From a18621e1cbdd7944ecb482816f6a7d6f922570bf Mon Sep 17 00:00:00 2001 From: Giulio Fidente Date: Tue, 18 Jul 2017 11:14:53 +0200 Subject: [PATCH] Use optimal (instead of default) tunables for Ceph on upgrade With the default setting, after the majority of the monitors have been upgraded the cluster will go in WARN state because of legacy tunables. This changes the tunables we set after each monitor is upgraded from 'default' to 'optimal' [1]. 1. http://docs.ceph.com/docs/master/rados/operations/crush-map/#warning-when-tunables-are-non-optimal While different from the ocata (or pike) versions, this change implements the same fix submitted for ocata: https://review.openstack.org/#/c/484682/ Change-Id: I3635f095ae7d0cdf7f4384dcdc0ae2b39980721e Closes-Bug: #1704959 --- extraconfig/tasks/major_upgrade_ceph_mon.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/extraconfig/tasks/major_upgrade_ceph_mon.sh b/extraconfig/tasks/major_upgrade_ceph_mon.sh index e0d160f1ef..bf1dfa1268 100755 --- a/extraconfig/tasks/major_upgrade_ceph_mon.sh +++ b/extraconfig/tasks/major_upgrade_ceph_mon.sh @@ -73,7 +73,7 @@ elif [[ "$UPDATED_VERSION" =~ ^10\.2.* ]]; then # if tunables become legacy, cluster status will be HEALTH_WARN causing # upgrade to fail on following node - ceph osd crush tunables default + ceph osd crush tunables optimal echo INFO: Ceph was upgraded to Jewel else