[Lxc-users] Many containers and too many open files

[Daniel Lezcano]

On 02/26/2011 08:45 AM, Daniel Lezcano wrote:
> On 02/25/2011 06:30 PM, Andre Nathan wrote:
>> On Fri, 2011-02-25 at 13:13 -0300, Andre Nathan wrote:
>>>> Google says you can setup these tables with the following values if you
>>>> encounter this problem.
>>>>
>>>> echo 256>   /proc/sys/net/ipv4/neigh/default/gc_thresh1
>>>> echo 512>   /proc/sys/net/ipv4/neigh/default/gc_thresh2
>>>> echo 1024>   /proc/sys/net/ipv4/neigh/default/gc_thresh3
>>>>
>>>> Do they fix your problem ?
>>> I found that, but even with much larger values for the thresholds I'm
>>> still getting the error.
>> Also tried running arpd on br0, no luck so far.
> I am trying to understand the routing cache code in the kernel to
> understand why the table is full.
>
> At the first glance, the hash table lock is sized with the number of
> cpus. For less than 4 cpus, it is 256 entries, for 4<= nrcpus<  8 it is
> 512 and for 8<= nrcpus<  16 it is 1024.
>
> How many cpus do you have on your hardware ?

Andre,


Can you add to the kernel boot option:

     rhash_entries=2097152


You should have in your console outpout:

     "IP route cache hash table entries: 524288 (order: 10, 4194304 bytes)"

and check if the neighbor table overflows.

Thanks
   -- Daniel