On Mon, Jan 03, 2005 at 12:14:27AM +0100, Goswin von Brederlow wrote: > Hmm, I guessed it would ignore read errors and fill the missing spots > with 0 bytes. Force the move from a faulty disk. > How do I do that with pvmove now? That was -i in LVM1.
Not supported by LVM2 yet I'm afraid - may need small kernel patch. >From memory, what I think happens is that the kernel's kcopyd gets errors on some blocks and so is never in a position to report 100% of the copy complete, but userspace pvmove just sits and waits for 100% to be reported, so waits for ever. Feel free to look at the code and propose a suitable way of handling this. I think the original plan was for the 'status' line from the mirror target to report the failures as well as what's in sync. Userspace might want to know exactly which extents failed - we considered holding them in a circular buffer which userspace could query to get the data out of if it was fast enough. Or it might be deducible from the log state if it reports 'copy complete' but not all extents are in sync. Or it could just write them out with kprintf like some other subsystems do. Alasdair -- [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
