From 7178f1caa8cf7c6f65f8d72a28718da417426af8 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Mon, 5 Dec 2011 14:07:44 -0800
Subject: [PATCH] doc: document monitor cluster expansion/contraction

Pretty sure my rst syntax is wrong.

Signed-off-by: Sage Weil <sage@newdream.net>
---
 doc/ops/manage/grow/mon.rst | 75 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 74 insertions(+), 1 deletion(-)

diff --git a/doc/ops/manage/grow/mon.rst b/doc/ops/manage/grow/mon.rst
index eee28004ec0de..15b5df09a3cd4 100644
--- a/doc/ops/manage/grow/mon.rst
+++ b/doc/ops/manage/grow/mon.rst
@@ -2,4 +2,77 @@
  Resizing the monitor cluster
 ==============================
 
-.. todo:: Choosing and adjusting cluster size
+Adding a monitor
+----------------
+
+1) Initialize the new monitor's data directory with the ``ceph-mon
+--mkfs`` command.  You need to provide the new monitor with three
+pieces of information:
+
+   - the cluster fsid
+   - one or more existing monitors to join
+   - the monitor authentication key ``mon.``
+
+   The simplest way to do this is probably::
+
+     $ ceph mon getmap -o /tmp/monmap
+     $ ceph auth export mon. -o /tmp/monkey
+     $ ceph-mon -i newname --mkfs --monmap /tmp/foo --keyring /tmp/monkey
+
+2) Start the new monitor and it will automatically join the cluster::
+
+     $ ceph-mon -i newname
+
+3) If you would like other nodes to be able to use this monitor during
+their initial startup, you'll need to adjust ``ceph.conf`` to add a
+section and ``mon addr`` for the new monitor, or add it to the
+existing ``mon host`` list.
+
+
+Removing a monitor from a healthy cluster
+-----------------------------------------
+
+If the cluster is healthy, you can do::
+
+  $ ceph mon remove $id
+
+For example, if your cluster includes ``mon.a``, ``mon.b``, and ``mon.c``, then you can remove ``mon.c`` with::
+
+  $ ceph mon remove c
+
+Removing a monitor from an unhealth/down cluster
+------------------------------------------------
+
+The mon cluster may not be up because you have lost too many nodes to
+form a quorum.
+
+1) On a surviving monitor node, find the most recent monmap::
+
+     $ ls $mon_data/monmap
+     1  2  accepted_pn  last_committed  latest
+
+   in this case it is 2.
+
+2) Copy to a temporary location and modify the monmap to remove the
+node(s) you don't want.  Let's say the map has ``mon.a``, ``mon.b``,
+and ``mon.c``, but only ``mon.a`` is surviving::
+
+     $ cp $mon_data/monmap/2 /tmp/foo
+     $ monmaptool /tmp/foo --rm b
+     $ monmaptool /tmp/foo --rm c
+
+3) Make sure ceph-mon isn't running::
+
+     $ service ceph stop mon
+
+4) Inject the modified map on any surviving nodes.  For example, for
+``mon.a``::
+
+     $ ceph-mon -i a --inject-monmap /tmp/foo   # for each surviving monitor
+
+5) Start the surviving monitor(s)::
+
+     $ service ceph start mon           # on each node with a surviving monitor
+
+6) Remove the old monitors from ``ceph.conf`` so that nobody tries to
+connect to the old instances.
\ No newline at end of file
-- 
2.39.5