Patch up the CRUSH map compatibility guards #2072
Conversation
We need to see if there's a feature which is not in the quorum_features, not if there are no features in common! Signed-off-by: Greg Farnum <greg@inktank.com>
…ster features When we change the tunables, or set a new CRUSH map, we need to make sure it's supported by all the monitors and OSDs currently participating in the cluster. Fixes: #8738 Signed-off-by: Greg Farnum <greg@inktank.com>
|
I ran local tests on this with vstart, verifying that it protects changes while there are old daemons, and allows changes when the daemons are updated. Haven't done anything else with it. |
|
It would be nice to have a minimal test that shows the problem as it was and prove that it is gone after the patch. |
|
Looks good to me! |
|
@dachary, we don't have any of the mocks we'd need to do that reasonably, and as it stands testing this requires old and new daemons running in the same cluster to demonstrate. |
|
|
||
| stringstream features_ss; | ||
|
|
||
| int r = check_cluster_features(features, features_ss); |
There was a problem hiding this comment.
Then we won't apply the change, and the user will get output saying a pending OSD doesn't support it.
|
@gregsfortytwo I'm asking because I've recently discovered the appeal of the upgrade suite ;-) It's non trivial work however, that's true. |
|
the patches look good to me. |
1 similar comment
|
the patches look good to me. |
Patch up the CRUSH map compatibility guards Reviewed-by: Loic Dachary <loic@dachary.org> Reviewed-by: Joao Eduardo Luis <joao.luis@inktank.com> Reviewed-by: Sage Weil <sage@redhat.com>
The "ceph osd crush tunables optimal" and "ceph osd setcrushmap" commands left an incompatibility window. Close it by testing the proposed CRUSH change feature requirements against the existing cluster's feature bits.