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 IntBlockAllocator(bytesUsed);
this.DeleteQueue = deleteQueue;
Debug.Assert(numDocsInRAM == 0, "num docs " + numDocsInRAM);
PendingUpdates.Clear();
DeleteSlice = deleteQueue.NewSlice();
SegmentInfo_Renamed = 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);
}
/// <summary>
/// Does the actual (time-consuming) work of the merge,
/// but without holding synchronized lock on IndexWriter
/// instance
/// </summary>
private int MergeMiddle(MergePolicy.OneMerge merge)
{
merge.CheckAborted(directory);
string mergedName = merge.info.Info.Name;
IList<SegmentCommitInfo> sourceSegments = merge.Segments;
IOContext context = new IOContext(merge.MergeInfo);
MergeState.CheckAbort checkAbort = new MergeState.CheckAbort(merge, directory);
TrackingDirectoryWrapper dirWrapper = new TrackingDirectoryWrapper(directory);
if (infoStream.IsEnabled("IW"))
{
infoStream.Message("IW", "merging " + SegString(merge.Segments));
}
merge.Readers = new List<SegmentReader>();
// this is try/finally to make sure merger's readers are
// closed:
bool success = false;
try
{
int segUpto = 0;
while (segUpto < sourceSegments.Count)
{
SegmentCommitInfo info = sourceSegments[segUpto];
// Hold onto the "live" reader; we will use this to
// commit merged deletes
ReadersAndUpdates rld = readerPool.Get(info, true);
// Carefully pull the most recent live docs and reader
SegmentReader reader;
Bits liveDocs;
int delCount;
lock (this)
{
// Must sync to ensure BufferedDeletesStream cannot change liveDocs,
// pendingDeleteCount and field updates while we pull a copy:
reader = rld.GetReaderForMerge(context);
liveDocs = rld.ReadOnlyLiveDocs;
delCount = rld.PendingDeleteCount + info.DelCount;
Debug.Assert(reader != null);
Debug.Assert(rld.VerifyDocCounts());
if (infoStream.IsEnabled("IW"))
{
if (rld.PendingDeleteCount != 0)
{
infoStream.Message("IW", "seg=" + SegString(info) + " delCount=" + info.DelCount + " pendingDelCount=" + rld.PendingDeleteCount);
}
else if (info.DelCount != 0)
{
infoStream.Message("IW", "seg=" + SegString(info) + " delCount=" + info.DelCount);
}
else
{
infoStream.Message("IW", "seg=" + SegString(info) + " no deletes");
}
}
}
// Deletes might have happened after we pulled the merge reader and
// before we got a read-only copy of the segment's actual live docs
// (taking pending deletes into account). In that case we need to
// make a new reader with updated live docs and del count.
if (reader.NumDeletedDocs != delCount)
{
// fix the reader's live docs and del count
Debug.Assert(delCount > reader.NumDeletedDocs); // beware of zombies
SegmentReader newReader = new SegmentReader(info, reader, liveDocs, info.Info.DocCount - delCount);
bool released = false;
try
{
rld.Release(reader);
released = true;
}
finally
{
if (!released)
{
newReader.DecRef();
}
}
reader = newReader;
}
merge.Readers.Add(reader);
Debug.Assert(delCount <= info.Info.DocCount, "delCount=" + delCount + " info.docCount=" + info.Info.DocCount + " rld.pendingDeleteCount=" + rld.PendingDeleteCount + " info.getDelCount()=" + info.DelCount);
segUpto++;
}
// System.out.println("[" + Thread.currentThread().getName() + "] IW.mergeMiddle: merging " + merge.getMergeReaders());
// we pass merge.getMergeReaders() instead of merge.readers to allow the
// OneMerge to return a view over the actual segments to merge
SegmentMerger merger = new SegmentMerger(merge.MergeReaders, merge.info.Info, infoStream, dirWrapper, Config_Renamed.TermIndexInterval, checkAbort, GlobalFieldNumberMap, context, Config_Renamed.CheckIntegrityAtMerge);
merge.CheckAborted(directory);
// this is where all the work happens:
MergeState mergeState;
bool success3 = false;
try
{
if (!merger.ShouldMerge())
{
// would result in a 0 document segment: nothing to merge!
mergeState = new MergeState(new List<AtomicReader>(), merge.info.Info, infoStream, checkAbort);
}
else
{
mergeState = merger.Merge();
}
success3 = true;
}
finally
{
if (!success3)
{
lock (this)
{
Deleter.Refresh(merge.info.Info.Name);
}
}
}
Debug.Assert(mergeState.SegmentInfo == merge.info.Info);
merge.info.Info.Files = new HashSet<string>(dirWrapper.CreatedFiles);
// Record which codec was used to write the segment
if (infoStream.IsEnabled("IW"))
{
if (merge.info.Info.DocCount == 0)
{
infoStream.Message("IW", "merge away fully deleted segments");
}
else
{
infoStream.Message("IW", "merge codec=" + Codec + " docCount=" + merge.info.Info.DocCount + "; merged segment has " + (mergeState.FieldInfos.HasVectors() ? "vectors" : "no vectors") + "; " + (mergeState.FieldInfos.HasNorms() ? "norms" : "no norms") + "; " + (mergeState.FieldInfos.HasDocValues() ? "docValues" : "no docValues") + "; " + (mergeState.FieldInfos.HasProx() ? "prox" : "no prox") + "; " + (mergeState.FieldInfos.HasProx() ? "freqs" : "no freqs"));
}
}
// Very important to do this before opening the reader
// because codec must know if prox was written for
// this segment:
//System.out.println("merger set hasProx=" + merger.hasProx() + " seg=" + merge.info.name);
bool useCompoundFile;
lock (this) // Guard segmentInfos
{
useCompoundFile = mergePolicy.UseCompoundFile(segmentInfos, merge.info);
}
if (useCompoundFile)
{
success = false;
ICollection<string> filesToRemove = merge.info.Files();
try
{
filesToRemove = CreateCompoundFile(infoStream, directory, checkAbort, merge.info.Info, context);
success = true;
}
catch (System.IO.IOException ioe)
{
lock (this)
{
if (merge.Aborted)
{
// this can happen if rollback or close(false)
// is called -- fall through to logic below to
// remove the partially created CFS:
}
else
{
HandleMergeException(ioe, merge);
}
}
}
catch (Exception t)
{
HandleMergeException(t, merge);
}
finally
{
if (!success)
{
if (infoStream.IsEnabled("IW"))
{
infoStream.Message("IW", "hit exception creating compound file during merge");
}
lock (this)
{
Deleter.DeleteFile(Lucene.Net.Index.IndexFileNames.SegmentFileName(mergedName, "", Lucene.Net.Index.IndexFileNames.COMPOUND_FILE_EXTENSION));
Deleter.DeleteFile(Lucene.Net.Index.IndexFileNames.SegmentFileName(mergedName, "", Lucene.Net.Index.IndexFileNames.COMPOUND_FILE_ENTRIES_EXTENSION));
Deleter.DeleteNewFiles(merge.info.Files());
}
}
}
// So that, if we hit exc in deleteNewFiles (next)
// or in commitMerge (later), we close the
// per-segment readers in the finally clause below:
success = false;
lock (this)
{
// delete new non cfs files directly: they were never
// registered with IFD
Deleter.DeleteNewFiles(filesToRemove);
if (merge.Aborted)
{
if (infoStream.IsEnabled("IW"))
{
infoStream.Message("IW", "abort merge after building CFS");
}
Deleter.DeleteFile(Lucene.Net.Index.IndexFileNames.SegmentFileName(mergedName, "", Lucene.Net.Index.IndexFileNames.COMPOUND_FILE_EXTENSION));
Deleter.DeleteFile(Lucene.Net.Index.IndexFileNames.SegmentFileName(mergedName, "", Lucene.Net.Index.IndexFileNames.COMPOUND_FILE_ENTRIES_EXTENSION));
return 0;
}
}
merge.info.Info.UseCompoundFile = true;
}
else
{
// So that, if we hit exc in commitMerge (later),
// we close the per-segment readers in the finally
// clause below:
success = false;
}
// Have codec write SegmentInfo. Must do this after
// creating CFS so that 1) .si isn't slurped into CFS,
// and 2) .si reflects useCompoundFile=true change
// above:
bool success2 = false;
try
{
Codec.SegmentInfoFormat().SegmentInfoWriter.Write(directory, merge.info.Info, mergeState.FieldInfos, context);
success2 = true;
}
finally
{
if (!success2)
{
lock (this)
{
Deleter.DeleteNewFiles(merge.info.Files());
}
}
}
// TODO: ideally we would freeze merge.info here!!
// because any changes after writing the .si will be
// lost...
if (infoStream.IsEnabled("IW"))
{
infoStream.Message("IW", string.Format(CultureInfo.InvariantCulture, "merged segment size=%.3f MB vs estimate=%.3f MB", merge.info.SizeInBytes() / 1024.0 / 1024.0, merge.EstimatedMergeBytes / 1024 / 1024.0));
}
IndexReaderWarmer mergedSegmentWarmer = Config_Renamed.MergedSegmentWarmer;
if (PoolReaders && mergedSegmentWarmer != null && merge.info.Info.DocCount != 0)
{
ReadersAndUpdates rld = readerPool.Get(merge.info, true);
SegmentReader sr = rld.GetReader(IOContext.READ);
try
{
mergedSegmentWarmer.Warm(sr);
}
finally
{
lock (this)
{
rld.Release(sr);
readerPool.Release(rld);
}
}
}
// Force READ context because we merge deletes onto
// this reader:
if (!CommitMerge(merge, mergeState))
{
// commitMerge will return false if this merge was
// aborted
return 0;
}
success = true;
}
finally
{
// Readers are already closed in commitMerge if we didn't hit
// an exc:
if (!success)
{
CloseMergeReaders(merge, true);
}
}
return merge.info.Info.DocCount;
}
/// <summary>
/// Merges the provided indexes into this index.
///
/// <p>
/// The provided IndexReaders are not closed.
///
/// <p>
/// See <seealso cref="#addIndexes"/> for details on transactional semantics, temporary
/// free space required in the Directory, and non-CFS segments on an Exception.
///
/// <p>
/// <b>NOTE</b>: if this method hits an OutOfMemoryError you should immediately
/// close the writer. See <a href="#OOME">above</a> for details.
///
/// <p>
/// <b>NOTE:</b> empty segments are dropped by this method and not added to this
/// index.
///
/// <p>
/// <b>NOTE:</b> this method merges all given <seealso cref="IndexReader"/>s in one
/// merge. If you intend to merge a large number of readers, it may be better
/// to call this method multiple times, each time with a small set of readers.
/// In principle, if you use a merge policy with a {@code mergeFactor} or
/// {@code maxMergeAtOnce} parameter, you should pass that many readers in one
/// call. Also, if the given readers are <seealso cref="DirectoryReader"/>s, they can be
/// opened with {@code termIndexInterval=-1} to save RAM, since during merge
/// the in-memory structure is not used. See
/// <seealso cref="DirectoryReader#open(Directory, int)"/>.
///
/// <p>
/// <b>NOTE</b>: if you call <seealso cref="#close(boolean)"/> with <tt>false</tt>, which
/// aborts all running merges, then any thread still running this method might
/// hit a <seealso cref="MergePolicy.MergeAbortedException"/>.
/// </summary>
/// <exception cref="CorruptIndexException">
/// if the index is corrupt </exception>
/// <exception cref="IOException">
/// if there is a low-level IO error </exception>
public virtual void AddIndexes(params IndexReader[] readers)
{
EnsureOpen();
int numDocs = 0;
try
{
if (infoStream.IsEnabled("IW"))
{
infoStream.Message("IW", "flush at addIndexes(IndexReader...)");
}
Flush(false, true);
string mergedName = NewSegmentName();
IList<AtomicReader> mergeReaders = new List<AtomicReader>();
foreach (IndexReader indexReader in readers)
{
numDocs += indexReader.NumDocs;
foreach (AtomicReaderContext ctx in indexReader.Leaves)
{
mergeReaders.Add(ctx.AtomicReader);
}
}
IOContext context = new IOContext(new MergeInfo(numDocs, -1, true, -1));
// TODO: somehow we should fix this merge so it's
// abortable so that IW.close(false) is able to stop it
TrackingDirectoryWrapper trackingDir = new TrackingDirectoryWrapper(directory);
SegmentInfo info = new SegmentInfo(directory, Constants.LUCENE_MAIN_VERSION, mergedName, -1, false, Codec, null);
SegmentMerger merger = new SegmentMerger(mergeReaders, info, infoStream, trackingDir, Config_Renamed.TermIndexInterval, MergeState.CheckAbort.NONE, GlobalFieldNumberMap, context, Config_Renamed.CheckIntegrityAtMerge);
if (!merger.ShouldMerge())
{
return;
}
MergeState mergeState;
bool success = false;
try
{
mergeState = merger.Merge(); // merge 'em
success = true;
}
finally
{
if (!success)
{
lock (this)
{
Deleter.Refresh(info.Name);
}
}
}
SegmentCommitInfo infoPerCommit = new SegmentCommitInfo(info, 0, -1L, -1L);
info.Files = new HashSet<string>(trackingDir.CreatedFiles);
trackingDir.CreatedFiles.Clear();
SetDiagnostics(info, SOURCE_ADDINDEXES_READERS);
bool useCompoundFile;
lock (this) // Guard segmentInfos
{
if (StopMerges)
{
Deleter.DeleteNewFiles(infoPerCommit.Files());
return;
}
EnsureOpen();
useCompoundFile = mergePolicy.UseCompoundFile(segmentInfos, infoPerCommit);
}
// Now create the compound file if needed
if (useCompoundFile)
{
ICollection<string> filesToDelete = infoPerCommit.Files();
try
{
CreateCompoundFile(infoStream, directory, MergeState.CheckAbort.NONE, info, context);
}
finally
{
// delete new non cfs files directly: they were never
// registered with IFD
lock (this)
{
Deleter.DeleteNewFiles(filesToDelete);
}
}
info.UseCompoundFile = true;
}
// Have codec write SegmentInfo. Must do this after
// creating CFS so that 1) .si isn't slurped into CFS,
// and 2) .si reflects useCompoundFile=true change
// above:
success = false;
try
{
Codec.SegmentInfoFormat().SegmentInfoWriter.Write(trackingDir, info, mergeState.FieldInfos, context);
success = true;
}
finally
{
if (!success)
{
lock (this)
{
Deleter.Refresh(info.Name);
}
}
}
info.AddFiles(trackingDir.CreatedFiles);
// Register the new segment
lock (this)
{
if (StopMerges)
{
Deleter.DeleteNewFiles(info.Files);
return;
}
EnsureOpen();
segmentInfos.Add(infoPerCommit);
Checkpoint();
}
}
catch (System.OutOfMemoryException oom)
{
HandleOOM(oom, "addIndexes(IndexReader...)");
}
}
/// <summary>
/// Copies the segment files as-is into the IndexWriter's directory. </summary>
private SegmentCommitInfo CopySegmentAsIs(SegmentCommitInfo info, string segName, IDictionary<string, string> dsNames, ISet<string> dsFilesCopied, IOContext context, ISet<string> copiedFiles)
{
// Determine if the doc store of this segment needs to be copied. It's
// only relevant for segments that share doc store with others,
// because the DS might have been copied already, in which case we
// just want to update the DS name of this SegmentInfo.
string dsName = Lucene3xSegmentInfoFormat.GetDocStoreSegment(info.Info);
Debug.Assert(dsName != null);
string newDsName;
if (dsNames.ContainsKey(dsName))
{
newDsName = dsNames[dsName];
}
else
{
dsNames[dsName] = segName;
newDsName = segName;
}
// note: we don't really need this fis (its copied), but we load it up
// so we don't pass a null value to the si writer
FieldInfos fis = SegmentReader.ReadFieldInfos(info);
ISet<string> docStoreFiles3xOnly = Lucene3xCodec.GetDocStoreFiles(info.Info);
IDictionary<string, string> attributes;
// copy the attributes map, we might modify it below.
// also we need to ensure its read-write, since we will invoke the SIwriter (which might want to set something).
if (info.Info.Attributes() == null)
{
attributes = new Dictionary<string, string>();
}
else
{
attributes = new Dictionary<string, string>(info.Info.Attributes());
}
if (docStoreFiles3xOnly != null)
{
// only violate the codec this way if it's preflex &
// shares doc stores
// change docStoreSegment to newDsName
attributes[Lucene3xSegmentInfoFormat.DS_NAME_KEY] = newDsName;
}
//System.out.println("copy seg=" + info.info.name + " version=" + info.info.getVersion());
// Same SI as before but we change directory, name and docStoreSegment:
SegmentInfo newInfo = new SegmentInfo(directory, info.Info.Version, segName, info.Info.DocCount, info.Info.UseCompoundFile, info.Info.Codec, info.Info.Diagnostics, attributes);
SegmentCommitInfo newInfoPerCommit = new SegmentCommitInfo(newInfo, info.DelCount, info.DelGen, info.FieldInfosGen);
HashSet<string> segFiles = new HashSet<string>();
// Build up new segment's file names. Must do this
// before writing SegmentInfo:
foreach (string file in info.Files())
{
string newFileName;
if (docStoreFiles3xOnly != null && docStoreFiles3xOnly.Contains(file))
{
newFileName = newDsName + Lucene.Net.Index.IndexFileNames.StripSegmentName(file);
}
else
{
newFileName = segName + Lucene.Net.Index.IndexFileNames.StripSegmentName(file);
}
segFiles.Add(newFileName);
}
newInfo.Files = segFiles;
// We must rewrite the SI file because it references
// segment name (its own name, if its 3.x, and doc
// store segment name):
TrackingDirectoryWrapper trackingDir = new TrackingDirectoryWrapper(directory);
Codec currentCodec = newInfo.Codec;
try
{
currentCodec.SegmentInfoFormat().SegmentInfoWriter.Write(trackingDir, newInfo, fis, context);
}
catch (System.NotSupportedException uoe)
{
if (currentCodec is Lucene3xCodec)
{
// OK: 3x codec cannot write a new SI file;
// SegmentInfos will write this on commit
}
else
{
throw uoe;
}
}
ICollection<string> siFiles = trackingDir.CreatedFiles;
bool success = false;
try
{
// Copy the segment's files
foreach (string file in info.Files())
{
string newFileName;
if (docStoreFiles3xOnly != null && docStoreFiles3xOnly.Contains(file))
{
newFileName = newDsName + Lucene.Net.Index.IndexFileNames.StripSegmentName(file);
if (dsFilesCopied.Contains(newFileName))
{
continue;
}
dsFilesCopied.Add(newFileName);
}
else
{
newFileName = segName + Lucene.Net.Index.IndexFileNames.StripSegmentName(file);
}
if (siFiles.Contains(newFileName))
{
// We already rewrote this above
continue;
}
Debug.Assert(!SlowFileExists(directory, newFileName), "file \"" + newFileName + "\" already exists; siFiles=" + siFiles);
Debug.Assert(!copiedFiles.Contains(file), "file \"" + file + "\" is being copied more than once");
copiedFiles.Add(file);
info.Info.Dir.Copy(directory, file, newFileName, context);
}
success = true;
}
finally
{
if (!success)
{
foreach (string file in newInfo.Files)
{
try
{
directory.DeleteFile(file);
}
catch (Exception)
{
}
}
}
}
return newInfoPerCommit;
}
private SegmentCommitInfo Merge(Directory dir, SegmentCommitInfo si1, SegmentCommitInfo si2, string merged, bool useCompoundFile)
{
IOContext context = NewIOContext(Random());
SegmentReader r1 = new SegmentReader(si1, DirectoryReader.DEFAULT_TERMS_INDEX_DIVISOR, context);
SegmentReader r2 = new SegmentReader(si2, DirectoryReader.DEFAULT_TERMS_INDEX_DIVISOR, context);
Codec codec = Codec.Default;
TrackingDirectoryWrapper trackingDir = new TrackingDirectoryWrapper(si1.Info.Dir);
SegmentInfo si = new SegmentInfo(si1.Info.Dir, Constants.LUCENE_MAIN_VERSION, merged, -1, false, codec, null);
SegmentMerger merger = new SegmentMerger(Arrays.AsList<AtomicReader>(r1, r2), si, InfoStream.Default, trackingDir, IndexWriterConfig.DEFAULT_TERM_INDEX_INTERVAL, MergeState.CheckAbort.NONE, new FieldInfos.FieldNumbers(), context, true);
MergeState mergeState = merger.Merge();
r1.Dispose();
r2.Dispose();
SegmentInfo info = new SegmentInfo(si1.Info.Dir, Constants.LUCENE_MAIN_VERSION, merged, si1.Info.DocCount + si2.Info.DocCount, false, codec, null);
info.Files = new HashSet<string>(trackingDir.CreatedFiles);
if (useCompoundFile)
{
ICollection<string> filesToDelete = IndexWriter.CreateCompoundFile(InfoStream.Default, dir, MergeState.CheckAbort.NONE, info, NewIOContext(Random()));
info.UseCompoundFile = true;
foreach (String fileToDelete in filesToDelete)
{
si1.Info.Dir.DeleteFile(fileToDelete);
}
}
return new SegmentCommitInfo(info, 0, -1L, -1L);
}
