Red Hat Bugzilla – Bug 1249042
[GSS] Add a use-case for the CRUSH OSD tunable "osd crush update on start" in RHCS1.3 Storage Strategies documentation
Last modified: 2015-12-18 05:21:17 EST
a) Description of problem:
Add a use-case for the CRUSH OSD tunable "osd crush update on start" under the section "4.1.1. ceph-crush-location hook" in RHCS1.3 Storage Strategies documentation.
If a user sets a custom hostname for an OSD node in CRUSH map, and the OSD process or the node is restarted, the CRUSH maps get reverted back to the initial state. This makes the custom settings vanish.
In order to overcome this, set "osd crush update on start = false" in /etc/ceph/ceph.conf and push the configuration to the OSD/MON nodes with :
# ceph-deploy --overwrite-conf config push <node1> <node2> ... <nodeN>
b) Version-Release number of selected component (if applicable):
RHCS1.2.3 Storage Strategies documentation.
NOTE: Editing a crush map and setting custom hostnames for rulesets and buckets are something customers do usually. The Crush map will revert back to the default as per the OSD hierarchy, if this tunable is not set. We need to get this added as a use-case.
This seems like it might be useful to mention in a few more places. For example, the installation guide's section on creating a CRUSH hierarchy, and maybe the details for creating a CRUSH hierarchy. Alternatively, perhaps we should instruct the user to add `crush_location` when adding an OSD to a CRUSH hierarchy so this setting doesn't need to be set.
Here's the doc link:
verified on 1.3.0 and it looks good