Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 13 Jan 2013, at 12:49, Lars Ellenberg <lars.ellenberg at linbit.com> wrote: > On Sat, Jan 12, 2013 at 10:15:18PM +0000, Robin Kearney wrote: >> Hi, >> >> I've installed drbd version 8.3.13 on a couple of machines running centos >> 5.8. I already have data on a logical volume on the primary and I had no >> space for a meta disk, so I've done the following: >> >> * umount my data LV >> * full fsck of the LV >> * resize it down by 1GB >> * mount it successfully and write to it for a few days without error >> >> And then I created a second 1GB LV out of the same VG for later use as the >> drbd meta disk. >> >> And then tonight I did the following: >> >> * unmounted my LV >> * configured drbd to use meta data on a second LV (created out of the 1GB >> now free in the VG) > > > And exactly *how* did you configured drbd to do so? > I guess you used the "indexed fixed size meta data"? > meta-disk /dev/whatever[0]; > That tells DRBD to use the first (index [0]) 128 MiB of that device > as its meta data. > 128 MiB (minus a few sectors) of bitmap > is just enough for about 4 TiB of backend storage. > Which is what you are describing below. > > You want > flexible-meta-disk /dev/whatever; > No index there, it uses as much of that device as is available > respectively necessary to cover the backend device. > > One GiB of bitmap is enough to cover 32 TiB of backend storage. Hi, yep that is exactly how I configured it and the solution was to do as you said. Due to list moderation lag, I'd fixed it just before your reply came through. Is it worth having this mentioned in the section of the docs where meta data is explained? And is it desirable behaviour to start a sync and to present the device as usable, even though the meta data is insufficiently sized? r.