public DocumentsWriterPerThread(string segmentName, Directory directory, LiveIndexWriterConfig indexWriterConfig, InfoStream infoStream, DocumentsWriterDeleteQueue deleteQueue, FieldInfos.Builder fieldInfos)
{
this.directoryOrig = directory;
this.directory = new TrackingDirectoryWrapper(directory);
this.fieldInfos = fieldInfos;
this.indexWriterConfig = indexWriterConfig;
this.infoStream = infoStream;
this.codec = indexWriterConfig.Codec;
this.docState = new DocState(this, infoStream);
this.docState.similarity = indexWriterConfig.Similarity;
bytesUsed = Counter.NewCounter();
byteBlockAllocator = new DirectTrackingAllocator(bytesUsed);
pendingUpdates = new BufferedUpdates();
intBlockAllocator = new Int32BlockAllocator(bytesUsed);
this.deleteQueue = deleteQueue;
Debug.Assert(numDocsInRAM == 0, "num docs " + numDocsInRAM);
pendingUpdates.Clear();
deleteSlice = deleteQueue.NewSlice();
segmentInfo = new SegmentInfo(directoryOrig, Constants.LUCENE_MAIN_VERSION, segmentName, -1, false, codec, null);
Debug.Assert(numDocsInRAM == 0);
if (INFO_VERBOSE && infoStream.IsEnabled("DWPT"))
{
infoStream.Message("DWPT", Thread.CurrentThread.Name + " init seg=" + segmentName + " delQueue=" + deleteQueue);
}
// this should be the last call in the ctor
// it really sucks that we need to pull this within the ctor and pass this ref to the chain!
consumer = indexWriterConfig.IndexingChain.GetChain(this);
}
private void EnsureInitialized(ThreadState state)
{
if (state.IsActive && state.dwpt == null)
{
FieldInfos.Builder infos = new FieldInfos.Builder(writer.globalFieldNumberMap);
state.dwpt = new DocumentsWriterPerThread(writer.NewSegmentName(), directory, config, infoStream, deleteQueue, infos);
}
}
/// <summary>
/// Call this to get the (merged) <see cref="FieldInfos"/> for a
/// composite reader.
/// <para/>
/// NOTE: the returned field numbers will likely not
/// correspond to the actual field numbers in the underlying
/// readers, and codec metadata (<see cref="FieldInfo.GetAttribute(string)"/>)
/// will be unavailable.
/// </summary>
public static FieldInfos GetMergedFieldInfos(IndexReader reader)
{
var builder = new FieldInfos.Builder();
foreach (AtomicReaderContext ctx in reader.Leaves)
{
builder.Add(ctx.AtomicReader.FieldInfos);
}
return(builder.Finish());
}
// note, just like in codec apis Directory 'dir' is NOT the same as segmentInfo.dir!!
internal SegmentMerger(IList <AtomicReader> readers, SegmentInfo segmentInfo, InfoStream infoStream, Directory dir, int termIndexInterval, CheckAbort checkAbort, FieldInfos.FieldNumbers fieldNumbers, IOContext context, bool validate)
{
// validate incoming readers
if (validate)
{
foreach (AtomicReader reader in readers)
{
reader.CheckIntegrity();
}
}
mergeState = new MergeState(readers, segmentInfo, infoStream, checkAbort);
directory = dir;
this.termIndexInterval = termIndexInterval;
this.codec = segmentInfo.Codec;
this.context = context;
this.fieldInfosBuilder = new FieldInfos.Builder(fieldNumbers);
mergeState.SegmentInfo.DocCount = SetDocMaps();
}
public virtual void WriteFieldUpdates(Directory dir, DocValuesFieldUpdates.Container dvUpdates)
{
UninterruptableMonitor.Enter(this);
try
{
if (Debugging.AssertsEnabled)
{
Debugging.Assert(UninterruptableMonitor.IsEntered(writer));
}
//System.out.println("rld.writeFieldUpdates: seg=" + info + " numericFieldUpdates=" + numericFieldUpdates);
if (Debugging.AssertsEnabled)
{
Debugging.Assert(dvUpdates.Any());
}
// Do this so we can delete any created files on
// exception; this saves all codecs from having to do
// it:
TrackingDirectoryWrapper trackingDir = new TrackingDirectoryWrapper(dir);
FieldInfos fieldInfos = null;
bool success = false;
try
{
Codec codec = Info.Info.Codec;
// reader could be null e.g. for a just merged segment (from
// IndexWriter.commitMergedDeletes).
SegmentReader reader = this.reader ?? new SegmentReader(Info, writer.Config.ReaderTermsIndexDivisor, IOContext.READ_ONCE);
try
{
// clone FieldInfos so that we can update their dvGen separately from
// the reader's infos and write them to a new fieldInfos_gen file
FieldInfos.Builder builder = new FieldInfos.Builder(writer.globalFieldNumberMap);
// cannot use builder.add(reader.getFieldInfos()) because it does not
// clone FI.attributes as well FI.dvGen
foreach (FieldInfo fi in reader.FieldInfos)
{
FieldInfo clone = builder.Add(fi);
// copy the stuff FieldInfos.Builder doesn't copy
if (fi.Attributes != null)
{
foreach (KeyValuePair <string, string> e in fi.Attributes)
{
clone.PutAttribute(e.Key, e.Value);
}
}
clone.DocValuesGen = fi.DocValuesGen;
}
// create new fields or update existing ones to have NumericDV type
foreach (string f in dvUpdates.numericDVUpdates.Keys)
{
builder.AddOrUpdate(f, NumericDocValuesField.TYPE);
}
// create new fields or update existing ones to have BinaryDV type
foreach (string f in dvUpdates.binaryDVUpdates.Keys)
{
builder.AddOrUpdate(f, BinaryDocValuesField.TYPE);
}
fieldInfos = builder.Finish();
long nextFieldInfosGen = Info.NextFieldInfosGen;
// LUCENENET specific: We created the segments names wrong in 4.8.0-beta00001 - 4.8.0-beta00015,
// so we added a switch to be able to read these indexes in later versions. This logic as well as an
// optimization on the first 100 segment values is implmeneted in SegmentInfos.SegmentNumberToString().
string segmentSuffix = SegmentInfos.SegmentNumberToString(nextFieldInfosGen);
SegmentWriteState state = new SegmentWriteState(null, trackingDir, Info.Info, fieldInfos, writer.Config.TermIndexInterval, null, IOContext.DEFAULT, segmentSuffix);
DocValuesFormat docValuesFormat = codec.DocValuesFormat;
DocValuesConsumer fieldsConsumer = docValuesFormat.FieldsConsumer(state);
bool fieldsConsumerSuccess = false;
try
{
// System.out.println("[" + Thread.currentThread().getName() + "] RLD.writeFieldUpdates: applying numeric updates; seg=" + info + " updates=" + numericFieldUpdates);
foreach (KeyValuePair <string, NumericDocValuesFieldUpdates> e in dvUpdates.numericDVUpdates)
{
string field = e.Key;
NumericDocValuesFieldUpdates fieldUpdates = e.Value;
FieldInfo fieldInfo = fieldInfos.FieldInfo(field);
if (Debugging.AssertsEnabled)
{
Debugging.Assert(fieldInfo != null);
}
fieldInfo.DocValuesGen = nextFieldInfosGen;
// write the numeric updates to a new gen'd docvalues file
fieldsConsumer.AddNumericField(fieldInfo, GetInt64Enumerable(reader, field, fieldUpdates));
}
// System.out.println("[" + Thread.currentThread().getName() + "] RAU.writeFieldUpdates: applying binary updates; seg=" + info + " updates=" + dvUpdates.binaryDVUpdates);
foreach (KeyValuePair <string, BinaryDocValuesFieldUpdates> e in dvUpdates.binaryDVUpdates)
{
string field = e.Key;
BinaryDocValuesFieldUpdates dvFieldUpdates = e.Value;
FieldInfo fieldInfo = fieldInfos.FieldInfo(field);
if (Debugging.AssertsEnabled)
{
Debugging.Assert(fieldInfo != null);
}
// System.out.println("[" + Thread.currentThread().getName() + "] RAU.writeFieldUpdates: applying binary updates; seg=" + info + " f=" + dvFieldUpdates + ", updates=" + dvFieldUpdates);
fieldInfo.DocValuesGen = nextFieldInfosGen;
// write the numeric updates to a new gen'd docvalues file
fieldsConsumer.AddBinaryField(fieldInfo, GetBytesRefEnumerable(reader, field, dvFieldUpdates));
}
codec.FieldInfosFormat.FieldInfosWriter.Write(trackingDir, Info.Info.Name, segmentSuffix, fieldInfos, IOContext.DEFAULT);
fieldsConsumerSuccess = true;
}
finally
{
if (fieldsConsumerSuccess)
{
fieldsConsumer.Dispose();
}
else
{
IOUtils.DisposeWhileHandlingException(fieldsConsumer);
}
}
}
finally
{
if (reader != this.reader)
{
// System.out.println("[" + Thread.currentThread().getName() + "] RLD.writeLiveDocs: closeReader " + reader);
reader.Dispose();
}
}
success = true;
}
finally
{
if (!success)
{
// Advance only the nextWriteDocValuesGen so that a 2nd
// attempt to write will write to a new file
Info.AdvanceNextWriteFieldInfosGen();
// Delete any partially created file(s):
foreach (string fileName in trackingDir.CreatedFiles)
{
try
{
dir.DeleteFile(fileName);
}
catch (Exception t) when(t.IsThrowable())
{
// Ignore so we throw only the first exc
}
}
}
}
Info.AdvanceFieldInfosGen();
// copy all the updates to mergingUpdates, so they can later be applied to the merged segment
if (isMerging)
{
foreach (KeyValuePair <string, NumericDocValuesFieldUpdates> e in dvUpdates.numericDVUpdates)
{
if (!mergingDVUpdates.TryGetValue(e.Key, out DocValuesFieldUpdates updates))
{
mergingDVUpdates[e.Key] = e.Value;
}
else
{
updates.Merge(e.Value);
}
}
foreach (KeyValuePair <string, BinaryDocValuesFieldUpdates> e in dvUpdates.binaryDVUpdates)
{
if (!mergingDVUpdates.TryGetValue(e.Key, out DocValuesFieldUpdates updates))
{
mergingDVUpdates[e.Key] = e.Value;
}
else
{
updates.Merge(e.Value);
}
}
}
// create a new map, keeping only the gens that are in use
IDictionary <long, ISet <string> > genUpdatesFiles = Info.UpdatesFiles;
IDictionary <long, ISet <string> > newGenUpdatesFiles = new Dictionary <long, ISet <string> >();
long fieldInfosGen = Info.FieldInfosGen;
foreach (FieldInfo fi in fieldInfos)
{
long dvGen = fi.DocValuesGen;
if (dvGen != -1 && !newGenUpdatesFiles.ContainsKey(dvGen))
{
if (dvGen == fieldInfosGen)
{
newGenUpdatesFiles[fieldInfosGen] = trackingDir.CreatedFiles;
}
else
{
newGenUpdatesFiles[dvGen] = genUpdatesFiles[dvGen];
}
}
}
Info.SetGenUpdatesFiles(newGenUpdatesFiles);
// wrote new files, should checkpoint()
writer.Checkpoint();
// if there is a reader open, reopen it to reflect the updates
if (reader != null)
{
SegmentReader newReader = new SegmentReader(Info, reader, liveDocs, Info.Info.DocCount - Info.DelCount - pendingDeleteCount);
bool reopened = false;
try
{
reader.DecRef();
reader = newReader;
reopened = true;
}
finally
{
if (!reopened)
{
newReader.DecRef();
}
}
}
}
finally
{
UninterruptableMonitor.Exit(this);
}
}
/// <summary>
/// Expert: create a <see cref="ParallelAtomicReader"/> based on the provided
/// <paramref name="readers"/> and <paramref name="storedFieldsReaders"/>; when a document is
/// loaded, only <paramref name="storedFieldsReaders"/> will be used.
/// </summary>
public ParallelAtomicReader(bool closeSubReaders, AtomicReader[] readers, AtomicReader[] storedFieldsReaders)
{
this.closeSubReaders = closeSubReaders;
if (readers.Length == 0 && storedFieldsReaders.Length > 0)
{
throw new ArgumentException("There must be at least one main reader if storedFieldsReaders are used.");
}
this.parallelReaders = (AtomicReader[])readers.Clone();
this.storedFieldsReaders = (AtomicReader[])storedFieldsReaders.Clone();
if (parallelReaders.Length > 0)
{
AtomicReader first = parallelReaders[0];
this.maxDoc = first.MaxDoc;
this.numDocs = first.NumDocs;
this.hasDeletions = first.HasDeletions;
}
else
{
this.maxDoc = this.numDocs = 0;
this.hasDeletions = false;
}
completeReaderSet.UnionWith(this.parallelReaders);
completeReaderSet.UnionWith(this.storedFieldsReaders);
// check compatibility:
foreach (AtomicReader reader in completeReaderSet)
{
if (reader.MaxDoc != maxDoc)
{
throw new ArgumentException("All readers must have same MaxDoc: " + maxDoc + "!=" + reader.MaxDoc);
}
}
// TODO: make this read-only in a cleaner way?
FieldInfos.Builder builder = new FieldInfos.Builder();
// build FieldInfos and fieldToReader map:
foreach (AtomicReader reader in this.parallelReaders)
{
FieldInfos readerFieldInfos = reader.FieldInfos;
foreach (FieldInfo fieldInfo in readerFieldInfos)
{
// NOTE: first reader having a given field "wins":
if (!fieldToReader.ContainsKey(fieldInfo.Name))
{
builder.Add(fieldInfo);
fieldToReader[fieldInfo.Name] = reader;
if (fieldInfo.HasVectors)
{
tvFieldToReader[fieldInfo.Name] = reader;
}
}
}
}
fieldInfos = builder.Finish();
// build Fields instance
foreach (AtomicReader reader in this.parallelReaders)
{
Fields readerFields = reader.Fields;
if (readerFields != null)
{
foreach (string field in readerFields)
{
// only add if the reader responsible for that field name is the current:
if (fieldToReader[field].Equals(reader))
{
this.fields.AddField(field, readerFields.GetTerms(field));
}
}
}
}
// do this finally so any Exceptions occurred before don't affect refcounts:
foreach (AtomicReader reader in completeReaderSet)
{
if (!closeSubReaders)
{
reader.IncRef();
}
reader.RegisterParentReader(this);
}
}
public abstract void ProcessDocument(FieldInfos.Builder fieldInfos);
public override void ProcessDocument(FieldInfos.Builder fieldInfos)
{
consumer.StartDocument();
storedConsumer.StartDocument();
fieldCount = 0;
int thisFieldGen = fieldGen++;
// Absorb any new fields first seen in this document.
// Also absorb any changes to fields we had already
// seen before (eg suddenly turning on norms or
// vectors, etc.):
foreach (IIndexableField field in docState.doc)
{
string fieldName = field.Name;
// Make sure we have a PerField allocated
int hashPos = fieldName.GetHashCode() & hashMask;
DocFieldProcessorPerField fp = fieldHash[hashPos];
while (fp != null && !fp.fieldInfo.Name.Equals(fieldName, StringComparison.Ordinal))
{
fp = fp.next;
}
if (fp == null)
{
// TODO FI: we need to genericize the "flags" that a
// field holds, and, how these flags are merged; it
// needs to be more "pluggable" such that if I want
// to have a new "thing" my Fields can do, I can
// easily add it
FieldInfo fi = fieldInfos.AddOrUpdate(fieldName, field.IndexableFieldType);
fp = new DocFieldProcessorPerField(this, fi);
fp.next = fieldHash[hashPos];
fieldHash[hashPos] = fp;
totalFieldCount++;
if (totalFieldCount >= fieldHash.Length / 2)
{
Rehash();
}
}
else
{
// need to addOrUpdate so that FieldInfos can update globalFieldNumbers
// with the correct DocValue type (LUCENE-5192)
FieldInfo fi = fieldInfos.AddOrUpdate(fieldName, field.IndexableFieldType);
Debug.Assert(fi == fp.fieldInfo, "should only have updated an existing FieldInfo instance");
}
if (thisFieldGen != fp.lastGen)
{
// First time we're seeing this field for this doc
fp.fieldCount = 0;
if (fieldCount == fields.Length)
{
int newSize = fields.Length * 2;
DocFieldProcessorPerField[] newArray = new DocFieldProcessorPerField[newSize];
Array.Copy(fields, 0, newArray, 0, fieldCount);
fields = newArray;
}
fields[fieldCount++] = fp;
fp.lastGen = thisFieldGen;
}
fp.AddField(field);
storedConsumer.AddField(docState.docID, field, fp.fieldInfo);
}
// If we are writing vectors then we must visit
// fields in sorted order so they are written in
// sorted order. TODO: we actually only need to
// sort the subset of fields that have vectors
// enabled; we could save [small amount of] CPU
// here.
ArrayUtil.IntroSort(fields, 0, fieldCount, fieldsComp);
for (int i = 0; i < fieldCount; i++)
{
DocFieldProcessorPerField perField = fields[i];
perField.consumer.ProcessFields(perField.fields, perField.fieldCount);
}
}
