@@ -4,43 +4,43 @@ Reconfigure a Replica Set when Members are Down
44
55.. default-domain:: mongodb
66
7- To reconfigure a :term:`replica set` when a majority of the members are
8- down or unreachable, you must manually change the set configuration.
9-
10- This includes situations where you have a network partition and where
11- neither side of the partition has a majority. In such cases the two
12- sides of the partition cannot see each other when determining whether a
13- majority exists (see
14- :ref:`replica-set-elections-and-network-partitions`). Therefore, do not
15- use scripts to reconfigure but do so manually.
16-
17- This section gives several procedures for reconfiguring when a majority
18- is down. Use the procedure appropriate to your version and situation.
19-
20- .. note:: To reconfigure a replica set when a majority of members are
21- running, run the :method:`rs.reconfig()` command on the current
22- :term:`primary`. For examples of using :method:`rs.reconfig()`, see
23- :ref:`replica-set-reconfiguration-usage`.
24-
25- Force a Reconfiguration when the Primary is Down
26- ------------------------------------------------
7+ To reconfigure a :term:`replica set` when only a *minority* of members
8+ are down or unreachable, run the :method:`rs.reconfig()` command on the
9+ current :term:`primary`. For examples of how to reconfigure a replica
10+ set using :method:`rs.reconfig()`, see
11+ :ref:`replica-set-reconfiguration-usage`.
12+
13+ To reconfigure a replica set when a *majority* of the members are down
14+ or unreachable, you must manually change the set configuration as
15+ described in the procedures in this tutorial. Use the procedure
16+ appropriate to your version and situation.
17+
18+ Reconfiguring when a *majority* of members are down can include
19+ situations where you have a network partition and where neither side of
20+ the partition has a majority. In such cases the two sides of the
21+ partition cannot see each other when determining whether a majority
22+ exists (see :ref:`replica-set-elections-and-network-partitions`). In
23+ these situations, never use scripts to reconfigure but instead
24+ reconfigure manually, as described in the procedures here.
25+
26+ .. index:: replica set; reconfiguration
27+ .. _replica-set-force-reconfiguration:
28+
29+ Reconfigure by Forcing the Reconfiguration
30+ ------------------------------------------
2731
2832.. versionchanged:: 2.0
2933
3034This procedure lets you recover while a majority of :term:`replica set`
31- members are down or unreachable. A member might be unreachable, for
32- example, if it is on the wrong side of a network partition.
33-
34- .. TODO Question: must the primary be down for you to use this procedure?
35-
36- You connect to any surviving member and use the the ``force : true``
37- option to force a reconfiguration of the replica set.
35+ members are down or unreachable. You connect to any surviving member and
36+ use :method:`rs.reconfig()`'s ``force`` option to force a
37+ reconfiguration of the replica set.
3838
39- The ``force : true `` option manually reconfigures the set. The option is
39+ The ``force`` option manually reconfigures the set. The option is
4040intended only for serious problems, such as a disaster recovery
41- failover. Do not use ``force : true `` every time you reconfigure. Also,
42- do not put ``force : true `` into any automatic scripts and do not use
43- ``force : true`` when there is still a primary.
41+ failover. Do not use ``force`` any time you reconfigure. Also, do not
42+ include ``force`` into any automatic scripts and do not use ``force``
43+ when there is still a primary.
4444
4545To force reconfiguration:
4646
@@ -69,7 +69,8 @@ To force reconfiguration:
6969 cfg.members = [cfg.members[0] , cfg.members[4] , cfg.members[7]]
7070
7171#. On the same member, reconfigure the set by using the
72- :method:`rs.reconfig()` command with the ``force : true`` option:
72+ :method:`rs.reconfig()` command with the ``force`` option set to
73+ ``true``:
7374
7475 .. code-block:: javascript
7576
@@ -79,29 +80,75 @@ To force reconfiguration:
7980 connected to.
8081
8182 .. note:: When you use ``force : true``, the version number in the
82- replica set configuration increases significantly, by tens or
83- hundreds of thousands. This is normal and designed to prevent set
84- version collisions if network partitioning ends.
83+ replica set configuration increases significantly, by tens or
84+ hundreds of thousands. This is normal and designed to prevent set
85+ version collisions if network partitioning ends.
8586
86- #. If the failure or partition was only temporary, then as members
87- recover they detect that they have been removed from the set and
88- enter a special state where they are up but refuse to answer
89- requests, as they are no longer syncing changes. You can now re-add
90- them to the configuration object:
87+ #. If any of the removed members come back online, there is the chance
88+ they will elect a new primary, resulting in two primaries. To ensure
89+ that the removed members do not elect a new primary, shut down or
90+ decommission the removed members as soon as possible.
9191
92- Be sure that each member has the same _id it had before.
92+ Reconfigure by Replacing the Replica Set
93+ ----------------------------------------
9394
94- .. TODO why does each member have to have the same ID as before?
95- .. The wiki said they'd have to run rs.reconfig(). Why?
95+ The procedures here are intended mainly for MongoDB versions *prior to*
96+ version 2.0. For post-2.0 version, the above procedure,
97+ :ref:`replica-set-force-reconfiguration`, is recommended.
9698
97- Consider the following example:
99+ These procedures are for situations where a *majority* of the
100+ :term:`replica set` members are down or unreachable. If a majority is
101+ *running*, then skip these procedures and instead use the
102+ :method:`rs.reconfig()` command according to the examples in
103+ :ref:`replica-set-reconfiguration-usage`.
98104
99- .. code-block:: javascript
105+ If you run a pre-2.0 version and a majority of your replica set is down,
106+ you have the two options described here. Both involve replacing the
107+ replica set.
108+
109+ Reconfigure by Turning Off Replication
110+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
111+
112+ This option replaces the :term:`replica set` with a :term:`standalone` server.
113+
114+ 1. Stop the surviving :program:`mongod` instances.
115+
116+ #. Perform a backup.
117+
118+ #. Move each surviving member's data directory to an archive folder. For example:
119+
120+ .. code-block:: sh
121+
122+ mv /data/db /data/db-old
123+
124+ .. optional:: You may remove the data instead.
125+
126+ #. Restart one of the :program:`mongod` instances *without* the
127+ ``--replSet`` parameter.
128+
129+ You are back online with a single server that is not a replica set
130+ member. Clients can use this server for both reads and writes.
131+
132+ Reconfigure by "Breaking the Mirror"
133+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
134+
135+ This option selects a surviving :term:`replica set` member to be the new
136+ :term:`primary` and to be the "seed" for a new replica set. All other
137+ members must resync from this new primary.
138+
139+ 1. Stop the surviving :program:`mongod` instances.
140+
141+ #. Perform a backup.
142+
143+ #. Move each surviving member's data directory to an archive folder. For example:
144+
145+ .. code-block:: sh
146+
147+ mv /data/db /data/db-old
100148
101- rs.add("example.com:30003")
149+ .. optional:: You may remove the data instead.
102150
103- Once you add the removed members back into the set, they detect they
104- have been added and synchronize to the current state of the set.
151+ #. Restart all :program:`mongod` instances with the new replica set name.
105152
106- Be aware that if the original primary was one of the removed members,
107- these members may need to rollback .
153+ #. On the new primary, add the other instances as members of the replica
154+ set. For more information, see :doc:`/tutorial/expand-replica-set` .
0 commit comments