From 6db3f9f3422400e408136501b817ded3cddc5505 Mon Sep 17 00:00:00 2001 From: Jeffrey Zhang Date: Mon, 5 Nov 2018 15:02:40 +0800 Subject: [PATCH] Disable ceph osd crush update on start in default The buggy come from ceph changes[0], which is included since ceph osd v11.0.0. The `osd crush update on start` logical is moved from `ceph-osd-prestart.sh` to ceph-osd startup process. So ceph-osd will create buckets by node hostname automatically. Whereas, kolla is creating buckets by node ip For the less confused and ceph upgrade impact, disable `osd crush update on start` is a better choice [0] https://github.com/ceph/ceph/commit/a28b71e3c9446541e14795324f2ec1f9d69c9187 Change-Id: Ibbeac9505c9957319126267dbe6bd7a2cac11f0c Closes-Bug: #1801662 --- ansible/roles/ceph/templates/ceph.conf.j2 | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/ansible/roles/ceph/templates/ceph.conf.j2 b/ansible/roles/ceph/templates/ceph.conf.j2 index 0ea2522f6f..0b788f8390 100644 --- a/ansible/roles/ceph/templates/ceph.conf.j2 +++ b/ansible/roles/ceph/templates/ceph.conf.j2 @@ -22,6 +22,11 @@ auth client required = cephx # chown operation in upgrade procedure. https://bugs.launchpad.net/kolla/+bug/1620702 setuser match path = /var/lib/ceph/$type/$cluster-$id +# NOTE(Jeffrey4l): kolla is using node ip as host bucket name, whereas `osd +# crush update on start` feature depends on hostname. Disable this feature for +# less confusion and upgrade impact. +osd crush update on start = false + [mon] # NOTE(SamYaple): The monitor files have been known to grow very large. The # only fix for that is to compact the files.