sysinfo virtualisation (totalram)

[zerkms]

I just posted #200 specifically about sysinfo.uptime, but now I found that at least some other fields are not virtualised either

#include <stdio.h>
#include <sys/sysinfo.h>

int main ()
{
 struct sysinfo si;
 sysinfo (&si);
 printf ("totalram %lu\n", si.totalram);
 return 0;
}

This outputs the total amount of memory the host machine has, not the container.

totalram 8369901568

/proc/meminfo has the correct numbers.

MemTotal:        1048576 kB

[hallyn]

Sorry, sysinfo cannot be handled by lxcfs.

[zerkms]

@hallyn but openvz could do that, or am I confusing something?

[hallyn]

@hallyn but openvz could do that, or am I confusing something?

Nope you're right, but it did that by having a custom kernel. See the do_sysinfo() function at https://github.com/OpenVZ/vzkernel/blob/branch-rh7-3.10.0-123.1.2-ovz/kernel/sys.c lxcfs only helps programs which use the procfs to get their info. If you can succeed in getting a patch like openvz's into upstream, that would be terrific.

[zerkms]

Oh, I see now. Thanks for the clarification!

[hallyn]

Note depending on your workload, one option would be to use seccomp with SECCOMP_RET_TRACE to have a ptracer intercept the syscall exit and fix up the results :)

[zerkms]

It actually is zabbix that does it, so I think it's time to report them when I have enough time to check what they have in the master branch.

[hallyn]

Right - the actual using seccomp-ptrace to fixup the results would be pretty trivial, the problem is that the container must be traced for that to work, which in turn means it can't do any tracing itself. I've not looked at zabbix source so don't know whether that would be a problem.