Ceph » RADOS

Reproducer(s):
0. be in build dir
1. run vstart.sh
2. edit stop.sh to not `rm -rf "${asok_dir}"`
3. do ls of /tmp/ceph-asok.XXXX and out dirs to see mgr, mon, osd asok files and radosgw.8000.asok
4. run ../src/stop.sh
5. look in out dir and /tmp/ceph-asok.XXX to see the .asok files still there

This is could be due to the order in which the daemons are killed in stop.sh.

This issue has been seen in clusters configured with ceph-ansible, see: https://bugzilla.redhat.com/show_bug.cgi?id=1794550

A way to fix this is to have the admin sockets be in the abstract socket namespace on Linux to ensure that the .asok files are always deleted when the socket is closed: https://troydhanson.github.io/network/Unix_domain_sockets.html

A branch has been created to implement this: https://github.com/alimaredia/ceph/commit/42a8ec3fbe6cb253960f65710ce0108a864eb1e1