// Delete by query
private static long applyQueryDeletes(
Iterable<QueryAndLimit> queriesIter, ReadersAndLiveDocs rld, final SegmentReader reader)
throws IOException {
long delCount = 0;
final AtomicReaderContext readerContext = reader.getContext();
boolean any = false;
for (QueryAndLimit ent : queriesIter) {
Query query = ent.query;
int limit = ent.limit;
final DocIdSet docs =
new QueryWrapperFilter(query).getDocIdSet(readerContext, reader.getLiveDocs());
if (docs != null) {
final DocIdSetIterator it = docs.iterator();
if (it != null) {
while (true) {
int doc = it.nextDoc();
if (doc >= limit) {
break;
}
if (!any) {
rld.initWritableLiveDocs();
any = true;
}
if (rld.delete(doc)) {
delCount++;
}
}
}
}
}
return delCount;
}
// Delete by Term
private synchronized long applyTermDeletes(
Iterable<Term> termsIter, ReadersAndLiveDocs rld, SegmentReader reader) throws IOException {
long delCount = 0;
Fields fields = reader.fields();
if (fields == null) {
// This reader has no postings
return 0;
}
TermsEnum termsEnum = null;
String currentField = null;
DocsEnum docs = null;
assert checkDeleteTerm(null);
boolean any = false;
// System.out.println(Thread.currentThread().getName() + " del terms reader=" + reader);
for (Term term : termsIter) {
// Since we visit terms sorted, we gain performance
// by re-using the same TermsEnum and seeking only
// forwards
if (!term.field().equals(currentField)) {
assert currentField == null || currentField.compareTo(term.field()) < 0;
currentField = term.field();
Terms terms = fields.terms(currentField);
if (terms != null) {
termsEnum = terms.iterator(null);
} else {
termsEnum = null;
}
}
if (termsEnum == null) {
continue;
}
assert checkDeleteTerm(term);
// System.out.println(" term=" + term);
if (termsEnum.seekExact(term.bytes(), false)) {
// we don't need term frequencies for this
DocsEnum docsEnum = termsEnum.docs(rld.getLiveDocs(), docs, DocsEnum.FLAG_NONE);
// System.out.println("BDS: got docsEnum=" + docsEnum);
if (docsEnum != null) {
while (true) {
final int docID = docsEnum.nextDoc();
// System.out.println(Thread.currentThread().getName() + " del term=" + term + " doc=" +
// docID);
if (docID == DocIdSetIterator.NO_MORE_DOCS) {
break;
}
// NOTE: there is no limit check on the docID
// when deleting by Term (unlike by Query)
// because on flush we apply all Term deletes to
// each segment. So all Term deleting here is
// against prior segments:
if (!any) {
rld.initWritableLiveDocs();
any = true;
}
if (rld.delete(docID)) {
delCount++;
}
}
}
}
}
return delCount;
}
/**
* Resolves the buffered deleted Term/Query/docIDs, into actual deleted docIDs in the liveDocs
* MutableBits for each SegmentReader.
*/
public synchronized ApplyDeletesResult applyDeletes(
IndexWriter.ReaderPool readerPool, List<SegmentInfoPerCommit> infos) throws IOException {
final long t0 = System.currentTimeMillis();
if (infos.size() == 0) {
return new ApplyDeletesResult(false, nextGen++, null);
}
assert checkDeleteStats();
if (!any()) {
if (infoStream.isEnabled("BD")) {
infoStream.message("BD", "applyDeletes: no deletes; skipping");
}
return new ApplyDeletesResult(false, nextGen++, null);
}
if (infoStream.isEnabled("BD")) {
infoStream.message("BD", "applyDeletes: infos=" + infos + " packetCount=" + deletes.size());
}
final long gen = nextGen++;
List<SegmentInfoPerCommit> infos2 = new ArrayList<SegmentInfoPerCommit>();
infos2.addAll(infos);
Collections.sort(infos2, sortSegInfoByDelGen);
CoalescedDeletes coalescedDeletes = null;
boolean anyNewDeletes = false;
int infosIDX = infos2.size() - 1;
int delIDX = deletes.size() - 1;
List<SegmentInfoPerCommit> allDeleted = null;
while (infosIDX >= 0) {
// System.out.println("BD: cycle delIDX=" + delIDX + " infoIDX=" + infosIDX);
final FrozenBufferedDeletes packet = delIDX >= 0 ? deletes.get(delIDX) : null;
final SegmentInfoPerCommit info = infos2.get(infosIDX);
final long segGen = info.getBufferedDeletesGen();
if (packet != null && segGen < packet.delGen()) {
// System.out.println(" coalesce");
if (coalescedDeletes == null) {
coalescedDeletes = new CoalescedDeletes();
}
if (!packet.isSegmentPrivate) {
/*
* Only coalesce if we are NOT on a segment private del packet: the segment private del packet
* must only applied to segments with the same delGen. Yet, if a segment is already deleted
* from the SI since it had no more documents remaining after some del packets younger than
* its segPrivate packet (higher delGen) have been applied, the segPrivate packet has not been
* removed.
*/
coalescedDeletes.update(packet);
}
delIDX--;
} else if (packet != null && segGen == packet.delGen()) {
assert packet.isSegmentPrivate
: "Packet and Segments deletegen can only match on a segment private del packet gen="
+ segGen;
// System.out.println(" eq");
// Lock order: IW -> BD -> RP
assert readerPool.infoIsLive(info);
final ReadersAndLiveDocs rld = readerPool.get(info, true);
final SegmentReader reader = rld.getReader(IOContext.READ);
int delCount = 0;
final boolean segAllDeletes;
try {
if (coalescedDeletes != null) {
// System.out.println(" del coalesced");
delCount += applyTermDeletes(coalescedDeletes.termsIterable(), rld, reader);
delCount += applyQueryDeletes(coalescedDeletes.queriesIterable(), rld, reader);
}
// System.out.println(" del exact");
// Don't delete by Term here; DocumentsWriterPerThread
// already did that on flush:
delCount += applyQueryDeletes(packet.queriesIterable(), rld, reader);
final int fullDelCount = rld.info.getDelCount() + rld.getPendingDeleteCount();
assert fullDelCount <= rld.info.info.getDocCount();
segAllDeletes = fullDelCount == rld.info.info.getDocCount();
} finally {
rld.release(reader);
readerPool.release(rld);
}
anyNewDeletes |= delCount > 0;
if (segAllDeletes) {
if (allDeleted == null) {
allDeleted = new ArrayList<SegmentInfoPerCommit>();
}
allDeleted.add(info);
}
if (infoStream.isEnabled("BD")) {
infoStream.message(
"BD",
"seg="
+ info
+ " segGen="
+ segGen
+ " segDeletes=["
+ packet
+ "]; coalesced deletes=["
+ (coalescedDeletes == null ? "null" : coalescedDeletes)
+ "] newDelCount="
+ delCount
+ (segAllDeletes ? " 100% deleted" : ""));
}
if (coalescedDeletes == null) {
coalescedDeletes = new CoalescedDeletes();
}
/*
* Since we are on a segment private del packet we must not
* update the coalescedDeletes here! We can simply advance to the
* next packet and seginfo.
*/
delIDX--;
infosIDX--;
info.setBufferedDeletesGen(gen);
} else {
// System.out.println(" gt");
if (coalescedDeletes != null) {
// Lock order: IW -> BD -> RP
assert readerPool.infoIsLive(info);
final ReadersAndLiveDocs rld = readerPool.get(info, true);
final SegmentReader reader = rld.getReader(IOContext.READ);
int delCount = 0;
final boolean segAllDeletes;
try {
delCount += applyTermDeletes(coalescedDeletes.termsIterable(), rld, reader);
delCount += applyQueryDeletes(coalescedDeletes.queriesIterable(), rld, reader);
final int fullDelCount = rld.info.getDelCount() + rld.getPendingDeleteCount();
assert fullDelCount <= rld.info.info.getDocCount();
segAllDeletes = fullDelCount == rld.info.info.getDocCount();
} finally {
rld.release(reader);
readerPool.release(rld);
}
anyNewDeletes |= delCount > 0;
if (segAllDeletes) {
if (allDeleted == null) {
allDeleted = new ArrayList<SegmentInfoPerCommit>();
}
allDeleted.add(info);
}
if (infoStream.isEnabled("BD")) {
infoStream.message(
"BD",
"seg="
+ info
+ " segGen="
+ segGen
+ " coalesced deletes=["
+ (coalescedDeletes == null ? "null" : coalescedDeletes)
+ "] newDelCount="
+ delCount
+ (segAllDeletes ? " 100% deleted" : ""));
}
}
info.setBufferedDeletesGen(gen);
infosIDX--;
}
}
assert checkDeleteStats();
if (infoStream.isEnabled("BD")) {
infoStream.message("BD", "applyDeletes took " + (System.currentTimeMillis() - t0) + " msec");
}
// assert infos != segmentInfos || !any() : "infos=" + infos + " segmentInfos=" + segmentInfos +
// " any=" + any;
return new ApplyDeletesResult(anyNewDeletes, gen, allDeleted);
}
