How does linux hide underlying disks partitions?

I'm trying to evaluate a complex mdadm-lvm-udev-anaconda issue
https://bugzilla.redhat.com/show_bug.cgi?id=1226305 and to undestand how mdadm works.

I have an intel imsm raid-1 volume /dev/md126 assembled with two disks /dev/sda and /dev/sdb.
This volume has two partitions /dev/md126p1 and /dev/md126p2.

Because it a simple mirror, both disks have te same information.

When the normal system boots up, I can see with journalctl that

1. at first /dev/sda1, /dev/sda2, /dev/sdb1, /dev/sdb2 appear
2. then mdadm assembles /dev/md126 from /dev/sda and /dev/sdb
3. /dev/sda1, /dev/sda2, /dev/sdb1, /dev/sdb2 disappear
4. /dev/md126p1 and /dev/md126p2 coming instead

Which program component is responsible for disappearing /dev/sda1, /dev/sda2, /dev/sdb1 and /dev/sdb2?

Why can they be left after starting anaconda installer?

mdadm has a function "remove_partitions" which uses the BLKPG ioctl to tell the kernel to forget about the partitions.

I don't know anything about anaconda - maybe someone else can help.

NeilBrown