Hello,

I have 2 250GB disks with 3 partitions each for /boot, / and swap cofigured as RAID1 md0, md1, md2.
Everything works fine, but when I unplug one physical disk (no matters which) system start to boot, GRUB is working and after while it stucks on "A start job is running for dev-md1.device" (md1 is / ) with moving asterisks and do not continue ( I waited about 3 hours ).

If I return disk and start again, system boot without problem, if I mark all hda or hdb partitions as fail and reboot, it boot OK again. Problem is, when disk is removed (failed) and array is not manualy configured as degraded.

I think, problem is somewhere in Dracut, but cannot get it to work.

Not dracut so much as systemd - but let's not start that old flame-fest again.
This seems a general shortcoming in systemd - could be just a timing issue. Open a bug against CentOS.
If you look at the manpage for dracut, it gives details for debugging - might be worth adding rd.shell to see if you get a shell rather than that continual wait. You need to look at the rdsosreport - the bugzilla folks will probably want it too.