christophe.varoqui@xxxxxxx wrote:
Selon Arjan van de Ven <arjan@xxxxxxxxxxxxx>:
On Mon, 2004-10-25 at 00:14 +0200, christophe varoqui wrote:
* multipath & multipathd now use the same path checkers.
Consequence being multipath now need sg loaded
why sg and not using SG_IO on the actual devices ?
mp-tools are 2.6 only anyway, right ?
Glad you notice :)
In fact all the sg stuff I use is done with the SG_IO ioctl (see
libcheckers/*.c)
But somehow, it doesn't work as expected. I don't remember how exactly it failed
in my tests but it did ... I remembered receiving bad path status and
experiencing blocked checkers.
I would really like someone experienced put his nose in this stuff. If you want,
I can code up a flag to switch the mp-tools ioctl submission between blockdev
and sgdev.
Note you don't need to have real multipathed hardware to run and test the thing
: parallel SCSI hw suffice and is seen as monopath.
A significant patch from Patrick Mansfield concerning the
error reporting of the block layer SG_IO ioctl is present
in lk 2.6.10-rc1 . It was not present in lk 2.6.9 and earlier.
Doug Gilbert