runtime-rs: Change default block device driver from virtio-scsi to virtio-blk-* #11491
Conversation
| # Examples include: | ||
| # - virtio-blk-pci | ||
| # - virtio-blk-ccw | ||
| # - nvidmm |
There was a problem hiding this comment.
At the moment we're dropping scsi (which I think may be a mistake, as it tends to perform better than virtio-blk in cases where one can take advantage of iothreads) .... shouldn't we also remove the options that are scsi specific? (such as enable_iothreads)?
There was a problem hiding this comment.
I hope we're not dropping scis but just changing the default, we're using SCSI for NVME Fabrics since this performs better. Need to create a mental note to change this in the GPU use-case to have the SCSI per default.
There was a problem hiding this comment.
Sure, I was hoping to hear your voice about dropping virtio-scsi. I can say yes, in some cases, it's still useful.
With your thoughts, I agree with just changing the default from virtio-scsi to virtio-blk-*.
One more thing, currently, hotplugging virtio-scsi is not supported, shall we have to support hot-plugging virtio-scsi device in qemu-rs ?
There was a problem hiding this comment.
@Apokleos, yep, that should be supported in qemu-rs. :-)
|
With the current go-runtime, I can see the following: For x86_64: For s390x: However, nothing is actually attached to it, so we can remove them from the command line. Therefore, I think not defaulting to But one question remains (similar to Fabi’s):
|
Honestly, I am thinking
As I know that, the cold plugging virtio-scsi in runtime-rs has been implemented by Pavel @pmores but now the hotp-lugging mode is still unimplemented. Honestly, I am thinking if we are planning to support it or just stop supporting it as it is. |
|
cc @pmores FYI |
545b5cc to
f02b7ad
Compare
|
@Apokleos The opinions above have answered my question. Please ignore my comment. Thanks! |
|
Hey @zvonkok @fidencio as we disscussed, I have a summary:
Are there any concerns about defaulting to virtio-blk-* as the block device driver ? |
No more concerns from my side, @Apokleos! |
f02b7ad to
b0fd50f
Compare
|
Hi @fidencio @zvonkok @BbolroC @stevenhorsman If we agree on defaulting the block device driver to virtio-blk-* in runtime-rs, should we move it forward ? Thx a lot. |
When we run a kata pod with runtime-rs/qemu and with a default configuration toml, it will fail with error "unsupported driver type virtio-scsi". As virtio-scsi within runtime-rs is not so popular, we set default block device driver with `virtio-blk-*`. Fixes kata-containers#11488 Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
The previous description for the `block_device_driver` was inaccurate or outdated. This commit updates the documentation to provide a more precise explanation of its function. Fixes kata-containers#11488 Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
af6311e to
9cc14e4
Compare
|
I've rebased this one and will get it merged after the tests are finished (and passing). |
runtime-rs: Change block device driver defualt with virtio-blk-* and update docs
When we run a kata pod with runtime-rs/qemu and with a default configuration toml,
it will fail with error "unsupported driver type virtio-scsi".
As virtio-scsi within runtime-rs is not so popular, we set default block device driver with
virtio-blk-*.And as the previous description for the
block_device_driverwas inaccurate oroutdated. To make it clear for users, This PR also updates the documentation to
provide a more precise explanation of its function.
Fixes #11488
Signed-off-by: alex.lyn alex.lyn@antgroup.com