[dm-devel] Revert "dm space maps: don't reset space map allocation cursor when committing"
luomeng
luomeng12 at huawei.com
Tue Mar 8 04:03:31 UTC 2022
Yes, ramdisk itself is a thinly provisioned device. But We can't
restrict what disks users use. So if disk doesn't support discard, it
will cause this problem.
在 2022/3/2 0:06, Edward Thornber 写道:
> I'm having trouble understanding your issue. Does your ramdisk only
> allocate backing memory on demand? (ie. is the ramdisk itself a thinly
> provisioned device?). If so, not supporting discard seems to be the
> problem.
>
> Thinp makes no promises about where it will allocate your data. If you
> write a file, delete it, discard and then rewrite the file there is no
> guarantee that the file will be written to the same location.
>
> On Tue, Mar 1, 2022 at 2:08 PM luomeng <luomeng12 at huawei.com
> <mailto:luomeng12 at huawei.com>> wrote:
>
> Because thin-pool is storage over-commitment, one of the following
> scenarios exists: constantly create and delete file, then the search
> doesn't hit the end of the metadata device, but ramdisk hits the end
> (not support discard). So the cursor doesn't reset.
>
> 在 2022/2/28 23:37, Mike Snitzer 写道:
> > What you're saying doesn't make any sense. Especially when you
> > consider this last part of the commit message:
> > "Fix these issues by leaving the cursor alone, only resetting
> when the
> > search hits the end of the metadata device."
>
More information about the dm-devel
mailing list