Improve docs for search preferences (#32098)

Today it is unclear what guarantees are offered by the search preference
feature, and we claim a guarantee that is stronger than what we really offer:

> A custom value will be used to guarantee that the same shards will be used
> for the same custom value.

This commit clarifies this documentation and explains more clearly why
`_primary`, `_replica`, etc. are deprecated in `6.x` and removed in `master`.

Relates #31929 #26335 #26791.

Author: DaveCTurner

docs/reference/search/request/preference.asciidoc

@@ -1,56 +1,76 @@
 [[search-request-preference]]
 === Preference
 
-Controls a `preference` of which shard copies on which to execute the
-search. By default, the operation is randomized among the available shard 
-copies, unless allocation awareness is used.
+Controls a `preference` of which shard copies on which to execute the search.
+By default, Elasticsearch selects from the available shard copies in an
+unspecified order, taking the <<allocation-awareness,allocation awareness>> and
+<<search-adaptive-replica,adaptive replica selection>> configuration into
+account. However, it may sometimes be desirable to try and route certain
+searches to certain sets of shard copies, for instance to make better use of
+per-copy caches.
 
 The `preference` is a query string parameter which can be set to:
 
 [horizontal]
-`_primary`:: 
-	The operation will go and be executed only on the primary
-	shards. deprecated[6.1.0, will be removed in 7.0, use `_only_nodes` or `_prefer_nodes`]
+`_primary`::
+	The operation will be executed only on primary shards.  deprecated[6.1.0,
+	will be removed in 7.0. See the warning below for more information.]
 
-`_primary_first`:: 
-	The operation will go and be executed on the primary
-	shard, and if not available (failover), will execute on other shards.
-	deprecated[6.1.0, will be removed in 7.0, use `_only_nodes` or `_prefer_nodes`]
+`_primary_first`::
+	The operation will be executed on primary shards if possible, but will fall
+	back to other shards if not. deprecated[6.1.0, will be removed in 7.0. See
+	the warning below for more information.]
 
 `_replica`::
-  The operation will go and be executed only on a replica shard.
-  deprecated[6.1.0, will be removed in 7.0, use `_only_nodes` or `_prefer_nodes`]
+	The operation will be executed only on replica shards. If there are multiple
+	replicas then the order of preference between them is unspecified.
+	deprecated[6.1.0, will be removed in 7.0. See the warning below for more
+	information.]
 
 `_replica_first`::
-  The operation will go and be executed only on a replica shard, and if
-  not available (failover), will execute on other shards.
-  deprecated[6.1.0, will be removed in 7.0, use `_only_nodes` or `_prefer_nodes`]
+	The operation will be executed on replica shards if possible, but will fall
+	back to other shards if not. If there are multiple replicas then the order of
+	preference between them is unspecified.  deprecated[6.1.0, will be removed in
+	7.0. See the warning below for more information.]
 
-`_local`:: 
-	The operation will prefer to be executed on a local
-	allocated shard if possible.
+`_only_local`::
+	The operation will be executed only on shards allocated to the local
+	node.
+
+`_local`::
+	The operation will be executed on shards allocated to the local node if
+	possible, and will fall back to other shards if not.
 
 `_prefer_nodes:abc,xyz`::
-	Prefers execution on the nodes with the provided
-	node ids (`abc` or `xyz` in this case) if applicable.
+	The operation will be executed on nodes with one of the provided node
+	ids (`abc` or `xyz` in this case) if possible. If suitable shard copies
+	exist on more than one of the selected nodes then the order of
+	preference between these copies is unspecified.
 
-`_shards:2,3`:: 
-	Restricts the operation to the specified shards. (`2`
-	and `3` in this case). This preference can be combined with other
-	preferences but it has to appear first: `_shards:2,3|_local`
+`_shards:2,3`::
+	Restricts the operation to the specified shards. (`2` and `3` in this
+	case).  This preference can be combined with other preferences but it
+	has to appear first: `_shards:2,3|_local`
 
-`_only_nodes`::
-    Restricts the operation to nodes specified in <<cluster,node specification>>
+`_only_nodes:abc*,x*yz,...`::
+	Restricts the operation to nodes specified according to the
+	<<cluster,node specification>>. If suitable shard copies exist on more
+	than one of the selected nodes then the order of preference between
+	these copies is unspecified.
 
-Custom (string) value:: 
-	A custom value will be used to guarantee that
-	the same shards will be used for the same custom value. This can help
-	with "jumping values" when hitting different shards in different refresh
-	states. A sample value can be something like the web session id, or the
-	user name.
+Custom (string) value::
+	Any value that does not start with `_`. If two searches both give the same
+	custom string value for their preference and the underlying cluster state
+	does not change then the same ordering of shards will be used for the
+	searches. This does not guarantee that the exact same shards will be used
+	each time: the cluster state, and therefore the selected shards, may change
+	for a number of reasons including shard relocations and shard failures, and
+	nodes may sometimes reject searches causing fallbacks to alternative nodes.
+	However, in practice the ordering of shards tends to remain stable for long
+	periods of time. A good candidate for a custom preference value is something
+	like the web session id or the user name.
 
-For instance, use the user's session ID to ensure consistent ordering of results
-for the user:
+For instance, use the user's session ID `xyzabc123` as follows:
 
 [source,js]
 ------------------------------------------------
@@ -65,3 +85,22 @@ GET /_search?preference=xyzabc123
 ------------------------------------------------
 // CONSOLE
 
+NOTE: The `_only_local` preference guarantees only to use shard copies on the
+local node, which is sometimes useful for troubleshooting. All other options do
+not _fully_ guarantee that any particular shard copies are used in a search,
+and on a changing index this may mean that repeated searches may yield
+different results if they are executed on different shard copies which are in
+different refresh states.
+
+WARNING: The `_primary`, `_primary_first`, `_replica` and `_replica_first` are
+deprecated as their use is not recommended. They do not help to avoid
+inconsistent results that arise from the use of shards that have different
+refresh states, and Elasticsearch uses synchronous replication so the primary
+does not in general hold fresher data than its replicas. The `_primary_first`
+and `_replica_first` preferences silently fall back to non-preferred copies if
+it is not possible to search the preferred copies. The `_primary` and
+`_replica` preferences will silently change their preferred shards if a replica
+is promoted to primary, which can happen at any time. The `_primary` preference
+can also put undesirable extra load on the primary shards. The cache-related
+benefits of these options can also be obtained using `_only_nodes`,
+`_prefer_nodes`, or a custom string value instead.