stop raid device
# mdadm -S /dev/md2
when a raid 5 or 6 fail you have to rebuild using latest version of data :
get events of drives
# mdadm --examine /dev/sd[abcgf]2 | grep Event
Events : 11369
Events : 11369
Events : 11369
Events : 34
Events : 34
force rebuild raid without the 2 latest drives
# mdadm -Af /dev/md2 /dev/sda2 /dev/sdb2 /dev/sdc2
then add back 2 bad drives
# mdadm -a /dev/md2 /dev/sdg2 /dev/sdf2
watch rebuild
# watch -n1 "cat /proc/mdstat"
...