Date: Wed, 18 Nov 1998 13:40:48 -0500 From: "USWW (United States Wide Web)" <unix@usww.com> To: freebsd-isp@FreeBSD.ORG Subject: Cannot Fork/kvm read:Bad address Message-ID: <365314B0.51BE@usww.com>
next in thread | raw e-mail | index | archive | help
System Configuration: ASUS TPX4 CPU: Pentium (233.86-MHz 586-class CPU) Origin = "GenuineIntel" Id = 0x543 Stepping=3, Features=0x8001bf<FPU,VME,DE,PSE,TSC,MSR,MCE,CX8,MMX>, 256 meg pc 100, barracuda 4.3 scsi, wide controller. OS 2.2.7 FreeBSD, tried apache 1.2.6 reg and recompiled with mod perl. Increased maxusers to 253, sommaxconn=512 in kernel. Changed login.conf to infinity for all default, root, daemon processes and 256 meg for all mem limits. Additions to GENERIC Kernel are at the bottom of the email. I Still have a massive problem with "500 Internal Server Error" with kvm_read errors running cgi's. I maintain a total of 47 machines around the US all of them that have any anount of traffic are having these problems. I have 5-6 locations ready to bailout and go back to Linux if I can not solve it. I have sent Email to some groups with no response. I have read several hundred reports about this online with no resolve. Jordan or some one I need your help quickly. Tech support told me to email you since they did not have an up to date phone number. I have tried mod_perl with the same result. I tried the new 1.3.0, 1.3.1 and 1.3.2 and saw too many problems and went back to 1.2.6. What am I missing? What is slapping me in the face that I don't see? Here is a bit more information about the system problem it appears that kvm is coming in to play here. Once the cannot fork error starts showing up in the apache error log file. If I HUP apache it will work ok for about 30 minutes. In top it appears as though I have lost memory and vmstat is reporting major page faults. Under no conditions can I get the kvm_read: Bad Address to re-cycle. Is there a way while the system is running that I can re-init the kernels virtual memory? Server loads are running at average 2.05, 3.20, 2.15. I can fix the problem by rebooting the box it will go about 24 hour (2 million hits) before occuring again. Reminds me of a BRG (Bill Reboot Gates) NT box which has to be re-cycled daily because it looses memory pointers and looses rescources. Jordan: Have you tried asking the true experts when it comes to this topic? freebsd-isp@freebsd.org would be my very first avenue of recourse if our positions were reversed. I'd explain exactly what I was trying to implement (# of desired simultaneous users, type of CGI script(s) being run) along with the Apache version, OS info, hardware manifest and other such information as you have already provided me. The folks in freebsd-isp don't just contemplate such problems as an abstract challenge, as I do, they actually solve them every day. - Jordan ## Start of mods # changed maxuser because http://apache.org/docs/misc/perf-bsd44.html # also added SOMAXCONN NMBCLUSTERS cause of the same # changed SOMAX in /usr/include/sys/socket.h, /usr/src/sys/sys/socket.h # instead of defining here from http://apache.org/docs/misc/perf-bsd.html #options SOMAXCONN=256 # set w/sysctl -w kern.somaxconn=512 # in /etc/rc.local at boot options NMBCLUSTERS=4096 # mbuf clusters at 4096 options CHILD_MAX=512 options OPEN_MAX=512 #added the following because 500 cgi errors occur when loaded options "MAXDSIZ=(256*1024*1024)" options "DFLDSIZ=(256*1024*1024)" options "MAXMEM=(256*1024)" options MROUTING # Multicast routing options IPFIREWALL #firewall options IPFIREWALL_VERBOSE #print information about # dropped packets options "IPFIREWALL_VERBOSE_LIMIT=100" #limit verbosity options IPFIREWALL_DEFAULT_TO_ACCEPT #allow everything by default options IPDIVERT #divert sockets options TCPDEBUG options MFS #Memory filesystem To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?365314B0.51BE>