planner: add usage of paging copr in optimizer (pingcap#30536)

close pingcap#30578

distsql/distsql.go

@@ -133,6 +133,7 @@ func Select(ctx context.Context, sctx sessionctx.Context, kvReq *kv.Request, fie
 		memTracker: kvReq.MemTracker,
 		encodeType: encodetype,
 		storeType:  kvReq.StoreType,
+		paging:     kvReq.Paging,
 	}, nil
 }
 

distsql/select_result.go

@@ -152,7 +152,8 @@ type selectResult struct {
 	durationReported bool
 	memTracker       *memory.Tracker
 
-	stats *selectResultRuntimeStats
+	stats  *selectResultRuntimeStats
+	paging bool
 }
 
 func (r *selectResult) fetchResp(ctx context.Context) error {
@@ -206,7 +207,11 @@ func (r *selectResult) fetchResp(ctx context.Context) error {
 				// final round of fetch
 				// TODO: Add a label to distinguish between success or failure.
 				// https://github.com/pingcap/tidb/issues/11397
-				metrics.DistSQLQueryHistogram.WithLabelValues(r.label, r.sqlType).Observe(r.fetchDuration.Seconds())
+				if r.paging {
+					metrics.DistSQLQueryHistogram.WithLabelValues(r.label, r.sqlType, "paging").Observe(r.fetchDuration.Seconds())
+				} else {
+					metrics.DistSQLQueryHistogram.WithLabelValues(r.label, r.sqlType, "common").Observe(r.fetchDuration.Seconds())
+				}
 				r.durationReported = true
 			}
 			return nil

distsql/stream.go

@@ -82,7 +82,7 @@ func (r *streamResult) readDataFromResponse(ctx context.Context, resp kv.Respons
 		if !r.durationReported {
 			// TODO: Add a label to distinguish between success or failure.
 			// https://github.com/pingcap/tidb/issues/11397
-			metrics.DistSQLQueryHistogram.WithLabelValues(r.label, r.sqlType).Observe(r.fetchDuration.Seconds())
+			metrics.DistSQLQueryHistogram.WithLabelValues(r.label, r.sqlType, "streaming").Observe(r.fetchDuration.Seconds())
 			r.durationReported = true
 		}
 		return true, nil

executor/builder.go

@@ -3409,6 +3409,11 @@ func buildNoRangeIndexLookUpReader(b *executorBuilder, v *plannercore.PhysicalIn
 	if err != nil {
 		return nil, err
 	}
+	indexPaging := false
+	if v.Paging {
+		indexPaging = true
+		indexStreaming = false
+	}
 	tableReq, tableStreaming, tbl, err := buildTableReq(b, v.Schema().Len(), v.TablePlans)
 	if err != nil {
 		return nil, err
@@ -3430,6 +3435,7 @@ func buildNoRangeIndexLookUpReader(b *executorBuilder, v *plannercore.PhysicalIn
 		columns:           ts.Columns,
 		indexStreaming:    indexStreaming,
 		tableStreaming:    tableStreaming,
+		indexPaging:       indexPaging,
 		dataReaderBuilder: &dataReaderBuilder{executorBuilder: b},
 		corColInIdxSide:   b.corColInDistPlan(v.IndexPlans),
 		corColInTblSide:   b.corColInDistPlan(v.TablePlans),

executor/distsql.go

@@ -361,6 +361,7 @@ type IndexLookUpExecutor struct {
 
 	indexStreaming bool
 	tableStreaming bool
+	indexPaging    bool
 
 	corColInIdxSide bool
 	corColInTblSide bool
@@ -560,6 +561,7 @@ func (e *IndexLookUpExecutor) startIndexWorker(ctx context.Context, workCh chan<
 			SetDesc(e.desc).
 			SetKeepOrder(e.keepOrder).
 			SetStreaming(e.indexStreaming).
+			SetPaging(e.indexPaging).
 			SetReadReplicaScope(e.readReplicaScope).
 			SetIsStaleness(e.isStaleness).
 			SetFromSessionVars(e.ctx.GetSessionVars()).

metrics/distsql.go

@@ -27,7 +27,7 @@ var (
 			Name:      "handle_query_duration_seconds",
 			Help:      "Bucketed histogram of processing time (s) of handled queries.",
 			Buckets:   prometheus.ExponentialBuckets(0.0005, 2, 29), // 0.5ms ~ 1.5days
-		}, []string{LblType, LblSQLType})
+		}, []string{LblType, LblSQLType, LblCoprType})
 
 	DistSQLScanKeysPartialHistogram = prometheus.NewHistogram(
 		prometheus.HistogramOpts{

metrics/grafana/tidb.json

@@ -6168,10 +6168,10 @@
           "steppedLine": false,
           "targets": [
             {
-              "expr": "sum(rate(tidb_distsql_handle_query_duration_seconds_count{tidb_cluster=\"$tidb_cluster\"}[1m]))",
+              "expr": "sum(rate(tidb_distsql_handle_query_duration_seconds_count{tidb_cluster=\"$tidb_cluster\"}[1m])) by (copr_type)",
               "format": "time_series",
               "intervalFactor": 2,
-              "legendFormat": "",
+              "legendFormat": "{{copr_type}}",
               "metric": "tidb_distsql_query_total",
               "refId": "A",
               "step": 4

metrics/session.go

@@ -142,6 +142,7 @@ const (
 	LblDb          = "db"
 	LblResult      = "result"
 	LblSQLType     = "sql_type"
+	LblCoprType    = "copr_type"
 	LblGeneral     = "general"
 	LblInternal    = "internal"
 	LbTxnMode      = "txn_mode"

planner/core/explain.go

@@ -456,17 +456,22 @@ func (p *PhysicalIndexReader) accessObject(sctx sessionctx.Context) string {
 
 // ExplainInfo implements Plan interface.
 func (p *PhysicalIndexLookUpReader) ExplainInfo() string {
+	var str strings.Builder
 	// The children can be inferred by the relation symbol.
 	if p.PushedLimit != nil {
-		var str strings.Builder
 		str.WriteString("limit embedded(offset:")
 		str.WriteString(strconv.FormatUint(p.PushedLimit.Offset, 10))
 		str.WriteString(", count:")
 		str.WriteString(strconv.FormatUint(p.PushedLimit.Count, 10))
 		str.WriteString(")")
-		return str.String()
 	}
-	return ""
+	if p.Paging {
+		if p.PushedLimit != nil {
+			str.WriteString(", ")
+		}
+		str.WriteString("paging:true")
+	}
+	return str.String()
 }
 
 func (p *PhysicalIndexLookUpReader) accessObject(sctx sessionctx.Context) string {

planner/core/find_best_task.go

@@ -1249,6 +1249,7 @@ func (ds *DataSource) convertToIndexScan(prop *property.PhysicalProperty, candid
 		indexPlan:   is,
 		tblColHists: ds.TblColHists,
 		tblCols:     ds.TblCols,
+		expectCnt:   uint64(prop.ExpectedCnt),
 	}
 	cop.partitionInfo = PartitionInfo{
 		PruningConds:   ds.allConds,

planner/core/physical_plans.go

@@ -270,6 +270,7 @@ type PhysicalIndexLookUpReader struct {
 	TablePlans []PhysicalPlan
 	indexPlan  PhysicalPlan
 	tablePlan  PhysicalPlan
+	Paging     bool
 
 	ExtraHandleCol *expression.Column
 	// PushedLimit is used to avoid unnecessary table scan tasks of IndexLookUpReader.

planner/core/plan_test.go

@@ -18,6 +18,7 @@ import (
 	"bytes"
 	"fmt"
 	"strings"
+	"testing"
 
 	. "github.com/pingcap/check"
 	"github.com/pingcap/tidb/config"
@@ -26,6 +27,7 @@ import (
 	"github.com/pingcap/tidb/parser/model"
 	"github.com/pingcap/tidb/planner/core"
 	"github.com/pingcap/tidb/sessionctx/variable"
+	kit "github.com/pingcap/tidb/testkit"
 	"github.com/pingcap/tidb/util/israce"
 	"github.com/pingcap/tidb/util/plancodec"
 	"github.com/pingcap/tidb/util/testkit"
@@ -639,3 +641,65 @@ func (s *testPlanNormalize) TestIssue25729(c *C) {
 	tk.MustExec("insert into t1 values(\"a\", \"adwa\");")
 	tk.MustQuery("select * from t1  where concat(a, b) like \"aadwa\" and a = \"a\";").Check(testkit.Rows("a adwa"))
 }
+
+func TestCopPaging(t *testing.T) {
+	store, clean := kit.CreateMockStore(t)
+	defer clean()
+
+	tk := kit.NewTestKit(t, store)
+	tk.MustExec("use test")
+	tk.MustExec("drop table if exists t")
+	tk.MustExec("set session tidb_enable_paging = 1")
+	tk.MustExec("create table t(id int, c1 int, c2 int, primary key (id), key i(c1))")
+	defer tk.MustExec("drop table t")
+	for i := 0; i < 1024; i++ {
+		tk.MustExec("insert into t values(?, ?, ?)", i, i, i)
+	}
+	tk.MustExec("analyze table t")
+
+	// limit 960 should go paging
+	for i := 0; i < 10; i++ {
+		tk.MustQuery("explain format='brief' select * from t force index(i) where id <= 1024 and c1 >= 0 and c1 <= 1024 and c2 in (2, 4, 6, 8) order by c1 limit 960").Check(kit.Rows(
+			"Limit 4.00 root  offset:0, count:960",
+			"└─IndexLookUp 4.00 root  paging:true",
+			"  ├─Selection(Build) 1024.00 cop[tikv]  le(test.t.id, 1024)",
+			"  │ └─IndexRangeScan 1024.00 cop[tikv] table:t, index:i(c1) range:[0,1024], keep order:true",
+			"  └─Selection(Probe) 4.00 cop[tikv]  in(test.t.c2, 2, 4, 6, 8)",
+			"    └─TableRowIDScan 1024.00 cop[tikv] table:t keep order:false"))
+	}
+
+	// selection between limit and indexlookup, limit 960 should also go paging
+	for i := 0; i < 10; i++ {
+		tk.MustQuery("explain format='brief' select * from t force index(i) where mod(id, 2) > 0 and id <= 1024 and c1 >= 0 and c1 <= 1024 and c2 in (2, 4, 6, 8) order by c1 limit 960").Check(kit.Rows(
+			"Limit 3.20 root  offset:0, count:960",
+			"└─Selection 2.56 root  gt(mod(test.t.id, 2), 0)",
+			"  └─IndexLookUp 3.20 root  paging:true",
+			"    ├─Selection(Build) 819.20 cop[tikv]  le(test.t.id, 1024)",
+			"    │ └─IndexRangeScan 1024.00 cop[tikv] table:t, index:i(c1) range:[0,1024], keep order:true",
+			"    └─Selection(Probe) 3.20 cop[tikv]  in(test.t.c2, 2, 4, 6, 8)",
+			"      └─TableRowIDScan 819.20 cop[tikv] table:t keep order:false"))
+	}
+
+	// limit 961 exceeds the threshold, it should not go paging
+	for i := 0; i < 10; i++ {
+		tk.MustQuery("explain format='brief' select * from t force index(i) where id <= 1024 and c1 >= 0 and c1 <= 1024 and c2 in (2, 4, 6, 8) order by c1 limit 961").Check(kit.Rows(
+			"Limit 4.00 root  offset:0, count:961",
+			"└─IndexLookUp 4.00 root  ",
+			"  ├─Selection(Build) 1024.00 cop[tikv]  le(test.t.id, 1024)",
+			"  │ └─IndexRangeScan 1024.00 cop[tikv] table:t, index:i(c1) range:[0,1024], keep order:true",
+			"  └─Selection(Probe) 4.00 cop[tikv]  in(test.t.c2, 2, 4, 6, 8)",
+			"    └─TableRowIDScan 1024.00 cop[tikv] table:t keep order:false"))
+	}
+
+	// selection between limit and indexlookup, limit 961 should not go paging too
+	for i := 0; i < 10; i++ {
+		tk.MustQuery("explain format='brief' select * from t force index(i) where mod(id, 2) > 0 and id <= 1024 and c1 >= 0 and c1 <= 1024 and c2 in (2, 4, 6, 8) order by c1 limit 961").Check(kit.Rows(
+			"Limit 3.20 root  offset:0, count:961",
+			"└─Selection 2.56 root  gt(mod(test.t.id, 2), 0)",
+			"  └─IndexLookUp 3.20 root  ",
+			"    ├─Selection(Build) 819.20 cop[tikv]  le(test.t.id, 1024)",
+			"    │ └─IndexRangeScan 1024.00 cop[tikv] table:t, index:i(c1) range:[0,1024], keep order:true",
+			"    └─Selection(Probe) 3.20 cop[tikv]  in(test.t.c2, 2, 4, 6, 8)",
+			"      └─TableRowIDScan 819.20 cop[tikv] table:t keep order:false"))
+	}
+}

planner/core/task.go

@@ -35,6 +35,7 @@ import (
 	"github.com/pingcap/tidb/util/chunk"
 	"github.com/pingcap/tidb/util/collate"
 	"github.com/pingcap/tidb/util/logutil"
+	"github.com/pingcap/tidb/util/paging"
 	"github.com/pingcap/tidb/util/plancodec"
 	"github.com/pingcap/tipb/go-tipb"
 	"go.uber.org/zap"
@@ -89,6 +90,10 @@ type copTask struct {
 
 	// For table partition.
 	partitionInfo PartitionInfo
+
+	// expectCnt is the expected row count of upper task, 0 for unlimited.
+	// It's used for deciding whether using paging distsql.
+	expectCnt uint64
 }
 
 func (t *copTask) invalid() bool {
@@ -914,7 +919,17 @@ func buildIndexLookUpTask(ctx sessionctx.Context, t *copTask) *rootTask {
 	// (indexRows / batchSize) * batchSize * CPUFactor
 	// Since we don't know the number of copTasks built, ignore these network cost now.
 	indexRows := t.indexPlan.statsInfo().RowCount
-	newTask.cst += indexRows * sessVars.CPUFactor
+	idxCst := indexRows * sessVars.CPUFactor
+	// if the expectCnt is below the paging threshold, using paging API, recalculate idxCst.
+	// paging API reduces the count of index and table rows, however introduces more seek cost.
+	if ctx.GetSessionVars().EnablePaging && t.expectCnt > 0 && t.expectCnt <= paging.Threshold {
+		p.Paging = true
+		pagingCst := calcPagingCost(ctx, t)
+		// prevent enlarging the cost because we take paging as a better plan,
+		// if the cost is enlarged, it'll be easier to go another plan.
+		idxCst = math.Min(idxCst, pagingCst)
+	}
+	newTask.cst += idxCst
 	// Add cost of worker goroutines in index lookup.
 	numTblWorkers := float64(sessVars.IndexLookupConcurrency())
 	newTask.cst += (numTblWorkers + 1) * sessVars.ConcurrencyFactor
@@ -951,6 +966,41 @@ func buildIndexLookUpTask(ctx sessionctx.Context, t *copTask) *rootTask {
 	return newTask
 }
 
+func extractRows(p PhysicalPlan) float64 {
+	f := float64(0)
+	for _, c := range p.Children() {
+		if len(c.Children()) != 0 {
+			f += extractRows(c)
+		} else {
+			f += c.statsInfo().RowCount
+		}
+	}
+	return f
+}
+
+// calcPagingCost calculates the cost for paging processing which may increase the seekCnt and reduce scanned rows.
+func calcPagingCost(ctx sessionctx.Context, t *copTask) float64 {
+	sessVars := ctx.GetSessionVars()
+	indexRows := t.indexPlan.statsInfo().RowCount
+	expectCnt := t.expectCnt
+	sourceRows := extractRows(t.indexPlan)
+	// with paging, the scanned rows is always less than or equal to source rows.
+	if uint64(sourceRows) < expectCnt {
+		expectCnt = uint64(sourceRows)
+	}
+	seekCnt := paging.CalculateSeekCnt(expectCnt)
+	indexSelectivity := float64(1)
+	if sourceRows > indexRows {
+		indexSelectivity = indexRows / sourceRows
+	}
+	pagingCst := seekCnt*sessVars.GetSeekFactor(nil) + float64(expectCnt)*sessVars.CPUFactor
+	pagingCst *= indexSelectivity
+
+	// we want the diff between idxCst and pagingCst here,
+	// however, the idxCst does not contain seekFactor, so a seekFactor needs to be removed
+	return pagingCst - sessVars.GetSeekFactor(nil)
+}
+
 func (t *rootTask) convertToRootTask(_ sessionctx.Context) *rootTask {
 	return t.copy().(*rootTask)
 }

store/copr/coprocessor.go

@@ -42,6 +42,7 @@ import (
 	"github.com/pingcap/tidb/util/execdetails"
 	"github.com/pingcap/tidb/util/logutil"
 	"github.com/pingcap/tidb/util/memory"
+	"github.com/pingcap/tidb/util/paging"
 	"github.com/pingcap/tidb/util/trxevents"
 	"github.com/pingcap/tipb/go-tipb"
 	"github.com/tikv/client-go/v2/metrics"
@@ -61,18 +62,6 @@ const (
 	copNextMaxBackoff      = 20000
 )
 
-// A paging request may be separated into multi requests if there are more data than a page.
-// The paging size grows from min to max, it's not well tuned yet.
-// e.g. a paging request scans over range (r1, r200), it requires 64 rows in the first batch,
-// if it's not drained, then the paging size grows, the new range is calculated like (r100, r200), then send a request again.
-// Compare with the common unary request, paging request allows early access of data, it offers a streaming-like way processing data.
-// TODO: may make the paging parameters configurable.
-const (
-	minPagingSize  uint64 = 64
-	maxPagingSize         = minPagingSize * 128
-	pagingSizeGrow uint64 = 2
-)
-
 // CopClient is coprocessor client.
 type CopClient struct {
 	kv.RequestTypeSupportedChecker
@@ -212,7 +201,7 @@ func buildCopTasks(bo *Backoffer, cache *RegionCache, ranges *KeyRanges, req *kv
 			// the size will grow every round.
 			pagingSize := uint64(0)
 			if req.Paging {
-				pagingSize = minPagingSize
+				pagingSize = paging.MinPagingSize
 			}
 			tasks = append(tasks, &copTask{
 				region:     loc.Location.Region,
@@ -928,7 +917,7 @@ func (worker *copIteratorWorker) handleCopPagingResult(bo *Backoffer, rpcCtx *ti
 	if task.ranges.Len() == 0 {
 		return nil, nil
 	}
-	task.pagingSize = growPagingSize(task.pagingSize)
+	task.pagingSize = paging.GrowPagingSize(task.pagingSize)
 	return []*copTask{task}, nil
 }
 
@@ -1332,11 +1321,3 @@ func isolationLevelToPB(level kv.IsoLevel) kvrpcpb.IsolationLevel {
 		return kvrpcpb.IsolationLevel_SI
 	}
 }
-
-func growPagingSize(size uint64) uint64 {
-	size *= pagingSizeGrow
-	if size > maxPagingSize {
-		return maxPagingSize
-	}
-	return size
-}

store/copr/coprocessor_test.go

@@ -21,6 +21,7 @@ import (
 	"github.com/pingcap/kvproto/pkg/coprocessor"
 	"github.com/pingcap/tidb/kv"
 	"github.com/pingcap/tidb/store/driver/backoff"
+	"github.com/pingcap/tidb/util/paging"
 	"github.com/stretchr/testify/require"
 	"github.com/tikv/client-go/v2/testutils"
 	"github.com/tikv/client-go/v2/tikv"
@@ -318,7 +319,7 @@ func TestBuildPagingTasks(t *testing.T) {
 	require.Len(t, tasks, 1)
 	taskEqual(t, tasks[0], regionIDs[0], "a", "c")
 	require.True(t, tasks[0].paging)
-	require.Equal(t, tasks[0].pagingSize, minPagingSize)
+	require.Equal(t, tasks[0].pagingSize, paging.MinPagingSize)
 }
 
 func toCopRange(r kv.KeyRange) *coprocessor.KeyRange {