Prevent 'thundering herd' issue with the scale preset (#42410)

Add backoff settings to the `scale` performance preset, as described in elastic/elastic-agent#4469, to reduce load on large deployments.

(cherry picked from commit 5fcc26b)

Co-authored-by: Fae Charlton <[email protected]>

libbeat/outputs/elasticsearch/config_presets.go

@@ -63,6 +63,8 @@ var presetConfigs = map[string]*config.C{
 		"queue.mem.flush.timeout":    20 * time.Second,
 		"compression_level":          1,
 		"idle_connection_timeout":    1 * time.Second,
+		"backoff.init":               5 * time.Second,
+		"backoff.max":                300 * time.Second,
 	}),
 	presetLatency: config.MustNewConfigFrom(map[string]interface{}{
 		"bulk_max_size":              50,

libbeat/outputs/elasticsearch/docs/elasticsearch.asciidoc

@@ -680,6 +680,8 @@ Setting `bulk_max_size` to values less than or equal to 0 disables the
 splitting of batches. When splitting is disabled, the queue decides on the
 number of events to be contained in a batch.
 
+
+[[backoff-init-option]]
 ===== `backoff.init`
 
 The number of seconds to wait before trying to reconnect to Elasticsearch after
@@ -689,6 +691,7 @@ to `backoff.max`. After a successful connection, the backoff timer is reset. The
 default is `1s`.
 
 
+[[backoff-max-option]]
 ===== `backoff.max`
 
 The maximum number of seconds to wait before attempting to connect to
@@ -774,7 +777,7 @@ output.elasticsearch:
   preset: balanced
 ------------------------------------------------------------------------------
 
-Performance presets apply a set of configuration overrides based on a desired performance goal. If set, a performance preset will override other configuration flags to match the recommended settings for that preset. Valid options are:
+Performance presets apply a set of configuration overrides based on a desired performance goal. If set, a performance preset will override other configuration flags to match the recommended settings for that preset. If a preset doesn't set a value for a particular field, the user-specified value will be used if present, otherwise the default. Valid options are:
 * `balanced`: good starting point for general efficiency
 * `throughput`: good for high data volumes, may increase cpu and memory requirements
 * `scale`: reduces ambient resource use in large low-throughput deployments
@@ -830,6 +833,18 @@ Presets represent current recommendations based on the intended goal; their effe
 |`15s`
 |`1s`
 |`60s`
+
+|<<backoff-init-option,`backoff.init`>>
+|none
+|none
+|`5s`
+|none
+
+|<<backoff-max-option,`backoff.max`>>
+|none
+|none
+|`300s`
+|none
 |===
 
 [[es-apis]]