Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Reduce FST block size for BlockTreeTermsWriter #12604

Merged
merged 6 commits into from
Oct 4, 2023
Merged

Reduce FST block size for BlockTreeTermsWriter #12604

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
edfd968
reduce fst block size
gf2121 Sep 28, 2023
43e67cc
add CHANGES
gf2121 Sep 28, 2023
521d693
fix
gf2121 Sep 28, 2023
1eb6a99
estimate bit
gf2121 Oct 3, 2023
a4a4964
update changes
gf2121 Oct 3, 2023
21b6858
fix change
gf2121 Oct 4, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions lucene/CHANGES.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,9 @@ Optimizations
* GITHUB#12382: Faster top-level conjunctions on term queries when sorting by
descending score. (Adrien Grand)

* GITHUB#12604: Estimate the block size of FST BytesStore in BlockTreeTermsWriter
to reduce GC load during indexing. (Guo Feng)

Changes in runtime behavior
---------------------

Expand Down
13 changes: 13 additions & 0 deletions ...re/src/java/org/apache/lucene/codecs/lucene90/blocktree/Lucene90BlockTreeTermsWriter.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,6 +52,7 @@
import org.apache.lucene.util.fst.FST;
import org.apache.lucene.util.fst.FSTCompiler;
import org.apache.lucene.util.fst.Util;
import org.apache.lucene.util.packed.PackedInts;

/*
TODO:
Expand Down Expand Up @@ -490,10 +491,22 @@ public void compileIndex(
}
}

long estimateSize = prefix.length;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Too bad we don't have a writer that uses tiny (like 8 bytes) block at first, but doubles size for each new block (16 bytes, 32 bytes next, etc.). Then we would naturally use log(size) number of blocks without over-allocating.

But then reading bytes is a bit tricky because we'd need to take discrete log (base 2) of the address. Maybe it wouldn't be so bad -- we could do this with Long.numberOfLeadingZeros maybe? But that's a bigger change ... we can do this separately/later.

for (PendingBlock block : blocks) {
if (block.subIndices != null) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We also should really explore the TODO above to write vLong in opposite byte order -- this might save quite a bit of storage in the FST since outputs would share more prefixes. Again, separate issue 😀

for (FST<BytesRef> subIndex : block.subIndices) {
estimateSize += subIndex.numBytes();
}
}
}
int estimateBitsRequired = PackedInts.bitsRequired(estimateSize);
int pageBits = Math.min(15, Math.max(6, estimateBitsRequired));

final ByteSequenceOutputs outputs = ByteSequenceOutputs.getSingleton();
final FSTCompiler<BytesRef> fstCompiler =
new FSTCompiler.Builder<>(FST.INPUT_TYPE.BYTE1, outputs)
.shouldShareNonSingletonNodes(false)
.bytesPageBits(pageBits)
.build();
// if (DEBUG) {
// System.out.println(" compile index for prefix=" + prefix);
Expand Down
4 changes: 4 additions & 0 deletions lucene/core/src/java/org/apache/lucene/util/fst/FST.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -520,6 +520,10 @@ void finish(long newStartNode) throws IOException {
bytes.finish();
}

public long numBytes() {
return bytes.getPosition();
}

public T getEmptyOutput() {
return emptyOutput;
}
Expand Down
Loading