> I'm debugging an application (PHP witch pecl memcache) to see why > there is a small percent of timeouts when connecting to memcached. > Everything seemed to be ok until today where I found these values of > total_connections: > > STAT uptime 2174031 > STAT curr_connections 34 > STAT total_connections 1045845111 > > STAT uptime 2174032 > STAT curr_connections 47 > STAT total_connections 1045846625 > > In 1 second total_connections went from 1045845111 to 1045846625 > meaning 1514 new connections in one second. I check with one memcached > server for testing and don't see this behavior, total_connections > changes according to the requests I made to the server. So I'm > assuming that there is something wrong on the application. > Is ok if I'm taking total_connections value as a reference? Actually I > was monitoring curr_connections, which seems to be ok, and ignoring > the value of total_connections but now I'm monitoring both. Have you > seen this behavior on other servers? Which value is the one that I > need to compare with the maxconn parameter?
Not... quite sure what the confusion is here? I think you're looking for why you're getting timeouts? curr_connections == number of connected clients.. which is restricted by the maxconns limit. total_connections == each time a client connects this ticks up one. so connect, disconnect, connect, and it'll be at 2. if you're looking for if you've been running into maxconns "listen_disabled_num" is the counter for that. if it's nonzero you're running into problems. you seem to be connecting to it an awful lot. could also be causing churn in iptables or TmIME_WAIT buckets or whatever. Check dmesg, or try persistent connections.
