Hi Mike,
On 2023/5/24 10:28, Mike Snitzer wrote:
On Fri, May 19 2023 at 6:27P -0400,
Du Rui <durui@xxxxxxxxxxxxxxxxx> wrote:
OverlayBD is a novel layering block-level image format, which is design
for container, secure container and applicable to virtual machine,
published in USENIX ATC '20
https://www.usenix.org/system/files/atc20-li-huiba.pdf
OverlayBD already has a ContainerD non-core sub-project implementation
in userspace, as an accelerated container image service
https://github.com/containerd/accelerated-container-image
It could be much more efficient when do decompressing and mapping works
in the kernel with the framework of device-mapper, in many circumstances,
such as secure container runtime, mobile-devices, etc.
This patch contains a module, dm-overlaybd, provides two kinds of targets
dm-zfile and dm-lsmt, to expose a group of block-devices contains
OverlayBD image as a overlaid read-only block-device.
Signed-off-by: Du Rui <durui@xxxxxxxxxxxxxxxxx>
<snip, original patch here: [1] >
I appreciate that this work is being done with an eye toward
containerd "community" and standardization but based on my limited
research it appears that this format of OCI image storage/use is only
used by Alibaba? (but I could be wrong...)
Not necessarily Alibaba, actually OverlayBD solution is open-source to
containerd, at least I think it's an opensource project and I saw some
Microsoft Azure guys are also working on this.
But you'd do well to explain why the userspace solution isn't
acceptable. Are there security issues that moving the implementation
to kernel addresses?
OverlayBD user-space solution was actually the original Alibaba solution
widely used in Alibaba internally, and Nydus might be the another one
(used but limited, Ant group and Bytedance use Nydus more widely.) Since
Alibaba group is a big company, it's pretty normal to have two similiar
competing solutions together.
After I joined Alibaba, personally, I persuaded OverlayBD guys switching
from their stacked storage solution to a simple fs solution, because:
- It allows a simple on-disk format rather than a long storage stack
with a random fs, it increases the overall attack vector: which I
think this year LSF/MM already discuss about that;
- Different random fses cannot share page cache across images. IOWs,
many in-kernel fses actually doesn't suit for container image use
cases;
Also consider this one:
- Apart from the detailed on-disk design, this attempt is just a
read-only solution without 1) on-demand load; 2) write support;
- Very similar to the exist approaches:
dm-qcow2 https://lore.kernel.org/r/164846619932.251310.3668540533992131988.stgit@pro/
dm-vdo https://lore.kernel.org/r/20230523214539.226387-1-corwin@xxxxxxxxxx/
I also persuaded Nydus guys from their own format to erofs format, but
I failed to persuaded Overlaybd guys.
I also have doubts that this solution is _actually_ more performant
than a proper filesystem based solution that allows page cache sharing
of container image data across multiple containers.
Agreed.
There is an active discussion about, and active development effort
for, using overlayfs + erofs for container images. I'm reluctant to
merge this DM based container image approach without wider consensus
from other container stakeholders.
I'm too tired about these different container image solutions. I will
go on improve EROFS, and hopefully it will finally useful to everyone.
Thanks,
Gao Xiang
But short of reaching wider consensus on the need for these DM
targets: there is nothing preventing you from carrying these changes
in your alibaba kernel.
Mike
[1]: https://patchwork.kernel.org/project/dm-devel/patch/9505927dabc3b6695d62dfe1be371b12f5bdebf7.1684491648.git.durui@xxxxxxxxxxxxxxxxx/
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/dm-devel
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/dm-devel
