use fst for sstable index (#2268)

* read path for new fst based index

* implement BlockAddrStoreWriter

* extract slop/derivation computation

* use better linear approximator and allow negative correction to approximator

* document format and reorder some fields

* optimize single block sstable size

* plug backward compat

columnar/src/tests.rs

@@ -26,7 +26,7 @@ fn test_dataframe_writer_str() {
     assert_eq!(columnar.num_columns(), 1);
     let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
     assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 87);
+    assert_eq!(cols[0].num_bytes(), 73);
 }
 
 #[test]
@@ -40,7 +40,7 @@ fn test_dataframe_writer_bytes() {
     assert_eq!(columnar.num_columns(), 1);
     let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
     assert_eq!(cols.len(), 1);
-    assert_eq!(cols[0].num_bytes(), 87);
+    assert_eq!(cols[0].num_bytes(), 73);
 }
 
 #[test]

src/fastfield/mod.rs

@@ -131,7 +131,7 @@ mod tests {
         }
         let file = directory.open_read(path).unwrap();
 
-        assert_eq!(file.len(), 93);
+        assert_eq!(file.len(), 80);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let column = fast_field_readers
             .u64("field")
@@ -181,7 +181,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 121);
+        assert_eq!(file.len(), 108);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let col = fast_field_readers
             .u64("field")
@@ -214,7 +214,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 94);
+        assert_eq!(file.len(), 81);
         let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
         let fast_field_reader = fast_field_readers
             .u64("field")
@@ -246,7 +246,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 4489);
+        assert_eq!(file.len(), 4476);
         {
             let fast_field_readers = FastFieldReaders::open(file, SCHEMA.clone()).unwrap();
             let col = fast_field_readers
@@ -279,7 +279,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 265);
+        assert_eq!(file.len(), 252);
 
         {
             let fast_field_readers = FastFieldReaders::open(file, schema).unwrap();
@@ -773,7 +773,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 102);
+        assert_eq!(file.len(), 84);
         let fast_field_readers = FastFieldReaders::open(file, schema).unwrap();
         let bool_col = fast_field_readers.bool("field_bool").unwrap();
         assert_eq!(bool_col.first(0), Some(true));
@@ -805,7 +805,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 114);
+        assert_eq!(file.len(), 96);
         let readers = FastFieldReaders::open(file, schema).unwrap();
         let bool_col = readers.bool("field_bool").unwrap();
         for i in 0..25 {
@@ -830,7 +830,7 @@ mod tests {
             write.terminate().unwrap();
         }
         let file = directory.open_read(path).unwrap();
-        assert_eq!(file.len(), 104);
+        assert_eq!(file.len(), 86);
         let fastfield_readers = FastFieldReaders::open(file, schema).unwrap();
         let col = fastfield_readers.bool("field_bool").unwrap();
         assert_eq!(col.first(0), None);

sstable/Cargo.toml

@@ -11,6 +11,7 @@ description = "sstables for tantivy"
 
 [dependencies]
 common = {version= "0.6", path="../common", package="tantivy-common"}
+tantivy-bitpacker = { version= "0.5", path="../bitpacker" }
 tantivy-fst = "0.5"
 # experimental gives us access to Decompressor::upper_bound
 zstd = { version = "0.13", features = ["experimental"] }

sstable/README.md

@@ -89,33 +89,71 @@ Note: as the SSTable does not support redundant keys, there is no ambiguity betw
 
 ### SSTFooter
 ```
-+-------+-------+-----+-------------+---------+---------+
-| Block | Block | ... | IndexOffset | NumTerm | Version |
-+-------+-------+-----+-------------+---------+---------+
-|----( # of blocks)---|
++-----+----------------+-------------+-------------+---------+---------+
+| Fst | BlockAddrStore | StoreOffset | IndexOffset | NumTerm | Version |
++-----+----------------+-------------+-------------+---------+---------+
 ```
-- Block(SSTBlock): uses IndexValue for its Values format
+- Fst(Fst): finite state transducer mapping keys to a block number
+- BlockAddrStore(BlockAddrStore): store mapping a block number to its BlockAddr
+- StoreOffset(u64): Offset to start of the BlockAddrStore. If zero, see the SingleBlockSStable section
 - IndexOffset(u64): Offset to the start of the SSTFooter
 - NumTerm(u64): number of terms in the sstable
-- Version(u32): Currently equal to 2
+- Version(u32): Currently equal to 3
 
-### IndexValue
-```
-+------------+----------+-------+-------+-----+
-| EntryCount | StartPos | Entry | Entry | ... |
-+------------+----------+-------+-------+-----+
-                        |---( # of entries)---|
-```
+### Fst
 
-- EntryCount(VInt): number of entries
-- StartPos(VInt): the start pos of the first (data) block referenced by this (index) block
-- Entry (IndexEntry)
+Fst is in the format of tantivy\_fst
 
-### Entry
-```
-+----------+--------------+
-| BlockLen | FirstOrdinal |
-+----------+--------------+
-```
-- BlockLen(VInt): length of the block
-- FirstOrdinal(VInt): ordinal of the first element in the given block
+### BlockAddrStore
+
++---------+-----------+-----------+-----+-----------+-----------+-----+
+| MetaLen | BlockMeta | BlockMeta | ... | BlockData | BlockData | ... |
++---------+-----------+-----------+-----+-----------+-----------+-----+
+          |---------(N blocks)----------|---------(N blocks)----------|
+
+- MetaLen(u64): length of the BlockMeta section
+- BlockMeta(BlockAddrBlockMetadata): metadata to seek through BlockData
+- BlockData(CompactedBlockAddr): bitpacked per block metadata
+
+### BlockAddrBlockMetadata
+
++--------+------------+--------------+------------+--------------+-------------------+-----------------+----------+
+| Offset | RangeStart | FirstOrdinal | RangeSlope | OrdinalSlope | FirstOrdinalNBits | RangeStartNBits | BlockLen |
++--------+------------+--------------+------------+--------------+-------------------+-----------------+----------+
+
+- Offset(u64): offset of the corresponding BlockData in the datastream
+- RangeStart(u64): the start position of the first block
+- FirstOrdinal(u64): the first ordinal of the first block
+- RangeSlope(u32): slope predicted for start range evolution (see computation in BlockData)
+- OrdinalSlope(u64): slope predicted for first ordinal evolution (see computation in BlockData)
+- FirstOrdinalNBits(u8): number of bits per ordinal in datastream (see computation in BlockData)
+- RangeStartNBits(u8): number of bits per range start in datastream (see computation in BlockData)
+
+### BlockData
+
++-----------------+-------------------+---------------+
+| RangeStartDelta | FirstOrdinalDelta | FinalRangeEnd |
++-----------------+-------------------+---------------+
+|------(BlockLen repetitions)---------|
+
+- RangeStartDelta(var): RangeStartNBits *bits* of little endian number. See below for decoding
+- FirstOrdinalDelta(var): FirstOrdinalNBits *bits* of little endian number. See below for decoding
+- FinalRangeEnd(var): RangeStartNBits *bits* of integer. See below for decoding
+
+converting a BlockData of index Index and a BlockAddrBlockMetadata to an actual block address is done as follow:
+range\_prediction := RangeStart + Index * RangeSlop;
+range\_derivation := RangeStartDelta - (1 << (RangeStartNBits-1));
+range\_start := range\_prediction + range\_derivation
+
+The same computation can be done for ordinal.
+
+Note that `range_derivation` can take negative value. `RangeStartDelta` is just its translation to a positive range.
+
+
+## SingleBlockSStable
+
+The format used for the index is meant to be compact, however it has a constant cost of around 70
+bytes, which isn't negligible for a table containing very few keys.
+To limit the impact of that constant cost, single block sstable omit the Fst and BlockAddrStore from
+their index. Instead a block with first ordinal of 0, range start of 0 and range end of IndexOffset
+is implicitly used for every operations.

sstable/benches/ord_to_term.rs

@@ -40,6 +40,31 @@ pub fn criterion_benchmark(c: &mut Criterion) {
                 assert!(dict.ord_to_term(19_000_000, &mut res).unwrap());
             })
         });
+        c.bench_function("term_ord_suffix", |b| {
+            b.iter(|| {
+                assert_eq!(
+                    dict.term_ord(b"prefix.00186A0.suffix").unwrap().unwrap(),
+                    100_000
+                );
+                assert_eq!(
+                    dict.term_ord(b"prefix.121EAC0.suffix").unwrap().unwrap(),
+                    19_000_000
+                );
+            })
+        });
+        c.bench_function("open_and_term_ord_suffix", |b| {
+            b.iter(|| {
+                let dict = Dictionary::<MonotonicU64SSTable>::open(slice.clone()).unwrap();
+                assert_eq!(
+                    dict.term_ord(b"prefix.00186A0.suffix").unwrap().unwrap(),
+                    100_000
+                );
+                assert_eq!(
+                    dict.term_ord(b"prefix.121EAC0.suffix").unwrap().unwrap(),
+                    19_000_000
+                );
+            })
+        });
     }
     {
         let slice = make_test_sstable("");
@@ -59,6 +84,25 @@ pub fn criterion_benchmark(c: &mut Criterion) {
                 assert!(dict.ord_to_term(19_000_000, &mut res).unwrap());
             })
         });
+        c.bench_function("term_ord", |b| {
+            b.iter(|| {
+                assert_eq!(dict.term_ord(b"prefix.00186A0").unwrap().unwrap(), 100_000);
+                assert_eq!(
+                    dict.term_ord(b"prefix.121EAC0").unwrap().unwrap(),
+                    19_000_000
+                );
+            })
+        });
+        c.bench_function("open_and_term_ord", |b| {
+            b.iter(|| {
+                let dict = Dictionary::<MonotonicU64SSTable>::open(slice.clone()).unwrap();
+                assert_eq!(dict.term_ord(b"prefix.00186A0").unwrap().unwrap(), 100_000);
+                assert_eq!(
+                    dict.term_ord(b"prefix.121EAC0").unwrap().unwrap(),
+                    19_000_000
+                );
+            })
+        });
     }
 }
 

sstable/src/dictionary.rs

@@ -9,8 +9,11 @@ use common::{BinarySerializable, OwnedBytes};
 use tantivy_fst::automaton::AlwaysMatch;
 use tantivy_fst::Automaton;
 
+use crate::sstable_index_v3::SSTableIndexV3Empty;
 use crate::streamer::{Streamer, StreamerBuilder};
-use crate::{BlockAddr, DeltaReader, Reader, SSTable, SSTableIndex, TermOrdinal, VoidSSTable};
+use crate::{
+    BlockAddr, DeltaReader, Reader, SSTable, SSTableIndex, SSTableIndexV3, TermOrdinal, VoidSSTable,
+};
 
 /// An SSTable is a sorted map that associates sorted `&[u8]` keys
 /// to any kind of typed values.
@@ -180,24 +183,41 @@ impl<TSSTable: SSTable> Dictionary<TSSTable> {
     pub fn open(term_dictionary_file: FileSlice) -> io::Result<Self> {
         let (main_slice, footer_len_slice) = term_dictionary_file.split_from_end(20);
         let mut footer_len_bytes: OwnedBytes = footer_len_slice.read_bytes()?;
-
         let index_offset = u64::deserialize(&mut footer_len_bytes)?;
         let num_terms = u64::deserialize(&mut footer_len_bytes)?;
         let version = u32::deserialize(&mut footer_len_bytes)?;
-        if version != crate::SSTABLE_VERSION {
-            return Err(io::Error::new(
-                io::ErrorKind::Other,
-                format!(
-                    "Unsuported sstable version, expected {version}, found {}",
-                    crate::SSTABLE_VERSION,
-                ),
-            ));
-        }
-
         let (sstable_slice, index_slice) = main_slice.split(index_offset as usize);
         let sstable_index_bytes = index_slice.read_bytes()?;
-        let sstable_index = SSTableIndex::load(sstable_index_bytes)
-            .map_err(|_| io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption"))?;
+
+        let sstable_index = match version {
+            2 => SSTableIndex::V2(
+                crate::sstable_index_v2::SSTableIndex::load(sstable_index_bytes).map_err(|_| {
+                    io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption")
+                })?,
+            ),
+            3 => {
+                let (sstable_index_bytes, mut footerv3_len_bytes) = sstable_index_bytes.rsplit(8);
+                let store_offset = u64::deserialize(&mut footerv3_len_bytes)?;
+                if store_offset != 0 {
+                    SSTableIndex::V3(
+                        SSTableIndexV3::load(sstable_index_bytes, store_offset).map_err(|_| {
+                            io::Error::new(io::ErrorKind::InvalidData, "SSTable corruption")
+                        })?,
+                    )
+                } else {
+                    // if store_offset is zero, there is no index, so we build a pseudo-index
+                    // assuming a single block of sstable covering everything.
+                    SSTableIndex::V3Empty(SSTableIndexV3Empty::load(index_offset as usize))
+                }
+            }
+            _ => {
+                return Err(io::Error::new(
+                    io::ErrorKind::Other,
+                    format!("Unsuported sstable version, expected one of [2, 3], found {version}"),
+                ))
+            }
+        };
+
         Ok(Dictionary {
             sstable_slice,
             sstable_index,

sstable/src/lib.rs

@@ -10,8 +10,9 @@ pub mod merge;
 mod streamer;
 pub mod value;
 
-mod sstable_index;
-pub use sstable_index::{BlockAddr, SSTableIndex, SSTableIndexBuilder};
+mod sstable_index_v3;
+pub use sstable_index_v3::{BlockAddr, SSTableIndex, SSTableIndexBuilder, SSTableIndexV3};
+mod sstable_index_v2;
 pub(crate) mod vint;
 pub use dictionary::Dictionary;
 pub use streamer::{Streamer, StreamerBuilder};
@@ -28,7 +29,7 @@ use crate::value::{RangeValueReader, RangeValueWriter};
 pub type TermOrdinal = u64;
 
 const DEFAULT_KEY_CAPACITY: usize = 50;
-const SSTABLE_VERSION: u32 = 2;
+const SSTABLE_VERSION: u32 = 3;
 
 /// Given two byte string returns the length of
 /// the longest common prefix.
@@ -304,7 +305,8 @@ where
 
         let offset = wrt.written_bytes();
 
-        self.index_builder.serialize(&mut wrt)?;
+        let fst_len: u64 = self.index_builder.serialize(&mut wrt)?;
+        wrt.write_all(&fst_len.to_le_bytes())?;
         wrt.write_all(&offset.to_le_bytes())?;
         wrt.write_all(&self.num_terms.to_le_bytes())?;
 
@@ -385,13 +387,10 @@ mod test {
                 16, 17, 33, 18, 19, 17, 20, // data block
                 0, 0, 0, 0, // no more block
                 // index
-                8, 0, 0, 0, // size of index block
-                0, // compression
-                1, 0, 12, 0, 32, 17, 20, // index block
-                0, 0, 0, 0, // no more index block
+                0, 0, 0, 0, 0, 0, 0, 0, // fst lenght
                 16, 0, 0, 0, 0, 0, 0, 0, // index start offset
                 3, 0, 0, 0, 0, 0, 0, 0, // num term
-                2, 0, 0, 0, // version
+                3, 0, 0, 0, // version
             ]
         );
         let buffer = OwnedBytes::new(buffer);