Date: Fri, 11 Jun 2004 15:04:06 +0100 From: Pete French <petefrench@keithprowse.com> To: petefrench@keithprowse.com, silby@silby.com Cc: freebsd-stable@freebsd.org Subject: Portrange randomized problems - a little bit more info... Message-ID: <E1BYmda-0004qE-1d@dilbert.firstcallgroup.co.uk> In-Reply-To: <20040603153507.A23800@odysseus.silby.com>
next in thread | previous in thread | raw e-mail | index | archive | help
O.K., I have been doing a lot of testing with this, though have not yet managed to generate a lot of useful data sadly. To summarise - with net.inet.ip.portrange.randomized set to 1 I am seeing failures of mysql connections from a machine back to a server on the same machine. These happen rarely, but are freqnet enough that on a webserver I am seeing a handful every hour. The effect is real - if I disable net.inet.ip.portrange.randomized then it goes away completely. I have now tried this on a number of different machines with different configurations, and all of them give the same results. The error appears to be tthat the TCP socket cannot connect. I have several webservers here which are load balanced so they are all taking an equal share of the incomming requests. If I enable the randomisation on one of them and compare netstat -n outout what I see is that the randomised machine has far more of the mysql connecions stuck in the TIME_WAIT state. Foir example I just did a snapshot under very light load - there are 64 connections in that state on the randomised webserver, comapred to 3 on the non-randomised one. Does this help track down the problem at all ? Although turning off the randomisation is a workable workaround, I am concerned that this is the symptom of some rather more fundamental bug in the TCP code. -pcf.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E1BYmda-0004qE-1d>