Update dskit with -server.grpc.num-workers option (#6311)

* Update dskit with -server.grpc.num-workers option

This option configures `gprc.NumStreamWorkers()` option, that allows
running long-living workers to handle the requests.

Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>

* Update CHANGELOG, mark as experimental

Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>

* make doc

Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>

* make clean-white-noise

Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>

---------

Signed-off-by: Oleg Zaytsev <mail@olegzaytsev.com>

CHANGELOG.md

@@ -33,6 +33,8 @@
 * [ENHANCEMENT] Ingester / store-gateway: optimized regex matchers. #6168 #6250
 * [ENHANCEMENT] Distributor: Include ingester IDs in circuit breaker related metrics and logs. #6206
 * [ENHANCEMENT] Querier: improve errors and logging when streaming chunks from ingesters and store-gateways. #6194
+* [ENHANCEMENT] All: added an experimental `-server.grpc.num-workers` flag that configures the number of long-living workers used to process gRPC requests. This could decrease the CPU usage by reducing the number of stack allocations. #6311
+* [ENHANCEMENT] All: improved IPv6 support by using the proper host:port formatting. #6311
 * [BUGFIX] Query-frontend: Don't retry read requests rejected by the ingester due to utilization based read path limiting. #6032
 * [BUGFIX] Ring: Ensure network addresses used for component hash rings are formatted correctly when using IPv6. #6068
 * [BUGFIX] Query-scheduler: don't retain connections from queriers that have shut down, leading to gradually increasing enqueue latency over time. #6100 #6145

cmd/mimir/config-descriptor.json

@@ -559,6 +559,17 @@
           "fieldType": "boolean",
           "fieldCategory": "advanced"
         },
+        {
+          "kind": "field",
+          "name": "grpc_server_num_workers",
+          "required": false,
+          "desc": "If non-zero, configures the amount of GRPC server workers used to serve the requests.",
+          "fieldValue": null,
+          "fieldDefaultValue": 0,
+          "fieldFlag": "server.grpc.num-workers",
+          "fieldType": "int",
+          "fieldCategory": "experimental"
+        },
         {
           "kind": "field",
           "name": "log_format",

cmd/mimir/help-all.txt.tmpl

@@ -2475,6 +2475,8 @@ Usage of ./cmd/mimir/mimir:
     	Duration after which a keepalive probe is sent in case of no activity over the connection., Default: 2h (default 2h0m0s)
   -server.grpc.keepalive.timeout duration
     	After having pinged for keepalive check, the duration after which an idle connection should be closed, Default: 20s (default 20s)
+  -server.grpc.num-workers int
+    	[experimental] If non-zero, configures the amount of GRPC server workers used to serve the requests.
   -server.http-conn-limit int
     	Maximum number of simultaneous http connections, <=0 to disable
   -server.http-idle-timeout duration

docs/sources/mimir/configure/about-versioning.md

@@ -141,6 +141,7 @@ The following features are currently experimental:
     - `log.rate-limit-logs-per-second-burst`
 - Timeseries Unmarshal caching optimization in distributor (`-timeseries-unmarshal-caching-optimization-enabled`)
 - Reusing buffers for marshalling write requests in distributors (`-distributor.write-requests-buffer-pooling-enabled`)
+- Using a worker pool for handling GRPC requests (`-server.grpc.num-workers`)
 
 ## Deprecated features
 

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

@@ -653,6 +653,11 @@ grpc_tls_config:
 # CLI flag: -server.grpc.keepalive.ping-without-stream-allowed
 [grpc_server_ping_without_stream_allowed: <boolean> | default = true]
 
+# (experimental) If non-zero, configures the amount of GRPC server workers used
+# to serve the requests.
+# CLI flag: -server.grpc.num-workers
+[grpc_server_num_workers: <int> | default = 0]
+
 # Output log messages in the given format. Valid formats: [logfmt, json]
 # CLI flag: -log.format
 [log_format: <string> | default = "logfmt"]

go.mod

@@ -20,7 +20,7 @@ require (
 	github.com/golang/snappy v0.0.4
 	github.com/google/gopacket v1.1.19
 	github.com/gorilla/mux v1.8.0
-	github.com/grafana/dskit v0.0.0-20231009143556-6a6d7e8a6f05
+	github.com/grafana/dskit v0.0.0-20231010092604-9b82fbd87ce1
 	github.com/grafana/e2e v0.1.1
 	github.com/hashicorp/golang-lru v1.0.2 // indirect
 	github.com/json-iterator/go v1.1.12

go.sum

@@ -551,8 +551,8 @@ github.com/gosimple/slug v1.1.1 h1:fRu/digW+NMwBIP+RmviTK97Ho/bEj/C9swrCspN3D4=
 github.com/gosimple/slug v1.1.1/go.mod h1:ER78kgg1Mv0NQGlXiDe57DpCyfbNywXXZ9mIorhxAf0=
 github.com/grafana-tools/sdk v0.0.0-20220919052116-6562121319fc h1:PXZQA2WCxe85Tnn+WEvr8fDpfwibmEPgfgFEaC87G24=
 github.com/grafana-tools/sdk v0.0.0-20220919052116-6562121319fc/go.mod h1:AHHlOEv1+GGQ3ktHMlhuTUwo3zljV3QJbC0+8o2kn+4=
-github.com/grafana/dskit v0.0.0-20231009143556-6a6d7e8a6f05 h1:4htc/ow2X9ZGIott1BgvPdfswoekB6sfq1DmmGaCG70=
-github.com/grafana/dskit v0.0.0-20231009143556-6a6d7e8a6f05/go.mod h1:byPCvaG/pqi33Kq+Wvkp7WhLfmrlyy0RAoYG4yRh01I=
+github.com/grafana/dskit v0.0.0-20231010092604-9b82fbd87ce1 h1:u46tR3g8obq3eXKj6ifaPV8iC/98o1AymwHCOtHNkVc=
+github.com/grafana/dskit v0.0.0-20231010092604-9b82fbd87ce1/go.mod h1:byPCvaG/pqi33Kq+Wvkp7WhLfmrlyy0RAoYG4yRh01I=
 github.com/grafana/e2e v0.1.1 h1:/b6xcv5BtoBnx8cZnCiey9DbjEc8z7gXHO5edoeRYxc=
 github.com/grafana/e2e v0.1.1/go.mod h1:RpNLgae5VT+BUHvPE+/zSypmOXKwEu4t+tnEMS1ATaE=
 github.com/grafana/gomemcache v0.0.0-20230914135007-70d78eaabfe1 h1:MLYY2R60/74hfYl5vRRmC2VDo0Yuql1QQ1ig8hnvgSI=

pkg/util/fieldcategory/overrides.go

@@ -53,6 +53,7 @@ var overrides = map[string]Category{
 	"server.grpc.keepalive.ping-without-stream-allowed": Advanced,
 	"server.grpc.keepalive.time":                        Advanced,
 	"server.grpc.keepalive.timeout":                     Advanced,
+	"server.grpc.num-workers":                           Experimental,
 	"server.http-conn-limit":                            Advanced,
 	"server.http-idle-timeout":                          Advanced,
 	"server.http-listen-network":                        Advanced,