Connection retaining mode for p2p peer chooser (#6471)

cadence-workflow · Nov 12, 2024 · d2d1d47 · d2d1d47
1 parent 9d328c7
commit d2d1d47
Show file tree

Hide file tree

Showing 18 changed files with 564 additions and 146 deletions.
diff --git a/cmd/server/cadence/server.go b/cmd/server/cadence/server.go
@@ -155,8 +155,9 @@ func (s *server) startService() common.Daemon {
 	)
 
 	params.MetricScope = svcCfg.Metrics.NewScope(params.Logger, params.Name)
+	params.MetricsClient = metrics.NewClient(params.MetricScope, service.GetMetricsServiceIdx(params.Name, params.Logger))
 
-	rpcParams, err := rpc.NewParams(params.Name, s.cfg, dc, params.Logger)
+	rpcParams, err := rpc.NewParams(params.Name, s.cfg, dc, params.Logger, params.MetricsClient)
 	if err != nil {
 		log.Fatalf("error creating rpc factory params: %v", err)
 	}
@@ -182,8 +183,6 @@ func (s *server) startService() common.Daemon {
 		log.Fatalf("ringpop provider failed: %v", err)
 	}
 
-	params.MetricsClient = metrics.NewClient(params.MetricScope, service.GetMetricsServiceIdx(params.Name, params.Logger))
-
 	params.MembershipResolver, err = membership.NewResolver(
 		peerProvider,
 		params.Logger,

diff --git a/common/log/tag/tags.go b/common/log/tag/tags.go
@@ -408,6 +408,11 @@ func Service(sv string) Tag {
 	return newStringTag("service", sv)
 }
 
+// DestService returns tag for destination service
+func DestService(sv string) Tag {
+	return newStringTag("dest-service", sv)
+}
+
 // Addresses returns tag for Addresses
 func Addresses(ads []string) Tag {
 	return newObjectTag("addresses", ads)

diff --git a/common/metrics/defs.go b/common/metrics/defs.go
@@ -846,6 +846,9 @@ const (
 	// GlobalRatelimiterAggregator is the metrics scope for aggregator-side common/quotas/global behavior
 	GlobalRatelimiterAggregator
 
+	// P2PRPCPeerChooserScope is the metrics scope for P2P RPC peer chooser
+	P2PRPCPeerChooserScope
+
 	NumCommonScopes
 )
 
@@ -1741,6 +1744,8 @@ var ScopeDefs = map[ServiceIdx]map[int]scopeDefinition{
 		// currently used by both frontend and history, but may grow to other limiting-host-services.
 		GlobalRatelimiter:           {operation: "GlobalRatelimiter"},
 		GlobalRatelimiterAggregator: {operation: "GlobalRatelimiterAggregator"},
+
+		P2PRPCPeerChooserScope: {operation: "P2PRPCPeerChooser"},
 	},
 	// Frontend Scope Names
 	Frontend: {
@@ -2243,6 +2248,11 @@ const (
 	GlobalRatelimiterRemovedLimits
 	GlobalRatelimiterRemovedHostLimits
 
+	// p2p rpc metrics
+	P2PPeersCount
+	P2PPeerAdded
+	P2PPeerRemoved
+
 	NumCommonMetrics // Needs to be last on this list for iota numbering
 )
 
@@ -2936,6 +2946,10 @@ var MetricDefs = map[ServiceIdx]map[int]metricDefinition{
 		GlobalRatelimiterHostLimitsQueried: {metricName: "global_ratelimiter_host_limits_queried", metricType: Histogram, buckets: GlobalRatelimiterUsageHistogram},
 		GlobalRatelimiterRemovedLimits:     {metricName: "global_ratelimiter_removed_limits", metricType: Histogram, buckets: GlobalRatelimiterUsageHistogram},
 		GlobalRatelimiterRemovedHostLimits: {metricName: "global_ratelimiter_removed_host_limits", metricType: Histogram, buckets: GlobalRatelimiterUsageHistogram},
+
+		P2PPeersCount:  {metricName: "peers_count", metricType: Gauge},
+		P2PPeerAdded:   {metricName: "peer_added", metricType: Counter},
+		P2PPeerRemoved: {metricName: "peer_removed", metricType: Counter},
 	},
 	History: {
 		TaskRequests:             {metricName: "task_requests", metricType: Counter},

diff --git a/common/metrics/tags.go b/common/metrics/tags.go
@@ -53,6 +53,7 @@ const (
 	transport                 = "transport"
 	caller                    = "caller"
 	service                   = "service"
+	destService               = "dest_service"
 	signalName                = "signalName"
 	workflowVersion           = "workflow_version"
 	shardID                   = "shard_id"
@@ -219,11 +220,16 @@ func CallerTag(value string) Tag {
 	return simpleMetric{key: caller, value: value}
 }
 
-// CallerTag returns a new RPC Caller type tag.
+// ServiceTag returns a new service tag.
 func ServiceTag(value string) Tag {
 	return simpleMetric{key: service, value: value}
 }
 
+// DestServiceTag returns a new destination service tag.
+func DestServiceTag(value string) Tag {
+	return simpleMetric{key: destService, value: value}
+}
+
 // Hosttag emits the host identifier
 func HostTag(value string) Tag {
 	return metricWithUnknown(host, value)

diff --git a/common/resource/resource_impl.go b/common/resource/resource_impl.go
@@ -395,7 +395,9 @@ func (h *Impl) Start() {
 		h.logger.WithTags(tag.Error(err)).Fatal("fail to start PProf")
 	}
 
-	h.rpcFactory.Start(h.membershipResolver)
+	if err := h.rpcFactory.Start(h.membershipResolver); err != nil {
+		h.logger.WithTags(tag.Error(err)).Fatal("fail to start RPC factory")
+	}
 
 	if err := h.dispatcher.Start(); err != nil {
 		h.logger.WithTags(tag.Error(err)).Fatal("fail to start dispatcher")

diff --git a/common/resource/resource_impl_test.go b/common/resource/resource_impl_test.go
@@ -82,7 +82,7 @@ func TestStartStop(t *testing.T) {
 		"primary-cluster":   {InitialFailoverVersion: 1, Enabled: true, RPCTransport: "tchannel", RPCAddress: "localhost:0"},
 		"secondary-cluster": {InitialFailoverVersion: 1, Enabled: true, RPCTransport: "tchannel", RPCAddress: "localhost:0"},
 	}, nil, metricsCl, logger)
-	directOutboundPCF := rpc.NewDirectPeerChooserFactory(serviceName, logger)
+	directOutboundPCF := rpc.NewDirectPeerChooserFactory(serviceName, logger, metricsCl)
 	directConnRetainFn := func(opts ...dynamicconfig.FilterOption) bool { return false }
 	pcf := rpc.NewMockPeerChooserFactory(ctrl)
 	peerChooser := rpc.NewMockPeerChooser(ctrl)