[Systems] Freedom's root filesystem full

[Bernie Innocenti]

On 09/29/2014 03:10 AM, Sam P. wrote:
> Hi Bernie,
> 
> On Sep 29, 2014 12:58 AM, "Bernie Innocenti" <bernie at codewiz.org
> <mailto:bernie at codewiz.org>> wrote:
>>
>> Hello Sam,
>>
>> is the stuff in /var/lib/docker2 still needed? The root filesystem is
>> almost full.
> 
> No... I migrated all the stuff in the docker folder to a new partition
> by using that as temp folder... You can delete it.

I recursively deleted everything under /var/lib/docker2, but there was
still an aufs mounted under /var/lib/docker2, and it looks like it was
still in use by docker! I unmounted it

Can you please check that everything still works as intended?

In case I accidentally deleted something live, we should be able to
recover it from our daily backup.


>> Also, the two filesystems you mounted for docker instances are not in
>> /etc/fstab. You might want to add them.
> 
> I will be away from my laptop for a while now, but I will do this ;)

Ok.

Another question: which docker package are we using on freedom? We seem
to be using lxc-docker-1.2.0, but there were leftover config files from
an older docker.io version 0.9.1. I purged them to avoid confusion.


>> One last question: do we specifically need btrfs for snapshots or some
>> other feature? If not, I'd rather keep using ext4 exclusively on all
>> servers to reduce the number of kernel bugs we hit.
> 
> I will move back to ext4.  The only reason I used btrfs was some
> magazine article praising it as the filesystem of the future.

I've been following btrfs development for some 7 years, but it's been a
disappointment. It's not really... better. It's much slower than ext4
for common workloads and it's just barely starting to become stable now.
No distribution (other than Oracle's RHEL ripoff) made it the default,
and RHEL7 just switched to xfs, showing that RH is not interested in btrfs.

But btrfs has COW snapshots, which make it perfect for some
applications. Have you seen Lennart Poettering's b new package
management proposal?


>> It may sound silly, but constraining the amount of kernel code we run is
>> important for the uptime of dom0s. For example, justice just remounted
>> its root filesystem read-only and we don't even know why. The more
>> filesystems, the more different behaviors we observe.
> 
> Good idea!  I will try and follow it.
> 
> Sam
> 
> BTW: I'm not familiar with the h/w of freedom... But is it possible to
> make / bigger?  10GB is like tiny!

The root filesystem is needed for crash recovery, so it's not intened
to store anything other the base system.

Moreover, mounting per-application filesystems is a simple way to
isolate space usage, fragmentation and I/O bandwidth. It makes the whole
system easier to manage.

-- 
 _ // Bernie Innocenti
 \X/  http://codewiz.org