Hi Baptiste, That's very helpful.
I wish we could add your examples section to the official documentation because it helps clear up the maths. Komu W. On 04/11/2016, Baptiste <[email protected]> wrote: > On Fri, Nov 4, 2016 at 3:55 PM, komu wairagu <[email protected]> wrote: > >> Hi guys, >> >> I'm a little confused by the wording on the official haproxy ver1.6 >> documentation(https://cbonte.github.io/haproxy-dconv/1.6/ >> configuration.html) >> especially as concerning dns resolution. >> >> Here's a section of my haproxy config: >> >> ## start config >> >> resolvers my_dns_resolver >> #Google name servers >> nameserver dns1 8.8.8.8:53 >> nameserver dn2 8.8.4.4:53 >> resolve_retries 30 >> timeout retry 1s >> hold valid 2010 >> >> frontend my_frontend >> bind *:80 >> default_backend my_backend >> >> >> backend my_backend >> balance roundrobin >> server my_domain my_domain.com check resolvers my_dns_resolver inter >> 2000 >> >> ## end config >> >> So my question is, >> How often will haproxy (version 1.6) do a dns resolution? Is it every >> 2000 ms(health check interval) or >> every 10ms (2010 modulo 2000) ? >> >> The documenation says: >> >> `Note: since the name resolution is triggered by the health checks, a new >> resolution is triggered after <period> modulo the <inter> parameter >> of >> the healch check.` >> >> but I'm getting confused by the wording. >> I have two interpretations from reading that documentaion; >> 1. dns resolution is triggered by health checks and thus will happen >> (<period> modulo the <inter>) milliseconds after the health check. >> 2. dns resolution will happen every (<period> modulo the <inter>) >> milliseconds. >> >> So if we start at time 00.00.00.00(hh.mm.ss.ms) >> for case 1 above health check happens at 00.00.02.00 and first dns >> resolution happens at >> 00.00.02.10 and second one at 00.00.04.10 etc >> while for case 2 above health check happens at 00.00.02.00 and first >> dns resolution happens at >> 00.00.02.10 and second one at 00.00.02.20 etc >> >> >> thanks, >> Komu W. >> >> > > Hi Komu, > > Technically, HAProxy's internal scheduler will wake up the health check > task at <inter> period. > Before processing the check itself, HAProxy will check if the last > resolution result is obsolete, so it 'now > last-resolution + hold valid' > period. > > This means: > - smallest DNS resolution period will be <inter> if DNS resolution period > is lower than <inter> > - if DNS resolution period is greater than <inter>, then next DNS > resolution will be after X times <inter>, X being the rounded ratio between > <inter> and DNS resolution period > > With some examples: > - inter = 2s, DNS hold valid = 1s, then DNS resolution will happen every 2s > - inter = 2s, DNS hold valid = 5s, then DNS resolution will happen every 6s > - inter = 2s, DNS hold valid = 10s, then DNS resolution will happen every > 10s > > Baptiste >
