Hi Matevz, Thanks for helping me dig into this. While following your questions I realized that the original address on first connection 169.228.130.92 somehow gets transmuted into 169.228.230.92 which is a nonexistent host, does not have a DNS record, either. I feel somewhat stupid for not spotting this earlier, sorry :( How can this happen? Does server tell the client to which IP to connect? >>> Yes, the server responds with IP addresses for a locate request >>> (historic stuff). The question is the server responding that way or is >>> the client mangling the address. This will be one tough cookie to track >>> down if it's the client. So, try using the old client in the same way >>> and see if the same thing happens there. If so, it's a server issue. If >>> not, it's a client issue (assuming a sample of 2 is enough :-) Andy > Also, what is the result of: > > ]==> netstat -nt | grep 169.228.230.92 > > when the connections are 'in progress'? It just shows the connection in established state. Tried several times to catch some short-lived sockets but had no luck. For master / fedora 19 client (connection timeout error) I see this (apparently going through ip6): matevz@desire matevz> netstat -ntp | pcregrep '169.228.[12]30.92' tcp6 0 1 132.239.186.42:49331 169.228.230.92:9940 SYN_SENT 23539/xrdfs tcp6 0 0 132.239.186.42:52532 169.228.130.92:9940 ESTABLISHED 23539/xrdfs matevz@desire matevz> ll /proc/23539/fd | grep socket lrwx------ 1 matevz zh 64 Dec 3 14:03 10 -> socket:[6225802] lrwx------ 1 matevz zh 64 Dec 3 14:03 9 -> socket:[6225797] matevz@desire matevz> cat /proc/net/tcp6 sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode 8: 0000000000000000FFFF00002ABAEF84:CC72 0000000000000000FFFF00005C82E4A9:26D4 01 00000000:00000000 00:00000000 00000000 411 0 6225797 1 ffff88058335cd80 20 4 24 10 -1 9: 0000000000000000FFFF00002ABAEF84:BFF1 0000000000000000FFFF00005CE6E4A9:26D4 02 00000001:00000000 01:00000176 00000004 411 0 6225802 2 ffff88058335ae80 1600 0 0 1 5 ^ | !!! how come this is different ?? How could I have missed this :) Apparently 169.228.130.92 (the correct ip) is somehow changed into 169.228.230.92 (non-existent). This is also why Andy got confused For 3.3.3 client on slc5 (no route to host) I only see the established socket in the fd list. And it's an ip4 socket, there is no /proc/net/tcp6 on machine at all. My program that connects to the host several times shows up in ip4, /proc/net/tcp. Thanks a lot & cheers, Matevz >>> Lukasz >>> >>> On 03.12.2013 11:39, Lukasz Janyst wrote: >>>> Hi Andy, >>>> >>>> no, b is not the problem. There was an issue like the one you >>>> mention, but it was introduced after migration to IPv6 in master only >>>> and fixed immediately after it was discovered. It was never introduced >>>> to the stable-3.3.x branch. >>>> >>>> Matevz, what you see is really strange. It looks like the system >>>> won't let you connect to the host twice... A router issue perhaps? Can >>>> you telnet twice to this host:port from your test box? >>>> >>>> Cheers, >>>> Lukasz <snip> ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-DEV list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1