sysstat on debian

[yager]

After the recent patching activity, I'm experiencing a failure with the sysstat package. In particular, /usr/lib/sysstat/sa1 now gives a Segmentation fault.

Anybody else? confirm or deny?

Linux 2.6.9 #2 SMP Sat Sep 30 02:31:06 EDT 2006 i686 GNU/Linux

[Rick]

I'll check it out. does strace give any insight to the seg fault?

[Rick]

/proc/net/sockstat seems to be missing. I'll find the solution to this, thanks for reporting it...

[yager]

Thanks, Rick!

open("/proc/interrupts", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/proc/diskstats", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/proc/sys/fs/super-max", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/proc/sys/fs/dquot-max", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/proc/sys/kernel/rtsig-max", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/proc/net/sockstat", O_RDONLY) = -1 ENOENT (No such file or directory)

[Rick]

I checked the kernel source code and that file was removed from the current stable kernel branch because it is not being virtualized. the VZ project is moving towards complete proc virtualization. So, that program will crash for now, but in the next kernel release, it will function. Any data retrieved from the file in the old kernel was incorrect.

[yager]

Thanks Rick. Clearly virtualization is tricky business, especially at the resource level.

Maybe this links to the low values currently reported for memory usage by the resource history tool, which worked fine when you set it up.

Also, .... are we now running debug code? I get lots of sluggish "lag" in the terminal response lately; though it started ("getting thick") before the reboot.

[Rick]

Theres several ways to account for memory usage. Previously, I was treating all of your vps's allocated memory pages as used memory. However, now we are accounting on actual written pages. For example you can malloc a 1GB chunk and never use it. In this case, no memory is accounted; previously it was. Some software like named like to allocate gobs of memory but never write to it. Its more of a theoretical decision, but this method is better for the customer anyway.

[yager]

One other change I've noticed which I suspect is from recent patchings is that ssh tunnels to my public IP no longer work. I've had to reconfig the pop host to listen on localhost as well, and tunnel to the loopback rather than the public IP.

(In testing, I also discovered that tunnels back to the source IP (my desktop) also don't work.)

No Action Required :)

[Rick]

It took 2 months to get fixed but at least it is. This will get applied on our next kernel update.

http://git.openvz.org/?p=linux-2.6.18-openvz;a=commit;h=42ec77227671bedfc7b51bbf3892014c f3501498