Enable distributor write requests buffer pooling by default (#7165)

CHANGELOG.md

@@ -29,6 +29,7 @@
   * `-ingester.ring.spread-minimizing-zones`
   * `-ingester.ring.spread-minimizing-join-ring-in-order`
 * [CHANGE] Query-frontend: the default value of the CLI flag `-query-frontend.max-cache-freshness` (and its respective YAML configuration parameter) has been changed from `1m` to `10m`. #7161
+* [CHANGE] Distributor: default the optimization `-distributor.write-requests-buffer-pooling-enabled` to `true`. #7165
 * [FEATURE] Introduce `-tenant-federation.max-tenants` option to limit the max number of tenants allowed for requests when federation is enabled. #6959
 * [FEATURE] Cardinality API: added a new `count_method` parameter which enables counting active label values. #7085
 * [FEATURE] Querier / query-frontend: added `-querier.promql-experimental-functions-enabled` CLI flag (and respective YAML config option) to enable experimental PromQL functions. The experimental functions introduced are: `mad_over_time()`, `sort_by_label()` and `sort_by_label_desc()`. #7057

cmd/mimir/config-descriptor.json

@@ -1632,7 +1632,7 @@
           "required": false,
           "desc": "Enable pooling of buffers used for marshaling write requests.",
           "fieldValue": null,
-          "fieldDefaultValue": false,
+          "fieldDefaultValue": true,
           "fieldFlag": "distributor.write-requests-buffer-pooling-enabled",
           "fieldType": "boolean",
           "fieldCategory": "experimental"

cmd/mimir/help-all.txt.tmpl

@@ -1240,7 +1240,7 @@ Usage of ./cmd/mimir/mimir:
   -distributor.service-overload-status-code-on-rate-limit-enabled
     	[experimental] If enabled, rate limit errors will be reported to the client with HTTP status code 529 (Service is overloaded). If disabled, status code 429 (Too Many Requests) is used. Enabling -distributor.retry-after-header.enabled before utilizing this option is strongly recommended as it helps prevent premature request retries by the client.
   -distributor.write-requests-buffer-pooling-enabled
-    	[experimental] Enable pooling of buffers used for marshaling write requests.
+    	[experimental] Enable pooling of buffers used for marshaling write requests. (default true)
   -enable-go-runtime-metrics
     	Set to true to enable all Go runtime metrics, such as go_sched_* and go_memstats_*.
   -flusher.exit-after-flush

docs/sources/mimir/configure/configuration-parameters/index.md

@@ -921,7 +921,7 @@ instance_limits:
 
 # (experimental) Enable pooling of buffers used for marshaling write requests.
 # CLI flag: -distributor.write-requests-buffer-pooling-enabled
-[write_requests_buffer_pooling_enabled: <boolean> | default = false]
+[write_requests_buffer_pooling_enabled: <boolean> | default = true]
 
 # (experimental) Use experimental method of limiting push requests.
 # CLI flag: -distributor.limit-inflight-requests-using-grpc-method-limiter

pkg/distributor/distributor.go

@@ -222,7 +222,7 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet, logger log.Logger) {
 
 	f.IntVar(&cfg.MaxRecvMsgSize, "distributor.max-recv-msg-size", 100<<20, "Max message size in bytes that the distributors will accept for incoming push requests to the remote write API. If exceeded, the request will be rejected.")
 	f.DurationVar(&cfg.RemoteTimeout, "distributor.remote-timeout", 2*time.Second, "Timeout for downstream ingesters.")
-	f.BoolVar(&cfg.WriteRequestsBufferPoolingEnabled, "distributor.write-requests-buffer-pooling-enabled", false, "Enable pooling of buffers used for marshaling write requests.")
+	f.BoolVar(&cfg.WriteRequestsBufferPoolingEnabled, "distributor.write-requests-buffer-pooling-enabled", true, "Enable pooling of buffers used for marshaling write requests.")
 	f.BoolVar(&cfg.LimitInflightRequestsUsingGrpcMethodLimiter, "distributor.limit-inflight-requests-using-grpc-method-limiter", false, "Use experimental method of limiting push requests.")
 	f.IntVar(&cfg.ReusableIngesterPushWorkers, "distributor.reusable-ingester-push-workers", 2000, "Number of pre-allocated workers used to forward push requests to the ingesters. If 0, no workers will be used and a new goroutine will be spawned for each ingester push request. If not enough workers available, new goroutine will be spawned. (Note: this is a performance optimization, not a limiting feature.)")