-users mailing list
>>>         Pdns-users at mailman.powerdns.com
>>>         https://mailman.powerdns.com/mailman/listinfo/pdns-users
>>
>>
>> -- 
>> Diese Nachricht wurde von meinem Android-Gerät mit K-9 Mail gesendet.
>>
>> _______________________________________________
>> Pdns-users mailing list
>> Pdns-users at mailman.powerdns.com
>> https://mailman.powerdns.com/mailman/listinfo/pdns-users
>
>
> _______________________________________________
> Pdns-users mailing list
> Pdns-users at mailman.powerdns.com
> https://mailman.powerdns.com/mailman/listinfo/pdns-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20210726/3b8c4350/attachment.htm>


More information about the Pdns-users mailing list