From 2f6219151967b23e461f98c0c9ed4197f1fed2bf Mon Sep 17 00:00:00 2001
From: ti-srebot <66930949+ti-srebot@users.noreply.github.com>
Date: Fri, 16 Oct 2020 16:47:20 +0800
Subject: [PATCH] *: add gc duration in metric profile (#20379) (#20480)

Signed-off-by: ti-srebot <ti-srebot@pingcap.com>
---
 executor/inspection_profile.go | 49 +++++++++++++++++++++++++++-------
 infoschema/metric_table_def.go | 12 ++++-----
 2 files changed, 46 insertions(+), 15 deletions(-)

diff --git a/executor/inspection_profile.go b/executor/inspection_profile.go
index 1768519a70b70..36d78fdac96f8 100644
--- a/executor/inspection_profile.go
+++ b/executor/inspection_profile.go
@@ -310,11 +310,16 @@ func NewProfileBuilder(sctx sessionctx.Context, start, end time.Time, tp string)
 
 // Collect uses to collect the related metric information.
 func (pb *profileBuilder) Collect() error {
-	pb.buf.WriteString(fmt.Sprintf(`digraph "%s" {`, "tidb_profile"))
-	pb.buf.WriteByte('\n')
-	pb.buf.WriteString(`node [style=filled fillcolor="#f8f8f8"]`)
-	pb.buf.WriteByte('\n')
-	err := pb.addMetricTree(pb.genTiDBQueryTree(), "tidb_query")
+	tidbQuery := pb.genTiDBQueryTree()
+	err := pb.init(tidbQuery, "tidb_query")
+	if err != nil {
+		return err
+	}
+	err = pb.traversal(tidbQuery)
+	if err != nil {
+		return err
+	}
+	err = pb.traversal(pb.genTiDBGCTree())
 	if err != nil {
 		return err
 	}
@@ -337,8 +342,8 @@ func (pb *profileBuilder) getNameID(name string) uint64 {
 	return id
 }
 
-func (pb *profileBuilder) addMetricTree(root *metricNode, name string) error {
-	if root == nil {
+func (pb *profileBuilder) init(total *metricNode, name string) error {
+	if total == nil {
 		return nil
 	}
 	tp := "total_time"
@@ -348,9 +353,13 @@ func (pb *profileBuilder) addMetricTree(root *metricNode, name string) error {
 	case metricValueCnt:
 		tp = "total_count"
 	}
+	pb.buf.WriteString(fmt.Sprintf(`digraph "%s" {`, "tidb_profile"))
+	pb.buf.WriteByte('\n')
+	pb.buf.WriteString(`node [style=filled fillcolor="#f8f8f8"]`)
+	pb.buf.WriteByte('\n')
 	pb.buf.WriteString(fmt.Sprintf(`subgraph %[1]s { "%[1]s" [shape=box fontsize=16 label="Type: %[1]s\lTime: %s\lDuration: %s\l"] }`, name+"_"+tp, pb.start.String(), pb.end.Sub(pb.start).String()))
 	pb.buf.WriteByte('\n')
-	v, err := pb.GetTotalValue(root)
+	v, err := pb.GetTotalValue(total)
 	if err != nil {
 		return err
 	}
@@ -359,7 +368,7 @@ func (pb *profileBuilder) addMetricTree(root *metricNode, name string) error {
 	} else {
 		pb.totalValue = 1
 	}
-	return pb.traversal(root)
+	return nil
 }
 
 func (pb *profileBuilder) GetTotalValue(root *metricNode) (float64, error) {
@@ -609,6 +618,21 @@ func (pb *profileBuilder) dotColor(score float64, isBackground bool) string {
 	return fmt.Sprintf("#%02x%02x%02x", uint8(r*255.0), uint8(g*255.0), uint8(b*255.0))
 }
 
+func (pb *profileBuilder) genTiDBGCTree() *metricNode {
+	tidbGC := &metricNode{
+		table:          "tidb_gc",
+		isPartOfParent: true,
+		label:          []string{"stage"},
+		children: []*metricNode{
+			{
+				table:          "tidb_kv_request",
+				isPartOfParent: true,
+			},
+		},
+	}
+	return tidbGC
+}
+
 func (pb *profileBuilder) genTiDBQueryTree() *metricNode {
 	tidbKVRequest := &metricNode{
 		table:          "tidb_kv_request",
@@ -676,6 +700,10 @@ func (pb *profileBuilder) genTiDBQueryTree() *metricNode {
 							},
 						},
 					},
+					{
+						table: "tikv_gc_tasks",
+						label: []string{"task"},
+					},
 				},
 			},
 		},
@@ -700,6 +728,9 @@ func (pb *profileBuilder) genTiDBQueryTree() *metricNode {
 					{
 						table: "tidb_owner_handle_syncer",
 					},
+					{
+						table: "tidb_meta_operation",
+					},
 				},
 			},
 		},
diff --git a/infoschema/metric_table_def.go b/infoschema/metric_table_def.go
index 4e09f23bc40f5..a9df46bbded4e 100644
--- a/infoschema/metric_table_def.go
+++ b/infoschema/metric_table_def.go
@@ -605,8 +605,8 @@ var MetricTableMap = map[string]MetricTableDef{
 	},
 	"tidb_gc_duration": {
 		Comment:  "The quantile of kv storage garbage collection time durations",
-		PromQL:   "histogram_quantile($QUANTILE, sum(rate(tidb_tikvclient_gc_seconds_bucket{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (le,instance))",
-		Labels:   []string{"instance"},
+		PromQL:   "histogram_quantile($QUANTILE, sum(rate(tidb_tikvclient_gc_seconds_bucket{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (le,instance,stage))",
+		Labels:   []string{"instance", "stage"},
 		Quantile: 0.95,
 	},
 	"tidb_gc_config": {
@@ -2522,13 +2522,13 @@ var MetricTableMap = map[string]MetricTableDef{
 		Comment: "The total time of kv storage range worker processing one task duration",
 	},
 	"tidb_gc_total_count": {
-		PromQL:  "sum(increase(tidb_tikvclient_gc_seconds_count{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (instance)",
-		Labels:  []string{"instance"},
+		PromQL:  "sum(increase(tidb_tikvclient_gc_seconds_count{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (instance,stage)",
+		Labels:  []string{"instance", "stage"},
 		Comment: "The total count of kv storage garbage collection",
 	},
 	"tidb_gc_total_time": {
-		PromQL:  "sum(increase(tidb_tikvclient_gc_seconds_sum{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (instance)",
-		Labels:  []string{"instance"},
+		PromQL:  "sum(increase(tidb_tikvclient_gc_seconds_sum{$LABEL_CONDITIONS}[$RANGE_DURATION])) by (instance,stage)",
+		Labels:  []string{"instance", "stage"},
 		Comment: "The total time of kv storage garbage collection time durations",
 	},
 	"tidb_get_token_total_count": {