From: John Wilkins Date: Tue, 12 Mar 2013 21:15:51 +0000 (-0700) Subject: doc: Added troubleshooting entry for single OSD and single replica. X-Git-Tag: v0.60~91 X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=4df44bb1bbff4c5cc07a26cf2c39cefafa3938db;p=ceph.git doc: Added troubleshooting entry for single OSD and single replica. fixes: #3327 Signed-off-by: John Wilkins --- diff --git a/doc/rados/operations/troubleshooting-osd.rst b/doc/rados/operations/troubleshooting-osd.rst index e470968440e6..d167960c397e 100644 --- a/doc/rados/operations/troubleshooting-osd.rst +++ b/doc/rados/operations/troubleshooting-osd.rst @@ -468,6 +468,31 @@ Troubleshooting PG Errors ========================= +Placement Groups Never Get Clean +-------------------------------- + +There are a few cases where Ceph placement groups never get clean: + +#. **One OSD:** If you deviate from the quick start and use only one OSD, you + will likely run into problems. OSDs report other OSDs to the monitor, and + also interact with other OSDs when replicating data. If you have only one + OSD, a second OSD cannot check its heartbeat. Also, if you remove an OSD + and have only one OSD remaining, you may encounter problems. An secondary + or tertiary OSD expects another OSD to tell it which placement groups it + should have. The lack of another OSD prevents this from occurring. So a + placement group can remain stuck “stale” forever. + +#. **Pool Size = 1**: If you have only one copy of an object, no other OSD will + tell the OSD which objects it should have. For each placement group mapped + to the remaining OSD (see ``ceph pg dump``), you can force the OSD to notice + the placement groups it needs by running:: + + ceph pg force_create_pg + +As a general rule, you should run your cluster with more than one OSD and a +pool size greater than 1 object replica. + + Stuck Placement Groups ----------------------