Hi guys, On Thu, Sep 12, 2013 at 09:55:54PM +0200, Lukas Tribus wrote: > Hi Sergey, > > >> Or is CLOSE_WAIT growing on the HAProxy box? If thats the case, please > >> clarify whether frontend or backend sockets are affected. > > Sorry for ambiguity, I'm talking about HAproxy box, with frontend > > sockets which accepting client connections from internet. We are using > > mode tcp due to https traffic (https doesn't terminated on HAproxy box > > and forwarded to backend server) > > This is not supposed to happen and could be a bug in haproxy. > Willy, any ideas here?
Sounds extremely strange. However such sequences where the client closes first are extremely rare since they cause many issues on the clients (sockets ending in TIME_WAIT and *not* reusable for 60 seconds). So it would be possible that we're on an old bug that nobody has ever met yet. > Are you seeing this on a production system only? Can you somehow reproduce > this without affecting your production traffic? > > I could think of: > - reproduce in strace > - "show info", "show sess" and "show stat" from the unix socket: > > http://cbonte.github.io/haproxy-dconv/configuration-1.4.html#9.2 Yes, the show sess should help a lot with all the flags that are dumped. And clearly a pcap could help as well ! Sergey, this time we don't need the data in the pcap file, so if you can take one, you do not need to use -s0, you'll get a smaller file and won't disclose possibly sensitive information. Best regards, Willy
