Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-04-11 22:38:11 -0500 \ Jesse A Dubin: > > > > > I'm having a bit of a problem with my sync rate. It's never been > especially fast, but its even slower than usual. Slower than any of the > previous posts in the archive, so I felt it would be okay to post my > version of a common issue. (I hope I'm right.) > > This morning watchdog caused my primary to fail, and the secondary started > to take over. Before it could complete, the primary came back up, so I > have one node on the backup as primary, and the other 9 are on the main > server as primary. For about 18 hours now it's been resyncing, and there's > still a lot of time left. I understand that splitting the bandwidth among > all of these nodes will slow things down a bit, but I've only occasionally > seen the transer rate hit 124 K/sec. First: which version? We had a version out there that under certain circumstances would report in /proc/drbd it was still resyncing, but actually it was not, but stalled for some reason ... So you better make sure you tell us about your version of DRBD, and maybe your kernel version, distribution, and drbd.conf, maybe hardware... You did some dry runs, before you went production, no? What has been the sync rate then? Anyways. You have 10 configured devices. Do they share a physical storage device? If yes, then there are 10 processes with the highes available priority concurrently seeking all over your disk... Why did you not use sync-groups to serialize the synchronization runs? In case your version of DRBD does not yet support sync-groups, there is a "trick", by setting the min and max sync rate of all but one device to near zero, (like 0 and 1, or 1 and 2 K), and then as soon as one device finished sync, up them for one of the remaining devices... > Meanwhile my web server is down, so to say there's some urgency to my > getting this resolved is an understatement. well, you *should* be able to start the web services on the respective DRBD-Primary node. > Hopefully once this is figured out I can see where my data went. I > was expecting it to be present when the nodes stopped resyncing > (otherwise what's the point?), but it looks like they'll have to be > recovered from a backup. I don't understand this paragraph, what are you complaining about here? Lars Ellenberg