[DRBD-user] DRBD and write cache

[Siim Vahtre]

Hello,

I haven't really used DRBD, but I've read all the documentation I was able 
to find on the internet. However, there was one question that I didn't get 
good answer to: how (and if) DRBD is able to handle storage with 'write 
cache' turned on?

What happens if one node dies, and the data that disc reported as 
'written' isn't actually written? When the node is restarted, does it 
somehow know that it actually needs full sync (because the data in the 
disc's cache may not be actually written)? Or is it actually able to do 
partial sync thanks to the internal log so it knows what was really 
written?

AFAIK for the latter, you need to be sure that the data, even though maybe 
not actually written to disc, would be written in correct order (so that 
the counters and consistency flags are updated AFTER the other writes to 
the disc).

Is my assumptions correct? If so, is DRBD somehow able to enforce ordered 
writes? (Using write barriers or something?)

E.g short question: may I use write cache on both nodes and not worry 
about data integrity (that is - until two of them die at the same time...)

Thanks!