I run 8 disks/server as a jbod, 8 filesystems. I've noticed that on a live server, if I hot swap 1 disk and format it, mkfs makes a lot of dirty pages very quickly. That makes the interactive performance of my busy database servers kinda unhappy. Is there any way to convince mkfs to throw in some fsyncs?
The e2fs release notes claim that O_DIRECT support was added in version 1.34, I am running 1.39 (rhel 5), yet I still see the problem. Hm.