mgr/cephadm: verify spec service_id before applying

There are scenarios where the wrong spec is applied to the wrong OSDs,
likely due to some unknown process interruption or corruption of cache
data, to mitigate this issue I added a check for OSDs service_id to match
the spec service_id.

Fixes: https://tracker.ceph.com/issues/74885
Signed-off-by: Timothy Q Nguyen <timqn22@gmail.com>

diff --git a/src/pybind/mgr/cephadm/services/osd.py b/src/pybind/mgr/cephadm/services/osd.py
index 60a399149f9f9377ef77b34409f50ef656bdb9a8..ecd91f34c85abbafd4ed06411da20d6b37ae6b86 100644 (file)
--- a/src/pybind/mgr/cephadm/services/osd.py
+++ b/src/pybind/mgr/cephadm/services/osd.py
@@ -126,6 +126,9 @@ class OSDService(CephService):
                 if osd['tags']['ceph.cluster_fsid'] != fsid:
                     logger.debug('mismatched fsid, skipping %s' % osd)
                     continue
+                if osd['tags']['ceph.osdspec_affinity'] != spec.service_id:
+                    logger.debug('mismatched service id, skipping %s' % osd)
+                    continue
                 if osd_id in before_osd_uuid_map and osd_id not in replace_osd_ids:
                     # if it exists but is part of the replacement operation, don't skip
                     continue