Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2004-06-30 11:42:56 -0400 \ Jeff Tucker: > Hi, guys. I had several questions about DRBD that should > hopefully be easy to answer. For background, I'm bringing up a > new system with a pair of identical servers. I intend to > failover a RAID-0 array created from 12 SCSI drives. The drives > are configured as one big array and the entire thing will be > replicated. I'd prefer to use the 2.6 kernel and I'm currently > testing using 0.7 development releases of DRBD. > > - Since I'm failing over the entire /dev/md1 array, do I need to > specify a disk-size in my drbd.conf? I intend to put the > metadata on the array as well. The filesystem will be added > after the DRBD device is up (in other words, mkreiserfs > /dev/nb0). So, I think I want the reiserfs filesystem to take up > most of the md1 device and the metadata to take up the rest. > Does this happen automatically if I just say to put the metadata > on that same device and don't specify a disk-size? yes. on both nodes: /etc/drbd.conf: device /dev/nb0; disk /dev/md1; meta-data internal; (two times, in both host sections) drbdadm up all on chosen node: drbdadm primary your-drbd-resource-name mkfs ... > - I rebooted the primary to test a failover. The system failed > over to the secondary just fine. When the primary came up and > the units started to sync, I got a kernel panic due to a > Reiserfs journal-601 error that said it was trying to write past > the end of the device. I meant to save the actual numbers, but > it wasn't even close. It said the size was something like 400000 > and it was trying to write to 1500000. Could that be caused by a > lack of the disk-size? I was running the 20040528 snapshot at > the time but have since updated to the recent release candidate > 1. In my testing, the system usually fails over fine and > rebuilds. It was just this one time I saw an error. everything before -pre8 == rc1 is likely to corrupt data on resync. > - The drives being failed over are a RAID-0 array. If a drive > fails, I'll be replacing it. This means that a bunch of data out > of the middle of the array goes away. When I bring that machine > back online, I can't just write the data that the primary has > received while it has been offline, I need to write all that > data plus everything that was on the now-replaced drive. > Will DRBD handle this automatically? you can set on-io-error Detach; this will detach the device, and the next time some resync takes place it should be the full device, including that "bunch of data out of the middle". > Will I need to force a full resync somehow? sometimes you probably still have to force a full sync, e.g. if _you_ know the disk failed, but drbd did not notice yet, and can not know that you replaced the backing storage. the command to do so is drbdadm invalidate resource-name > - Pretty much the same question, but involving the metadata. If > I replace a drive that includes some or all of the DRBD > metadata, will I still be able to bring up the device when the > system is restarted? Will DRBD realize the metadata is missing > or corrupted and rebuild it? if on startup drbd does not find valid meta data, it will require a full sync on the next start. this is the same situation as when you setup the devices for the first time. lge