Am 05.01.2017 um 13:41 schrieb jtuchel@objektfabrik.de:
Am 05.01.17 um 13:34 schrieb Sven Van Caekenberghe:
I think this has to do with your particular server setup (Windows, Apache on Windows ?), there normally should not be any real difference (caching of DNS results). I am using localhost as upstream server reference everywhere (Linux mostly Nginx).
I guess youe are right: I just tried in our environment and the result is not distinguishable from the initial response times. We never had these problems, however.
And I'm glad that after 23 years of doing networking stuff I developed a gut feeling what might be wrong :) The problem here (I think) is precedence of ipv6 over ipv4. Modern systems are configured in a way to prefer ipv6 over v4. So while resolving the name localhost the loopback address ::1 is resolved making the apache proxy connecting to ::1. While pharo does not have ipv6 support the connection fails. In case of that in a ipv6 scenario the ipv4 alternative is tried making the call to 127.0.0.1 instead. Shouldn't be a reason taking so long so it might have to do with the apache proxy module code. There are multiple options to prevent that: 1. you make the use of localhost explicit by using 127.0.0.1 which here means localhost(ipv4) 2. You do echo 'precedence ::ffff:0:0/96 100' >> /etc/gai.conf to switch your system to ipv4 precedence 3. You remove ::1 localhost from the file /etc/hosts 4. You rename ::1 localhost to ::1 ip6-localhost In an ip dual-stack scenario there are these kind of problems. Hope that clarifies it! Norbert