os/bluestore: make bdev label parsing error more meaningful and less noisy #20090
Conversation
This happens during the normal initialization of a new bluestore osd and it is confusing for users. Make it less noisy. Signed-off-by: Sage Weil <sage@redhat.com>
If there is not a valid label, then the label is not found. This is a more reasonable error code than "Invalid argumnet". Signed-off-by: Sage Weil <sage@redhat.com>
| @@ -4200,10 +4200,10 @@ int BlueStore::_read_bdev_label(CephContext* cct, string path, | |||
| decode(expected_crc, p); | |||
| } | |||
| catch (buffer::error& e) { | |||
| derr << __func__ << " unable to decode label at offset " << p.get_off() | |||
| dout(2) << __func__ << " unable to decode label at offset " << p.get_off() | |||
There was a problem hiding this comment.
dout(1), still an error, not able to decode.
|
There isn't a way to conclude that this is the first time the OSD is being created which means it will hit this error? Ideally it wouldn't just send anything out. My concern is that if we make it less noisy, we won't catch it when it is a real problem |
|
failure tracked at http://tracker.ceph.com/issues/22846 @alfredodeza up to you to hold it back or merge it. |
|
Actual ticket is http://tracker.ceph.com/issues/22285 If there isn't a way to make this understand this is the first time starting up, then I guess we can just move forward with this. Anything is better than what we currently have that looks like an error :( |
|
I think this is good as is! |
|
How can we get this backported to Luminous? |
|
Never mind, backported to Luminous with #20326 |
No description provided.