func (udc *UpsideDownCouch) Update(doc *document.Document) (err error) {
// do analysis before acquiring write lock
analysisStart := time.Now()
resultChan := make(chan *index.AnalysisResult)
aw := index.NewAnalysisWork(udc, doc, resultChan)
// put the work on the queue
go func() {
udc.analysisQueue.Queue(aw)
}()
// wait for the result
result := <-resultChan
close(resultChan)
atomic.AddUint64(&udc.stats.analysisTime, uint64(time.Since(analysisStart)))
// start a writer for this update
indexStart := time.Now()
var kvwriter store.KVWriter
kvwriter, err = udc.store.Writer()
if err != nil {
return
}
defer func() {
if cerr := kvwriter.Close(); err == nil && cerr != nil {
err = cerr
}
}()
// first we lookup the backindex row for the doc id if it exists
// lookup the back index row
var backIndexRow *BackIndexRow
backIndexRow, err = udc.backIndexRowForDoc(kvwriter, doc.ID)
if err != nil {
atomic.AddUint64(&udc.stats.errors, 1)
return
}
// prepare a list of rows
addRows := make([]UpsideDownCouchRow, 0)
updateRows := make([]UpsideDownCouchRow, 0)
deleteRows := make([]UpsideDownCouchRow, 0)
addRows, updateRows, deleteRows = udc.mergeOldAndNew(backIndexRow, result.Rows, addRows, updateRows, deleteRows)
err = udc.batchRows(kvwriter, addRows, updateRows, deleteRows)
if err == nil && backIndexRow == nil {
udc.m.Lock()
udc.docCount++
udc.m.Unlock()
}
atomic.AddUint64(&udc.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
atomic.AddUint64(&udc.stats.updates, 1)
} else {
atomic.AddUint64(&udc.stats.errors, 1)
}
return
}
func (f *Firestorm) Update(doc *document.Document) (err error) {
// assign this document a number
doc.Number = atomic.AddUint64(&f.highDocNumber, 1)
// do analysis before acquiring write lock
analysisStart := time.Now()
numPlainTextBytes := doc.NumPlainTextBytes()
resultChan := make(chan *index.AnalysisResult)
aw := index.NewAnalysisWork(f, doc, resultChan)
// put the work on the queue
f.analysisQueue.Queue(aw)
// wait for the result
result := <-resultChan
close(resultChan)
atomic.AddUint64(&f.stats.analysisTime, uint64(time.Since(analysisStart)))
// start a writer for this update
indexStart := time.Now()
var kvwriter store.KVWriter
kvwriter, err = f.store.Writer()
if err != nil {
return
}
defer func() {
if cerr := kvwriter.Close(); err == nil && cerr != nil {
err = cerr
}
}()
var dictionaryDeltas map[string]int64
dictionaryDeltas, err = f.batchRows(kvwriter, [][]index.IndexRow{result.Rows}, nil)
if err != nil {
_ = kvwriter.Close()
atomic.AddUint64(&f.stats.errors, 1)
return
}
f.compensator.Mutate([]byte(doc.ID), doc.Number)
f.lookuper.NotifyBatch([]*InFlightItem{{[]byte(doc.ID), doc.Number}})
f.dictUpdater.NotifyBatch(dictionaryDeltas)
atomic.AddUint64(&f.stats.indexTime, uint64(time.Since(indexStart)))
atomic.AddUint64(&f.stats.numPlainTextBytesIndexed, numPlainTextBytes)
return
}
func (udc *UpsideDownCouch) Batch(batch *index.Batch) (err error) {
analysisStart := time.Now()
resultChan := make(chan *index.AnalysisResult)
var numUpdates uint64
for _, doc := range batch.IndexOps {
if doc != nil {
numUpdates++
}
}
var detectedUnsafeMutex sync.RWMutex
detectedUnsafe := false
go func() {
sofar := uint64(0)
for _, doc := range batch.IndexOps {
if doc != nil {
sofar++
if sofar > numUpdates {
detectedUnsafeMutex.Lock()
detectedUnsafe = true
detectedUnsafeMutex.Unlock()
return
}
aw := index.NewAnalysisWork(udc, doc, resultChan)
// put the work on the queue
udc.analysisQueue.Queue(aw)
}
}
}()
newRowsMap := make(map[string][]index.IndexRow)
// wait for the result
var itemsDeQueued uint64
for itemsDeQueued < numUpdates {
result := <-resultChan
newRowsMap[result.DocID] = result.Rows
itemsDeQueued++
}
close(resultChan)
detectedUnsafeMutex.RLock()
defer detectedUnsafeMutex.RUnlock()
if detectedUnsafe {
return UnsafeBatchUseDetected
}
atomic.AddUint64(&udc.stats.analysisTime, uint64(time.Since(analysisStart)))
indexStart := time.Now()
// start a writer for this batch
var kvwriter store.KVWriter
kvwriter, err = udc.store.Writer()
if err != nil {
return
}
// first lookup all the back index rows
var backIndexRows map[string]*BackIndexRow
backIndexRows, err = udc.backIndexRowsForBatch(kvwriter, batch)
if err != nil {
_ = kvwriter.Close()
return
}
// prepare a list of rows
addRows := make([]UpsideDownCouchRow, 0)
updateRows := make([]UpsideDownCouchRow, 0)
deleteRows := make([]UpsideDownCouchRow, 0)
docsAdded := uint64(0)
docsDeleted := uint64(0)
for docID, doc := range batch.IndexOps {
backIndexRow := backIndexRows[docID]
if doc == nil && backIndexRow != nil {
// delete
deleteRows = udc.deleteSingle(docID, backIndexRow, deleteRows)
docsDeleted++
} else if doc != nil {
addRows, updateRows, deleteRows = udc.mergeOldAndNew(backIndexRow, newRowsMap[docID], addRows, updateRows, deleteRows)
if backIndexRow == nil {
docsAdded++
}
}
}
// add the internal ops
for internalKey, internalValue := range batch.InternalOps {
if internalValue == nil {
// delete
deleteInternalRow := NewInternalRow([]byte(internalKey), nil)
deleteRows = append(deleteRows, deleteInternalRow)
} else {
updateInternalRow := NewInternalRow([]byte(internalKey), internalValue)
updateRows = append(updateRows, updateInternalRow)
}
}
err = udc.batchRows(kvwriter, addRows, updateRows, deleteRows)
if err != nil {
_ = kvwriter.Close()
atomic.AddUint64(&udc.stats.errors, 1)
return
}
err = kvwriter.Close()
atomic.AddUint64(&udc.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
udc.m.Lock()
udc.docCount += docsAdded
udc.docCount -= docsDeleted
udc.m.Unlock()
atomic.AddUint64(&udc.stats.updates, numUpdates)
atomic.AddUint64(&udc.stats.deletes, docsDeleted)
atomic.AddUint64(&udc.stats.batches, 1)
} else {
atomic.AddUint64(&udc.stats.errors, 1)
}
return
}
func (f *Firestorm) Batch(batch *index.Batch) (err error) {
// acquire enough doc numbers for all updates in the batch
// FIXME we actually waste doc numbers because deletes are in the
// same map and we don't need numbers for them
lastDocNumber := atomic.AddUint64(&f.highDocNumber, uint64(len(batch.IndexOps)))
firstDocNumber := lastDocNumber - uint64(len(batch.IndexOps)) + 1
analysisStart := time.Now()
resultChan := make(chan *index.AnalysisResult)
var docsUpdated uint64
var docsDeleted uint64
for _, doc := range batch.IndexOps {
if doc != nil {
doc.Number = firstDocNumber // actually assign doc numbers here
firstDocNumber++
docsUpdated++
} else {
docsDeleted++
}
}
var detectedUnsafeMutex sync.RWMutex
detectedUnsafe := false
go func() {
sofar := uint64(0)
for _, doc := range batch.IndexOps {
if doc != nil {
sofar++
if sofar > docsUpdated {
detectedUnsafeMutex.Lock()
detectedUnsafe = true
detectedUnsafeMutex.Unlock()
return
}
aw := index.NewAnalysisWork(f, doc, resultChan)
// put the work on the queue
f.analysisQueue.Queue(aw)
}
}
}()
// extra 1 capacity for internal updates.
collectRows := make([][]index.IndexRow, 0, docsUpdated+1)
// wait for the result
var itemsDeQueued uint64
for itemsDeQueued < docsUpdated {
result := <-resultChan
collectRows = append(collectRows, result.Rows)
itemsDeQueued++
}
close(resultChan)
detectedUnsafeMutex.RLock()
defer detectedUnsafeMutex.RUnlock()
if detectedUnsafe {
return UnsafeBatchUseDetected
}
atomic.AddUint64(&f.stats.analysisTime, uint64(time.Since(analysisStart)))
var deleteKeys [][]byte
if len(batch.InternalOps) > 0 {
// add the internal ops
updateInternalRows := make([]index.IndexRow, 0, len(batch.InternalOps))
for internalKey, internalValue := range batch.InternalOps {
if internalValue == nil {
// delete
deleteInternalRow := NewInternalRow([]byte(internalKey), nil)
deleteKeys = append(deleteKeys, deleteInternalRow.Key())
} else {
updateInternalRow := NewInternalRow([]byte(internalKey), internalValue)
updateInternalRows = append(updateInternalRows, updateInternalRow)
}
}
collectRows = append(collectRows, updateInternalRows)
}
inflightItems := make([]*InFlightItem, 0, len(batch.IndexOps))
for docID, doc := range batch.IndexOps {
if doc != nil {
inflightItems = append(inflightItems,
&InFlightItem{[]byte(docID), doc.Number})
} else {
inflightItems = append(inflightItems,
&InFlightItem{[]byte(docID), 0})
}
}
indexStart := time.Now()
// start a writer for this batch
var kvwriter store.KVWriter
kvwriter, err = f.store.Writer()
if err != nil {
return
}
var dictionaryDeltas map[string]int64
dictionaryDeltas, err = f.batchRows(kvwriter, collectRows, deleteKeys)
if err != nil {
_ = kvwriter.Close()
atomic.AddUint64(&f.stats.errors, 1)
return
}
f.compensator.MutateBatch(inflightItems, lastDocNumber)
err = kvwriter.Close()
f.lookuper.NotifyBatch(inflightItems)
f.dictUpdater.NotifyBatch(dictionaryDeltas)
atomic.AddUint64(&f.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
atomic.AddUint64(&f.stats.updates, docsUpdated)
atomic.AddUint64(&f.stats.deletes, docsDeleted)
atomic.AddUint64(&f.stats.batches, 1)
} else {
atomic.AddUint64(&f.stats.errors, 1)
}
return
}
func (udc *SmolderingCouch) Batch(batch *index.Batch) (err error) {
// acquire enough doc numbers for all updates in the batch
// FIXME we actually waste doc numbers because deletes are in the
// same map and we don't need numbers for them
lastDocNumber := atomic.AddUint64(&udc.maxInternalDocID, uint64(len(batch.IndexOps)))
nextDocNumber := lastDocNumber - uint64(len(batch.IndexOps)) + 1
analysisStart := time.Now()
resultChan := make(chan *index.AnalysisResult, len(batch.IndexOps))
var numUpdates uint64
var numPlainTextBytes uint64
for _, doc := range batch.IndexOps {
if doc != nil {
doc.Number = nextDocNumber // actually assign doc numbers here
nextDocNumber++
numUpdates++
numPlainTextBytes += doc.NumPlainTextBytes()
}
}
go func() {
for _, doc := range batch.IndexOps {
if doc != nil {
aw := index.NewAnalysisWork(udc, doc, resultChan)
// put the work on the queue
udc.analysisQueue.Queue(aw)
}
}
}()
// retrieve back index rows concurrent with analysis
docBackIndexRowErr := error(nil)
docBackIndexRowCh := make(chan *docBackIndexRow, len(batch.IndexOps))
udc.writeMutex.Lock()
defer udc.writeMutex.Unlock()
go func() {
defer close(docBackIndexRowCh)
// open a reader for backindex lookup
indexReader, err := udc.reader()
if err != nil {
docBackIndexRowErr = err
return
}
for docID, doc := range batch.IndexOps {
var backIndexRow *BackIndexRow
if udc.cf.Lookup([]byte(docID)) {
backIndexRow, err = indexReader.backIndexRowForDoc(nil, docID)
if err != nil {
docBackIndexRowErr = err
return
}
}
var docNumber []byte
if backIndexRow != nil {
docNumber = backIndexRow.docNumber
}
docBackIndexRowCh <- &docBackIndexRow{docNumber, doc, backIndexRow}
}
err = indexReader.Close()
if err != nil {
docBackIndexRowErr = err
return
}
}()
// wait for analysis result
newRowsMap := make(map[string][]index.IndexRow)
var itemsDeQueued uint64
for itemsDeQueued < numUpdates {
result := <-resultChan
newRowsMap[result.DocID] = result.Rows
itemsDeQueued++
}
close(resultChan)
atomic.AddUint64(&udc.stats.analysisTime, uint64(time.Since(analysisStart)))
docsAdded := uint64(0)
docsDeleted := uint64(0)
indexStart := time.Now()
// prepare a list of rows
var addRowsAll [][]SmolderingCouchRow
var updateRowsAll [][]SmolderingCouchRow
var deleteRowsAll [][]SmolderingCouchRow
// add the internal ops
var updateRows []SmolderingCouchRow
var deleteRows []SmolderingCouchRow
for internalKey, internalValue := range batch.InternalOps {
if internalValue == nil {
// delete
deleteInternalRow := NewInternalRow([]byte(internalKey), nil)
deleteRows = append(deleteRows, deleteInternalRow)
} else {
updateInternalRow := NewInternalRow([]byte(internalKey), internalValue)
updateRows = append(updateRows, updateInternalRow)
}
}
if len(updateRows) > 0 {
updateRowsAll = append(updateRowsAll, updateRows)
}
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
// process back index rows as they arrive
for dbir := range docBackIndexRowCh {
if dbir.doc == nil && dbir.backIndexRow != nil {
// delete
deleteRows := udc.deleteSingle(dbir.docID, dbir.backIndexRow, nil)
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
docsDeleted++
} else if dbir.doc != nil {
addRows, updateRows, deleteRows := udc.mergeOldAndNew(dbir.doc.ID, dbir.backIndexRow, newRowsMap[dbir.doc.ID])
if len(addRows) > 0 {
addRowsAll = append(addRowsAll, addRows)
}
if len(updateRows) > 0 {
updateRowsAll = append(updateRowsAll, updateRows)
}
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
if dbir.backIndexRow == nil {
docsAdded++
}
}
}
if docBackIndexRowErr != nil {
return docBackIndexRowErr
}
// start a writer for this batch
var kvwriter store.KVWriter
kvwriter, err = udc.store.Writer()
if err != nil {
return
}
err = udc.batchRows(kvwriter, addRowsAll, updateRowsAll, deleteRowsAll)
if err != nil {
_ = kvwriter.Close()
atomic.AddUint64(&udc.stats.errors, 1)
return
}
err = kvwriter.Close()
atomic.AddUint64(&udc.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
udc.m.Lock()
udc.docCount += docsAdded
udc.docCount -= docsDeleted
udc.m.Unlock()
for did, doc := range batch.IndexOps {
if doc != nil {
udc.cf.Insert([]byte(did))
} else {
udc.cf.Delete([]byte(did))
}
}
atomic.AddUint64(&udc.stats.updates, numUpdates)
atomic.AddUint64(&udc.stats.deletes, docsDeleted)
atomic.AddUint64(&udc.stats.batches, 1)
atomic.AddUint64(&udc.stats.numPlainTextBytesIndexed, numPlainTextBytes)
} else {
atomic.AddUint64(&udc.stats.errors, 1)
}
return
}
func (udc *SmolderingCouch) Update(doc *document.Document) (err error) {
// get the next available doc number
doc.Number = atomic.AddUint64(&udc.maxInternalDocID, 1)
analysisStart := time.Now()
numPlainTextBytes := doc.NumPlainTextBytes()
resultChan := make(chan *index.AnalysisResult)
aw := index.NewAnalysisWork(udc, doc, resultChan)
// put the work on the queue
udc.analysisQueue.Queue(aw)
// wait for the result
result := <-resultChan
close(resultChan)
atomic.AddUint64(&udc.stats.analysisTime, uint64(time.Since(analysisStart)))
udc.writeMutex.Lock()
defer udc.writeMutex.Unlock()
indexReader, err := udc.reader()
if err != nil {
return
}
// first we lookup the backindex row for the doc id if it exists
// lookup the back index row
var backIndexRow *BackIndexRow
if udc.cf.Lookup([]byte(doc.ID)) {
backIndexRow, err = indexReader.backIndexRowForDoc(nil, doc.ID)
if err != nil {
_ = indexReader.Close()
atomic.AddUint64(&udc.stats.errors, 1)
return
}
}
err = indexReader.Close()
if err != nil {
return
}
// start a writer for this update
indexStart := time.Now()
var kvwriter store.KVWriter
kvwriter, err = udc.store.Writer()
if err != nil {
return
}
defer func() {
if cerr := kvwriter.Close(); err == nil && cerr != nil {
err = cerr
}
}()
// prepare a list of rows
var addRowsAll [][]SmolderingCouchRow
var updateRowsAll [][]SmolderingCouchRow
var deleteRowsAll [][]SmolderingCouchRow
addRows, updateRows, deleteRows := udc.mergeOldAndNew(doc.ID, backIndexRow, result.Rows)
if len(addRows) > 0 {
addRowsAll = append(addRowsAll, addRows)
}
if len(updateRows) > 0 {
updateRowsAll = append(updateRowsAll, updateRows)
}
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
err = udc.batchRows(kvwriter, addRowsAll, updateRowsAll, deleteRowsAll)
if err == nil && backIndexRow == nil {
udc.m.Lock()
udc.docCount++
udc.m.Unlock()
}
atomic.AddUint64(&udc.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
udc.cf.Insert([]byte(doc.ID))
atomic.AddUint64(&udc.stats.updates, 1)
atomic.AddUint64(&udc.stats.numPlainTextBytesIndexed, numPlainTextBytes)
} else {
atomic.AddUint64(&udc.stats.errors, 1)
}
return
}
func (udc *UpsideDownCouch) Batch(batch *index.Batch) (err error) {
analysisStart := time.Now()
resultChan := make(chan *index.AnalysisResult, len(batch.IndexOps))
var numUpdates uint64
var numPlainTextBytes uint64
for _, doc := range batch.IndexOps {
if doc != nil {
numUpdates++
numPlainTextBytes += doc.NumPlainTextBytes()
}
}
go func() {
for _, doc := range batch.IndexOps {
if doc != nil {
aw := index.NewAnalysisWork(udc, doc, resultChan)
// put the work on the queue
udc.analysisQueue.Queue(aw)
}
}
}()
// retrieve back index rows concurrent with analysis
docBackIndexRowErr := error(nil)
docBackIndexRowCh := make(chan *docBackIndexRow, len(batch.IndexOps))
udc.writeMutex.Lock()
defer udc.writeMutex.Unlock()
go func() {
defer close(docBackIndexRowCh)
// open a reader for backindex lookup
var kvreader store.KVReader
kvreader, err = udc.store.Reader()
if err != nil {
docBackIndexRowErr = err
return
}
for docID, doc := range batch.IndexOps {
backIndexRow, err := backIndexRowForDoc(kvreader, index.IndexInternalID(docID))
if err != nil {
docBackIndexRowErr = err
return
}
docBackIndexRowCh <- &docBackIndexRow{docID, doc, backIndexRow}
}
err = kvreader.Close()
if err != nil {
docBackIndexRowErr = err
return
}
}()
// wait for analysis result
newRowsMap := make(map[string][]index.IndexRow)
var itemsDeQueued uint64
for itemsDeQueued < numUpdates {
result := <-resultChan
newRowsMap[result.DocID] = result.Rows
itemsDeQueued++
}
close(resultChan)
atomic.AddUint64(&udc.stats.analysisTime, uint64(time.Since(analysisStart)))
docsAdded := uint64(0)
docsDeleted := uint64(0)
indexStart := time.Now()
// prepare a list of rows
var addRowsAll [][]UpsideDownCouchRow
var updateRowsAll [][]UpsideDownCouchRow
var deleteRowsAll [][]UpsideDownCouchRow
// add the internal ops
var updateRows []UpsideDownCouchRow
var deleteRows []UpsideDownCouchRow
for internalKey, internalValue := range batch.InternalOps {
if internalValue == nil {
// delete
deleteInternalRow := NewInternalRow([]byte(internalKey), nil)
deleteRows = append(deleteRows, deleteInternalRow)
} else {
updateInternalRow := NewInternalRow([]byte(internalKey), internalValue)
updateRows = append(updateRows, updateInternalRow)
}
}
if len(updateRows) > 0 {
updateRowsAll = append(updateRowsAll, updateRows)
}
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
// process back index rows as they arrive
for dbir := range docBackIndexRowCh {
if dbir.doc == nil && dbir.backIndexRow != nil {
// delete
deleteRows := udc.deleteSingle(dbir.docID, dbir.backIndexRow, nil)
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
docsDeleted++
} else if dbir.doc != nil {
addRows, updateRows, deleteRows := udc.mergeOldAndNew(dbir.backIndexRow, newRowsMap[dbir.docID])
if len(addRows) > 0 {
addRowsAll = append(addRowsAll, addRows)
}
if len(updateRows) > 0 {
updateRowsAll = append(updateRowsAll, updateRows)
}
if len(deleteRows) > 0 {
deleteRowsAll = append(deleteRowsAll, deleteRows)
}
if dbir.backIndexRow == nil {
docsAdded++
}
}
}
if docBackIndexRowErr != nil {
return docBackIndexRowErr
}
// start a writer for this batch
var kvwriter store.KVWriter
kvwriter, err = udc.store.Writer()
if err != nil {
return
}
err = udc.batchRows(kvwriter, addRowsAll, updateRowsAll, deleteRowsAll)
if err != nil {
_ = kvwriter.Close()
atomic.AddUint64(&udc.stats.errors, 1)
return
}
err = kvwriter.Close()
atomic.AddUint64(&udc.stats.indexTime, uint64(time.Since(indexStart)))
if err == nil {
udc.m.Lock()
udc.docCount += docsAdded
udc.docCount -= docsDeleted
udc.m.Unlock()
atomic.AddUint64(&udc.stats.updates, numUpdates)
atomic.AddUint64(&udc.stats.deletes, docsDeleted)
atomic.AddUint64(&udc.stats.batches, 1)
atomic.AddUint64(&udc.stats.numPlainTextBytesIndexed, numPlainTextBytes)
} else {
atomic.AddUint64(&udc.stats.errors, 1)
}
return
}