On Wed, 10 Mar 2010 18:45:09 +0100, Henrik Nordström <henrik@xxxxxxxxxxxxxxxxxxx> wrote: > ons 2010-03-10 klockan 21:12 +1300 skrev Amos Jeffries: > >> Most likely. 3.1 adds IPv6 support where 3.0 would not even try to use >> it. Henrik is the one who would know if the "-3" patched package fixes >> the related issues from 3.1.0.16. > > 3.1.0.16-6 is the HTCP fix release, but introduced the IPv4-mapped-any > issue preventing transparent operation. > > 3.1.0.17-3 fixes the IPv4-mapped issue. Thanks. > >> FWIW: We are happily receiving Google IPv6 here with the code about to >> be released as 3.1.0.18. Available in daily snapshots now. > > 3.1.0.17-3 seems to work fine for me on Fedora 12, with IPv4-only > connectivity. (had to drop my IPv6 tunnel for other reasons). But I'll > give it some more in-depth testing of this issue later tonight after > dinner. > > I would suspect the issue is related to the massive amount of IPv6 > addresses Google have, possibly making Squid never getting to the point > that it tries IPv4. His trace earlier showed squid receiving the whole list of AAAA and A, then running through them in the proper order marking each bad as it went. Reporting error after the last IPv4 had been tried. Amos
