(Illustration by Gaich Muramatsu)
On Sun, Mar 14, 2004 at 07:18:40PM -0800, Steve Simitzis wrote: > On 03/14/04, Jan Harkes <jaharkes_at_cs.cmu.edu> wrote: > > This wouldn't happen all too often, a server restart should fix it. I'll > > have a look to see how I can avoid the race, either temporarily blocking > > the growth of the original volume, or by restarting the clone when the > > numbers don't match up. > > the problem i had was that restarting the server took about an hour, > which the server spent destroying a damaged backup volume during the > initial salvage run. :/ Yeah, when cloning the backup volume fails, it is destroyed. But because of the copy-on-write between original and backup volumes it should only have to destroy some of the metadata and not really touch any of the container files. I guess that updating the ftreedb file in /vicepa is probably one of the bigger slowdowns here, it explicitly fsyncs every time the linkcount is decremented but without profiling I can't really tell. An hour definitely sounds quite bad performance wise. JanReceived on 2004-03-15 01:19:21