Lxc Memory limit

[yena]

Hello,
can i modify the lxc behavior when the container reach the ram limit ?
Now on some containers i see that mysql is often killed:

Host syslog:
------------------------------------------------------------
May 8 08:07:30 nodo3 kernel: [1504328.952937] Task in /lxc/343/ns killed as a result of limit of /lxc/343
May 8 08:07:30 nodo3 kernel: [1504328.952942] memory: usage 2004860kB, limit 3145728kB, failcnt 0
May 8 08:07:30 nodo3 kernel: [1504328.952943] memory+swap: usage 3145728kB, limit 3145728kB, failcnt 61576889
May 8 08:07:30 nodo3 kernel: [1504328.952944] kmem: usage 0kB, limit 9007199254740988kB, failcnt 0
---------------------------------------------------------------

I have noticed it migrating some Openvz container on a new Lxc ..

Thanks!

 

[Rhinox]

I do not understand how exactly you'd like to modify it. What should kernel do when VM reached its limit of ram+swap usage, and yet some process wants to allocate more memory?

 

[guletz]

I do not understand how exactly you'd like to modify it. What should kernel do when VM reached its limit of ram+swap usage, and yet some process wants to allocate more memory?

He expect that when the memory is full, that any request for new memory allocation to be denied. In any case he did not want to see a process to be killed. I do not know how it is normal to be in lxc, but I am very sure that is true 100 % for openvz. So, the @yena question is not so dummy.... if someone read it carefully

 

[Rhinox]

He expect that when the memory is full, that any request for new memory allocation to be denied...

Fortunately, memory-management in linux works in much more sophisticated way...

 

[yena]

Exactly, i would like to know if i can i denied memory access without kill the process ..
A workoround should be a monit process on the guest ...
Thanks

 

[guletz]

You can do something, if you use zfs for this CTs with mysql. So, do you use zfs for them ?

 

[guletz]

Fortunately, memory-management in linux works in much more sophisticated way...

Thx.