.An OSD failure no longer causes significant delay
Previously, when an OSD and a Monitor were colocated on the same node and the node failed, Ceph waited some time before sending the note to the Monitor so the Monitor could decide if it wanted to mark the OSD as down. This could lead to a significant delay. With this update, when an OSD is known to be down, the cluster becomes aware immediately after the failure report, and Ceph sends the note to the Monitor right away.
Description of problem:
MON - OSD communication, OSDs are not marked down when co-located with MONs:
In the RHCS 2.1 co-located environment MONs and OSDs, if the co-located node is shuted down, the osds are NOT marked as down. The terminated monitor leaves quorum correctly.
If is is during HW outage and the noout flag is set, the osds stay marked as "up" forever. If we manually perform "ceph osd down X", after 15 minutes f.e., all osds from this node are recognized as down.
If the noout flag is not set, the osds are marked as down and out after 300s.
Nodes that are pure ODS only are marked down correctly every time.
Version-Release number of selected component (if applicable):
ceph version 10.2.3-17.el7cp
Steps to Reproduce:
1. create co-located RHCS 2
2. shutdown one co-located node
3. osds are not recognized as down
osds are not recognized as down
osds are recognized as down
on RHCS 1.3 works fine
Tested in RHCS 1.3.3(0.94.9-8.el7cp) all works fine. Added 'regression' keyword.
Under collocation both MONs and OSDs situation,
jewel backport tracker: http://tracker.ceph.com/issues/18104
jewel backport patch: https://github.com/ceph/ceph/pull/12677/commits/dcc9483cbd605997876983ec65f6fb80d5da8a52
Included in upstream 10.2.6.
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.