Hi all, Andrew Hanushevsky wrote: > Hi Gregory, > > ----- Original Message ----- > From: "Gregory J. Sharp" <[log in to unmask]> > To: "Xrootd Mailing List" <[log in to unmask]> > Sent: Thursday, December 16, 2004 10:22 AM > Subject: Xrootd network problems > > > >>I have stared at the code for nearly a day, and I can't figure this one >>out. (Maybe 4 hours sleep last night just wasn't enough?) > > You really should be getting more sleep (yes, mom :-) > > >>My xrootd data director on sol199 produces the following messages for >>every connection. It looks to my naive eye that the connections are not >>being closed cleanly, but perhaps "link read error" is just a poor >>choice of error message. It occurs in two places in the code, so it >>isn't clear which piece of code produces the error. Anyway, things >>pretty much work okay while this is going on... > > Not only poor but wrong. There is a typo in one spot and the "==" should be > "!=". So, most of the time, 'link read error' means that the client closed > the connection and there was nothing to read. > > >>Then suddenly I get this in the xrootd data server log... lots of >>connections being made but never terminated. >> >>041216 12:49:14 020 XrootdXeq: User logged in as gregor.31754:17@lnx7108 >>041216 12:51:22 016 XrootdXeq: User logged in as gregor.31754:18@lnx7108 >>041216 12:53:22 018 XrootdXeq: User logged in as gregor.31754:19@lnx7108 >>041216 12:53:51 017 XrootdXeq: User logged in as gregor.31764:20@lnx7108 >>041216 12:55:51 019 XrootdXeq: User logged in as gregor.31764:21@lnx7108 >>041216 12:55:54 021 XrootdXeq: User logged in as gregor.31769:22@lnx7108 >>041216 12:55:56 022 XrootdXeq: User logged in as gregor.31773:23@lnx7108 >> >>Meanwhile, the client doing the connecting keeps printing >> >>041216 12:51:22 001 Xrd: ReadPartialAnswer Error reading msg from >>connmgr (server [sol199.lns.cornell.edu:1094]). >>041216 12:53:22 001 Xrd: ReadPartialAnswer Error reading msg from >>connmgr (server [sol199.lns.cornell.edu:1094]). >> >>until I kill it. > > Does this mean that the client is connecting multiple times and never > closing the connection? That is, the client error messages correlate with > the logins at the server. If so, this is a client error (i.e., it isn't > closing the connection). Fabrizio, could you verify this? > I did this particular test really many times. The client closes the connection and puts the physical connection in a sort of zombie state, waiting for the garbage collector to come after some time, not imemdiately. The messages dealing with the closure of the connection are printed at a higher log level (2), so in this case we should check with a higher debug level. Fabrizio