@@ -98,27 +98,30 @@ Sharding
98
98
- Use :ref:`targeted operations <sharding-mongos-targeted>`
99
99
for workloads that need to scale with the number of shards.
100
100
101
- - For MongoDB 3.4 and earlier, read from the primary nodes for
101
+ - ** For MongoDB 3.4 and earlier** , read from the primary nodes for
102
102
:ref:`non-targeted or broadcast <sharding-mongos-broadcast>`
103
103
queries as these queries may be sensitive to `stale or orphaned
104
104
data
105
105
<http://blog.mongodb.org/post/74730554385/background-indexing-on-se condaries-and-orphaned>`_.
106
106
107
- - For MongoDB 3.6 and later, secondaries no longer return orphaned
108
- data unless using read concern :readconcern:`"available"`. All
109
- members of the shard replica set maintain chunk metadata, allowing
110
- them to filter out orphans when not using
111
- :readconcern:`"available"`. As such, :ref:`non-targeted or
112
- broadcast <sharding-mongos-broadcast>` queries that are not using
113
- :readconcern:`"available"` can be safely run on any member and
114
- will not return orphaned data.
115
-
116
- The :readconcern:`"available"` read concern can return
117
- :term:`orphaned documents <orphaned document>` from secondary
118
- members since it does not check for updated chunk metadata.
119
- However, if the return of orphaned documents is immaterial to an
120
- application, the :readconcern:`"available"` read concern provides
121
- the lowest latency reads possible among the various read concerns.
107
+ - | **For MongoDB 3.6 and later**, secondaries no longer return orphaned
108
+ data unless using read concern :readconcern:`"available"` (which
109
+ is the default read concern for reads against secondaries when not
110
+ associated with :ref:`causally consistent sessions <sessions>`).
111
+
112
+ | Starting in MongoDB 3.6, all members of the shard replica set
113
+ maintain chunk metadata, allowing them to filter out orphans
114
+ when not using :readconcern:`"available"`. As such,
115
+ :ref:`non-targeted or broadcast <sharding-mongos-broadcast>`
116
+ queries that are not using :readconcern:`"available"` can be
117
+ safely run on any member and will not return orphaned data.
118
+
119
+ | The :readconcern:`"available"` read concern can return
120
+ :term:`orphaned documents <orphaned document>` from secondary
121
+ members since it does not check for updated chunk metadata.
122
+ However, if the return of orphaned documents is immaterial to an
123
+ application, the :readconcern:`"available"` read concern provides
124
+ the lowest latency reads possible among the various read concerns.
122
125
123
126
- :doc:`Pre-split and manually balance chunks
124
127
</tutorial/create-chunks-in-sharded-cluster>` when inserting large
0 commit comments