| Emiliano Gabrielli wrote: | for every image elog has to serve one can see something similar to the above ... lot of time lost in selects.. then a lot of data (serving an image I suppose), then a lot of time in select again and again ... untill everything is sent, in a couple of minutes or more 
Maybe an issue related to the dns search you introduced in order to guess the correct host name ?? .. |
This is strange to me, since I did not change anything which could slow down the server this much. The dns search your mentioned is only evaluated once on startup of elogd, so it cannot be the cause. The select() statements with Timeouts are normal. If there is no HTTP request (elogd is idling), the select should time out after one second, to be able to check a changed config file for example. If a HTTP request arrives, the select() call is immediately terminated and the request served.
There is however some problem with DNS server which I saw on midas.psi.ch. If the DNS host name resolution is slow due to a slow DNS server, this could slow down elogd considerably significantly, but only occasionally. I saw elogd hanging on midas.psi.ch like once or twice a day for ~30 seconds.
I order to address this problem, I imlemented a global flag "resolve host names = 0|1". The default is "0", which means that elogd does not contact the DNS server, and rather save the raw IP address in log files etc.
Can you check the CVS version and see if it makes any difference? |