Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Tom Brown wrote: > On Mon, 11 May 2009, Ivars Strazdi wrote: > > >> Hi DRBD users, >> it looks like I am experiencing similar problem that other people have >> had. In a two node XEN cluster if the first node goes away, the second >> node panics with skb_copy_bits up in stack trace. >> >> Cluster nodes run stock Debian Lenny (2.6.26-2-xen-686) and DRBD 8.3.1. >> But I am absolutely positive have already disabled protocol offload >> options (ie rx,tx,sg) on both nodes... before DRBD starts (that is, in >> /etc/network/interfaces with post-up option) >> >> Did I miss anything? What else could I look for? >> Or should I wait for DRBD 8.3.2 which allegedly fixes this problem? >> > > Please either actually read the messages you are attempting to summarize > (or which somone else mis-summarized). > > 8.3.2 will not "fix" the problem. It will provide a parameter for turning > off sendpage support, which is a work-around. basically it provides a > clean module interface for the 1 line code change that Lars just posted. > > [I just want folks to be clear that upgrading from 8.3.1 to 8.3.2 isn't > going to make the problem go away... but it will give you a way to make a > change which DOES make the problem go away.] > > I had no luck with changing the scatter/gather and checksum parameters, > for me it was code-change or protocol A. > > And replicating the problem is simple enough if you have some test > machines you can work with... just ifdown the interface on the secondary, > type sync in the xen guest (domU) and wait about 30 seconds. > Tom, thanks for clarification. AFAIK protocol A cannot be used with "allow-two-primaries" option, I have to sacrifice live migration. Perhaps I am missing something [again] but I haven't seen in DRBD mailing list that 1 line code change you mentioned. Regards, Ivars