Graham Leggett <[EMAIL PROTECTED]> writes: > This bug describes something similar: > http://nagoya.apache.org/bugzilla/show_bug.cgi?id=23548
that is indeed quite similar, but not exactly the same. in that bug, LDAP connections end up in CLOSE_WAIT, probably because the LDAP server has decided to close the connection, but the apache child hasn't noticed yet. > Apparently the apr_reslist functions are the solution for this in apr-util. > > The new code does check for errors properly (in theory), so if a > connection has timed out, it will gracefully disconnect/cleanup that > connection before trying to use it again, which might solve > potentional problems you're having. that would work fine, I think, as long as the timeout calculation could be done on the apache side. that is, do not rely on the LDAP server timing out and handling the reconnect when apache notices that the connection has been closed. my problem was that I had a device that turned an apparently valid TCP connection into a tarpit. the LDAP server might decide to close a given connection after 60 minutes, but this device would, after 45 minutes (say) decide to drop the connection, and would do so in a way (by dropping frames on the floor) that was undetectable by apache. so it's hard for me to claim that this is a bug in apache, of course. and in our case, the problem was solved by having the LDAP server close the connection sooner, before the device did its evil business. still, it would seem to me to be a useful configuration tool to be able to say something like "don't keep connections to the LDAP server open longer than N seconds", or something approximating that, anyhow. -- joe
