kswapd0 consuming 50% of cpu

[JD]

On 11/26/2012 12:01 PM, Bruno Wolff III wrote:
> On Mon, Nov 26, 2012 at 10:19:05 -0700,
>   JD <jd1008 at gmail.com> wrote:
>>
>> This is why I decided to download source of kernel 3.7.0-rc6.git from 
>> fc19 and build it on my fc16, in the hopes of a better kernel. But 
>> no.... it has the kswap bug eating 50% of cpu.
>
> So was this happening with both 3.7 and 3.6 kernels? The wording above 
> suggests that it didn't happen until you tried a 3.7 kernel, but 
> originally you said it happened with a 3.6.6 kernel. If the problem 
> happens in 3.6 and is related to what is going on in 3.7 that info 
> might be relevant to solving the problem.
The 50% cpu consumption by kswapd appeared first in 3.7
The Read/Write failures started in 3.6.X kernels, but do not appear in 3.7.
3.7  is doing the right thing by first issuing a hard reset to the sleeping
drives before reading or committing the journals. 3.6.X was not sending any
hard resets to the sleeping drives.
I will be testing 3.6.7 to see if either of these issues is still present.
>
>> It is obvious that hardly any extensive testing goes into fedora 
>> before release, in the attitude that the lab rats (us) will do that.
>> Has this been the culture of the fedora developers form the beginning?
>
> For the kernel there is a lot of different hardware out in the wild, 
> so kernel developers may not see an issue affecting only some 
> hardware. That's why testing of development kernels is encouraged. 
> Lots of people do this this testing and since Fedora is pretty close 
> to upstream, Fedora benefits from this testing. Some of us Fedora 
> users also test the development kernels and provide feedback (as in 
> the kswapd case, module signing issues and other issues) before the 
> kernels get into stable updates or a release.
>
But Bruno, the "not sending a hard reset to sleeping drives" is in all 
3.6.X kernels.
You would think they would have caught it rather than propagate it.
I can tell you it did lead to data loss for me. Some of it was important 
to me.