Feature #4650
open
osd: separate OSD names from their IDs
Added by Greg Farnum about 11 years ago.
Updated almost 6 years ago.
Description
The MDS and monitors allow you to give each daemon a human-readable name which is distinct from their internal system IDs. Users can't do this for the OSD, and it tends to first confuse them, and then drive them nuts.
This will become more important as deployments get larger and more auto-generated — admins who bring up an entire cluster at once will end up with IDs that have no correlation to the device they reside on (not even a range of IDs being grouped together, due to initial startup racing), and we don't provide them the information to handle that well at all.
From a system administration point of view there is no need to know about the OSD id. Naming the OSDs with human readable names will probably encourage users to control the names of the OSDs instead of letting the Ceph cluster manage them in a fully automated way. Or am I missing something ?
We expose OSD IDs in lots of places — like error reporting. But users can't specify those IDs (although they could once) and it causes lots of confusion.
In particular, there's no easy way to go "oh, OSDs 7, 43, 89, and 90 are all down. Those live together on host N; it's probably broken!" Whereas the normal naming scheme for monitors and MDSes, if applied to the OSDs, would result in "oh, OSDs N.1, N.2, N.3, and N.4 are all down; N is probably broken!"
Labels which we expose to users should be controllable by the users.
- Project changed from Ceph to RADOS
- Subject changed from separate OSD names from their IDs to osd: separate OSD names from their IDs
- Component(RADOS) OSD added
Also available in: Atom
PDF