Merge #148657 #148658

148657: server: register `DistSQL`, `Tracing` service with DRPC server r=cthumuluru-crdb a=cthumuluru-crdb

Enable `DistSQL` and `Tracing` services on the DRPC server in addition to gRPC. This is controlled by `rpc.experimental_drpc.enabled` (off by default).

This change is part of a series and is similar to: #146926

Note: This only registers the service; the client is not updated to use the DRPC client, so this service will not have any functional effect.

Epic: CRDB-48925
Release note: None

148658: roachtest: more CPU profiling in rebalancing tests r=tbg a=tbg

Closes #147487.

Co-authored-by: Chandra Thumuluru <[email protected]>
Co-authored-by: Tobias Grieger <[email protected]>

Author: 3 people

pkg/cmd/roachtest/tests/rebalance_load.go

@@ -97,6 +97,12 @@ func registerRebalanceLoad(r registry.Registry) {
 		settings.ClusterSettings["kv.allocator.load_based_rebalancing"] = rebalanceMode
 		settings.ClusterSettings["kv.range_split.by_load_enabled"] = "false"
 
+		// Take a 10s profile every minute.
+		settings.ClusterSettings["server.cpu_profile.duration"] = "10s"
+		settings.ClusterSettings["server.cpu_profile.interval"] = "1m"
+		settings.ClusterSettings["server.cpu_profile.cpu_usage_combined_threshold"] = "1" // basically always true
+		settings.ClusterSettings["server.cpu_profile.total_dump_size_limit"] = "256 MiB"
+
 		if mixedVersion {
 			mvt := mixedversion.NewTest(ctx, t, t.L(), c, roachNodes, mixedversion.NeverUseFixtures,
 				mixedversion.ClusterSettingOption(
@@ -119,8 +125,6 @@ func registerRebalanceLoad(r registry.Registry) {
 				})
 			mvt.Run()
 		} else {
-			// Note that CPU profiling is already enabled by default, should there be
-			// a failure it will be available in the artifacts.
 			c.Start(ctx, t.L(), startOpts, settings, roachNodes)
 			require.NoError(t, rebalanceByLoad(
 				ctx, t, t.L(), c, maxDuration,

pkg/server/BUILD.bazel

@@ -375,6 +375,7 @@ go_library(
         "@com_github_prometheus_client_model//go",
         "@com_github_prometheus_common//expfmt",
         "@in_gopkg_yaml_v2//:yaml_v2",
+        "@io_storj_drpc//:drpc",
         "@io_storj_drpc//drpcerr",
         "@io_storj_drpc//drpcmigrate",
         "@org_golang_google_grpc//:grpc",

pkg/server/server.go

@@ -1152,6 +1152,7 @@ func NewServer(cfg Config, stopper *stop.Stopper) (serverctl.ServerStartupInterf
 			nodeLiveness:             optionalnodeliveness.MakeContainer(nodeLiveness),
 			gossip:                   gossip.MakeOptionalGossip(g),
 			grpcServer:               grpcServer.Server,
+			drpcMux:                  drpcServer.DRPCServer,
 			nodeIDContainer:          idContainer,
 			externalStorage:          externalStorage,
 			externalStorageFromURI:   externalStorageFromURI,

pkg/server/server_sql.go

@@ -140,6 +140,7 @@ import (
 	"github.com/marusama/semaphore"
 	"github.com/nightlyone/lockfile"
 	"google.golang.org/grpc"
+	"storj.io/drpc"
 )
 
 // SQLServer encapsulates the part of a CRDB server that is dedicated to SQL
@@ -232,6 +233,7 @@ type sqlServerOptionalKVArgs struct {
 	gossip gossip.OptionalGossip
 	// To register blob and DistSQL servers.
 	grpcServer *grpc.Server
+	drpcMux    drpc.Mux
 	// For the temporaryObjectCleaner.
 	isMeta1Leaseholder func(context.Context, hlc.ClockTimestamp) (bool, error)
 	// DistSQL, lease management, and others want to know the node they're on.
@@ -555,6 +557,9 @@ func newSQLServer(ctx context.Context, cfg sqlServerArgs) (*SQLServer, error) {
 	// Create trace service for inter-node sharing of inflight trace spans.
 	tracingService := service.New(cfg.Tracer)
 	tracingservicepb.RegisterTracingServer(cfg.grpcServer, tracingService)
+	if err := tracingservicepb.DRPCRegisterTracing(cfg.drpcMux, tracingService); err != nil {
+		return nil, err
+	}
 
 	// If the node id is already populated, we only need to create a placeholder
 	// instance provider without initializing the instance, since this is not a
@@ -873,6 +878,9 @@ func newSQLServer(ctx context.Context, cfg sqlServerArgs) (*SQLServer, error) {
 
 	distSQLServer := distsql.NewServer(ctx, distSQLCfg, cfg.remoteFlowRunner)
 	execinfrapb.RegisterDistSQLServer(cfg.grpcServer, distSQLServer)
+	if err := execinfrapb.DRPCRegisterDistSQL(cfg.drpcMux, distSQLServer.AsDRPCServer()); err != nil {
+		return nil, err
+	}
 
 	// Set up Executor
 

pkg/server/tenant.go

@@ -1335,6 +1335,7 @@ func makeTenantSQLServerArgs(
 			nodeLiveness:      optionalnodeliveness.MakeContainer(nil),
 			gossip:            gossip.MakeOptionalGossip(nil),
 			grpcServer:        grpcServer.Server,
+			drpcMux:           drpcServer.DRPCServer,
 			isMeta1Leaseholder: func(_ context.Context, _ hlc.ClockTimestamp) (bool, error) {
 				return false, errors.New("isMeta1Leaseholder is not available to secondary tenants")
 			},