06.06.2019 17:07, Vladimir Sementsov-Ogievskiy wrote: > 06.06.2019 16:55, Eric Blake wrote: >> On 6/6/19 8:48 AM, Vladimir Sementsov-Ogievskiy wrote: >>> Hi all! >>> >>> Here is small new io API: blk_co_pcache, which does copy-on-read without >>> extra buffer for read data. This means that only parts that needs COR >>> will be actually read and only corresponding buffers allocated, no more. >>> >>> This allows to improve a bit block-stream and NBD_CMD_CACHE >> >> I'd really like to see qemu-io gain access to calling this command, so >> that we can add iotests coverage of this new feature. Note that the >> in-development libnbd >> (https://github.com/libguestfs/libnbd/commits/master) is also usable as >> an NBD client that can drive NBD_CMD_CACHE, although it's still new >> enough that we probably don't want to rely on it being available yet. >> > > Hmm, don't you think that blk_co_pcache sends NBD_CMD_CACHE if called on nbd > driver? > I didn't implement it. But may be I should.. > > May aim was only to avoid extra allocation and unnecessary reads. But if we > implement > full-featured io request, what should it do? > > On qcow2 with backing it should pull data from backing to top, like in > copy-on-read. > And for nbd it will send NBD_CMD_CACHE? > These semantics seems different, but why not? >
Any opinions here? Should I resend or could we use it as a first step, not touching external API but improving things a little bit? -- Best regards, Vladimir
