wip

lib/logstorage/block_result.go

@@ -208,9 +208,22 @@ func (br *blockResult) sizeBytes() int {
 	return n
 }
 
+// addResultColumns adds the given rcs to br.
+//
+// The br is valid only until rcs are modified.
+func (br *blockResult) addResultColumns(rcs []resultColumn) {
+	if len(rcs) == 0 || len(rcs[0].values) == 0 {
+		return
+	}
+
+	for i := range rcs {
+		br.addResultColumn(&rcs[i])
+	}
+}
+
 // setResultColumns sets the given rcs as br columns.
 //
-// The returned result is valid only until rcs are modified.
+// The br is valid only until rcs are modified.
 func (br *blockResult) setResultColumns(rcs []resultColumn) {
 	br.reset()
 
@@ -220,24 +233,29 @@ func (br *blockResult) setResultColumns(rcs []resultColumn) {
 
 	br.timestamps = fastnum.AppendInt64Zeros(br.timestamps[:0], len(rcs[0].values))
 
-	csBuf := br.csBuf
-	for _, rc := range rcs {
+	for i := range rcs {
+		br.addResultColumn(&rcs[i])
+	}
+}
+
+func (br *blockResult) addResultColumn(rc *resultColumn) {
+	if len(rc.values) != len(br.timestamps) {
+		logger.Panicf("BUG: column %q must contain %d rows, but it contains %d rows", rc.name, len(br.timestamps), len(rc.values))
+	}
 	if areConstValues(rc.values) {
 		// This optimization allows reducing memory usage after br cloning
-			csBuf = append(csBuf, blockResultColumn{
-				name:          br.a.copyString(rc.name),
+		br.csBuf = append(br.csBuf, blockResultColumn{
+			name:          rc.name,
 			isConst:       true,
 			valuesEncoded: rc.values[:1],
 		})
 	} else {
-			csBuf = append(csBuf, blockResultColumn{
-				name:          br.a.copyString(rc.name),
+		br.csBuf = append(br.csBuf, blockResultColumn{
+			name:          rc.name,
 			valueType:     valueTypeString,
 			valuesEncoded: rc.values,
 		})
 	}
-	}
-	br.csBuf = csBuf
 	br.csInitialized = false
 }
 
@@ -1765,37 +1783,25 @@ func (c *blockResultColumn) sumValues(br *blockResult) (float64, int) {
 	}
 }
 
-// resultColumn represents a column with result values
+// resultColumn represents a column with result values.
+//
+// It doesn't own the result values.
 type resultColumn struct {
 	// name is column name.
 	name string
 
-	// a contains values data.
-	a arena
-
-	// values is the result values. They are backed by data inside a.
+	// values is the result values.
 	values []string
 }
 
-func (rc *resultColumn) sizeBytes() int {
-	return len(rc.name) + rc.a.sizeBytes() + len(rc.values)*int(unsafe.Sizeof(rc.values[0]))
-}
-
 func (rc *resultColumn) resetKeepName() {
-	rc.a.reset()
 	clear(rc.values)
 	rc.values = rc.values[:0]
 }
 
 // addValue adds the given values v to rc.
 func (rc *resultColumn) addValue(v string) {
-	values := rc.values
-	if len(values) > 0 && string(v) == values[len(values)-1] {
-		v = values[len(values)-1]
-	} else {
-		v = rc.a.copyString(v)
-	}
-	rc.values = append(values, v)
+	rc.values = append(rc.values, v)
 }
 
 func truncateTimestampToMonth(timestamp int64) int64 {

lib/logstorage/pipe_extract.go

@@ -37,9 +37,15 @@ func (pe *pipeExtract) updateNeededFields(neededFields, unneededFields fieldsSet
 func (pe *pipeExtract) newPipeProcessor(workersCount int, stopCh <-chan struct{}, _ func(), ppBase pipeProcessor) pipeProcessor {
 	shards := make([]pipeExtractProcessorShard, workersCount)
 	for i := range shards {
+		ef := newExtractFormat(pe.steps)
+		rcs := make([]resultColumn, len(ef.fields))
+		for j := range rcs {
+			rcs[j].name = ef.fields[j].name
+		}
 		shards[i] = pipeExtractProcessorShard{
 			pipeExtractProcessorShardNopad: pipeExtractProcessorShardNopad{
-				ef: newExtractFormat(pe.steps),
+				ef:  ef,
+				rcs: rcs,
 			},
 		}
 	}
@@ -71,6 +77,8 @@ type pipeExtractProcessorShard struct {
 
 type pipeExtractProcessorShardNopad struct {
 	ef *extractFormat
+
+	rcs []resultColumn
 }
 
 func (pep *pipeExtractProcessor) writeBlock(workerID uint, br *blockResult) {
@@ -83,12 +91,18 @@ func (pep *pipeExtractProcessor) writeBlock(workerID uint, br *blockResult) {
 	values := c.getValues(br)
 
 	ef := shard.ef
+	rcs := shard.rcs
 	for _, v := range values {
 		ef.apply(v)
-		/*		for i, result := range ef.results {
-					rcs[i].addValue(result)
+		for i, f := range ef.fields {
+			rcs[i].addValue(*f.value)
 		}
-		*/
+	}
+	br.addResultColumns(rcs)
+	pep.ppBase.writeBlock(workerID, br)
+
+	for i := range rcs {
+		rcs[i].resetKeepName()
 	}
 }
 

lib/logstorage/pipe_sort.go

@@ -11,7 +11,6 @@ import (
 	"sync/atomic"
 	"unsafe"
 
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/stringsutil"
 )
@@ -211,7 +210,8 @@ func (shard *pipeSortProcessorShard) writeBlock(br *blockResult) {
 		shard.columnValues = columnValues
 
 		// Generate byColumns
-		var rc resultColumn
+		valuesEncoded := make([]string, len(br.timestamps))
+		shard.stateSizeBudget -= len(valuesEncoded) * int(unsafe.Sizeof(valuesEncoded[0]))
 
 		bb := bbPool.Get()
 		for rowIdx := range br.timestamps {
@@ -223,7 +223,12 @@ func (shard *pipeSortProcessorShard) writeBlock(br *blockResult) {
 				bb.B = marshalJSONKeyValue(bb.B, cs[i].name, v)
 				bb.B = append(bb.B, ',')
 			}
-			rc.addValue(bytesutil.ToUnsafeString(bb.B))
+			if rowIdx > 0 && valuesEncoded[rowIdx-1] == string(bb.B) {
+				valuesEncoded[rowIdx] = valuesEncoded[rowIdx-1]
+			} else {
+				valuesEncoded[rowIdx] = string(bb.B)
+				shard.stateSizeBudget -= len(bb.B)
+			}
 		}
 		bbPool.Put(bb)
 
@@ -236,13 +241,13 @@ func (shard *pipeSortProcessorShard) writeBlock(br *blockResult) {
 			{
 				c: &blockResultColumn{
 					valueType:     valueTypeString,
-					valuesEncoded: rc.values,
+					valuesEncoded: valuesEncoded,
 				},
 				i64Values: i64Values,
 				f64Values: f64Values,
 			},
 		}
-		shard.stateSizeBudget -= rc.sizeBytes() + int(unsafe.Sizeof(byColumns[0])+unsafe.Sizeof(*byColumns[0].c))
+		shard.stateSizeBudget -= int(unsafe.Sizeof(byColumns[0]) + unsafe.Sizeof(*byColumns[0].c))
 
 		// Append br to shard.blocks.
 		shard.blocks = append(shard.blocks, sortBlock{