Hi all, After looking at the source for pdcraid.o I can see that read requests are balanced across two the two disks in a RAID 1 array. For reasons I stated before I think this has the potential to cause massive filesystem damage if the disks go out of sync. I realize that balancing the read requests gives a performance increase but I think that I would rather have data security than performance. Either way it should be configurable. I will investigate a patch to pdcraid.c that allows this behaviour to be configured on or off. For example say the an inode was being updated as a result of a directory move. If one disk 1 got updated but disk 2 did not properly then disk two may need to be fsck'ed. The problem is that on reboot when fsck runs, fsck may read the data from disk 1 assume it is correct and pass the test with no write to either disk. If the directory that was moved was listed (ls) then the resultant output would depend on whatever disk was read from. BAD. If the fsck always read from disk 1 it probably would have fixed the error, written it to disk and both disks would be in sync. This is what I propose. Any comments about this, Michael.
