Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
> you are very sure that you did not create a DRBD with internel meta data > over an existing file system without shrinking the fs first, thus > destroying the last 128 MB of the fs, and corrupting it badly? Yes, I am very sure, I got external metadata. The filesystem has been existing for a few weeks now. There's one thing that *might* have gone wrong due to my stupidity- I bonnied the underlying device on each machine (obviously I stopped drbd before) to measure the performance of the underlying device. I am pretty sure I invalidated the one I bonnied on while the the other node primary thus syncing it in the right direction. Yepp, the logs tell me that I synced them fully and in the right direction. I have to look deeper into it. > you are very sure you don't have bad ram or a bad southbridge or > some other kaputt hardware ? Hmm, pretty sure but who knows. I was going to run memtest over it anyway. > > After this the machine locked up. Fscking it didn't help. Now I thougt > > I'd try Ext3 and see how it performs. Mkfs stuck. > > stuck? or just very slow? (mke2fs is _very_ slow on large devices). Good question. It might just have been very slow- I just see that the underlying raid is rebuilding itself as a result of the crash which isn't exactly helping either. > > we disallow invalidation of a Primary for obvious reasons. > you only can invalidate a Secondary. That makes sense. Sorry for bothering. The mixture of me staring at the message "Only in 'Connected' cstate possible." while also looking at a connected resource mixed with my stupidity was just too much. > you can invalidate the Secondary from the Primary by saying > "invalidate_remote". > > but why would you want to invalidate? > what do you expect this to solve? That's also a good question. I wanted to start from scratch and tell both nodes to forget about their metadata. I might as well have zeroed them but then I got stuck with the 'need to be connected' confusion. I just wanted to know everything is really at zero. Now I know this isn't the best way- I just made both secondary, invalidated them and ended up with an oops on one of them. Shouldn't it just be 'I don't have access to the good data' or so ?