Beispiel #1
0
func (f *Firestorm) Open() (err error) {

	// open the kv store
	storeConstructor := registry.KVStoreConstructorByName(f.storeName)
	if storeConstructor == nil {
		err = index.ErrorUnknownStorageType
		return
	}

	// now open the store
	f.store, err = storeConstructor(&mergeOperator, f.storeConfig)
	if err != nil {
		return
	}

	// start a reader
	var kvreader store.KVReader
	kvreader, err = f.store.Reader()
	if err != nil {
		return
	}

	// assert correct version, and find out if this is new index
	var newIndex bool
	newIndex, err = f.checkVersion(kvreader)
	if err != nil {
		return
	}

	if !newIndex {
		// process existing index before opening
		err = f.warmup(kvreader)
		if err != nil {
			return
		}
	}

	err = kvreader.Close()
	if err != nil {
		return
	}

	if newIndex {
		// prepare a new index
		err = f.bootstrap()
		if err != nil {
			return
		}
	}

	// start the garbage collector
	f.garbageCollector.Start()

	// start the lookuper
	f.lookuper.Start()

	// start the dict updater
	f.dictUpdater.Start()

	return
}
Beispiel #2
0
func (udc *SmolderingCouch) Open() (err error) {
	//acquire the write mutex for the duratin of Open()
	udc.writeMutex.Lock()
	defer udc.writeMutex.Unlock()

	// open the kv store
	storeConstructor := registry.KVStoreConstructorByName(udc.storeName)
	if storeConstructor == nil {
		err = index.ErrorUnknownStorageType
		return
	}

	// now open the store
	udc.store, err = storeConstructor(&mergeOperator, udc.storeConfig)
	if err != nil {
		return
	}

	udc.cf = cuckoofilter.NewDefaultCuckooFilter()

	// start a reader to look at the index
	var kvreader store.KVReader
	kvreader, err = udc.store.Reader()
	if err != nil {
		return
	}

	var value []byte
	value, err = kvreader.Get(VersionKey)
	if err != nil {
		_ = kvreader.Close()
		return
	}

	if value != nil {
		err = udc.loadSchema(kvreader)
		if err != nil {
			_ = kvreader.Close()
			return
		}

		// set doc count
		udc.m.Lock()
		udc.docCount, udc.maxInternalDocID, err = udc.countDocs(kvreader)
		udc.m.Unlock()

		err = kvreader.Close()
	} else {
		// new index, close the reader and open writer to init
		err = kvreader.Close()
		if err != nil {
			return
		}

		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
			}
		}()

		// init the index
		err = udc.init(kvwriter)
	}

	return
}
Beispiel #3
0
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
}
Beispiel #4
0
func (udc *UpsideDownCouch) Delete(id string) (err error) {
	indexStart := time.Now()

	udc.writeMutex.Lock()
	defer udc.writeMutex.Unlock()

	// open a reader for backindex lookup
	var kvreader store.KVReader
	kvreader, err = udc.store.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
	backIndexRow, err = backIndexRowForDoc(kvreader, index.IndexInternalID(id))
	if err != nil {
		_ = kvreader.Close()
		atomic.AddUint64(&udc.stats.errors, 1)
		return
	}

	err = kvreader.Close()
	if err != nil {
		return
	}

	if backIndexRow == nil {
		atomic.AddUint64(&udc.stats.deletes, 1)
		return
	}

	// start a writer for this delete
	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
		}
	}()

	var deleteRowsAll [][]UpsideDownCouchRow

	deleteRows := udc.deleteSingle(id, backIndexRow, nil)
	if len(deleteRows) > 0 {
		deleteRowsAll = append(deleteRowsAll, deleteRows)
	}

	err = udc.batchRows(kvwriter, nil, nil, deleteRowsAll)
	if err == 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.deletes, 1)
	} else {
		atomic.AddUint64(&udc.stats.errors, 1)
	}
	return
}
Beispiel #5
0
func (udc *UpsideDownCouch) Update(doc *document.Document) (err error) {
	// do analysis before acquiring write lock
	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()

	// open a reader for backindex lookup
	var kvreader store.KVReader
	kvreader, err = udc.store.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
	backIndexRow, err = backIndexRowForDoc(kvreader, index.IndexInternalID(doc.ID))
	if err != nil {
		_ = kvreader.Close()
		atomic.AddUint64(&udc.stats.errors, 1)
		return
	}

	err = kvreader.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 [][]UpsideDownCouchRow
	var updateRowsAll [][]UpsideDownCouchRow
	var deleteRowsAll [][]UpsideDownCouchRow

	addRows, updateRows, deleteRows := udc.mergeOldAndNew(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 {
		atomic.AddUint64(&udc.stats.updates, 1)
		atomic.AddUint64(&udc.stats.numPlainTextBytesIndexed, numPlainTextBytes)
	} else {
		atomic.AddUint64(&udc.stats.errors, 1)
	}
	return
}
Beispiel #6
0
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()

	udc.writeMutex.Lock()
	defer udc.writeMutex.Unlock()

	// open a reader for backindex lookup
	var kvreader store.KVReader
	kvreader, err = udc.store.Reader()
	if err != nil {
		return
	}

	// first lookup all the back index rows
	var backIndexRows map[string]*BackIndexRow
	backIndexRows, err = udc.backIndexRowsForBatch(kvreader, batch)
	if err != nil {
		_ = kvreader.Close()
		return
	}

	err = kvreader.Close()
	if err != nil {
		return
	}

	// start a writer for this batch
	var kvwriter store.KVWriter
	kvwriter, err = udc.store.Writer()
	if err != nil {
		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
}