Compare commits

..

4 Commits

Author SHA1 Message Date
Paul Masurel
08f7706973 test store 2021-01-09 10:27:03 +09:00
Paul Masurel
bf6e6e8a7c Merge pull request #972 from tantivy-search/issue/969
Issue/969
2021-01-07 22:49:31 +09:00
Paul Masurel
203b0256a3 Minor renaming 2021-01-07 22:47:57 +09:00
Paul Masurel
caf2a38b7e Closes #969.
The segment stacking optimization is not updating "first_doc_in_block".
2021-01-07 22:43:56 +09:00
8 changed files with 91 additions and 16 deletions

View File

@@ -47,6 +47,7 @@ murmurhash32 = "0.2"
chrono = "0.4"
smallvec = "1"
rayon = "1"
env_logger = "0.8"
lru = "0.6"
[target.'cfg(windows)'.dependencies]

View File

@@ -35,12 +35,18 @@ fn load_metas(
inventory: &SegmentMetaInventory,
) -> crate::Result<IndexMeta> {
let meta_data = directory.atomic_read(&META_FILEPATH)?;
let meta_string = String::from_utf8_lossy(&meta_data);
let meta_string = String::from_utf8(meta_data)
.map_err(|utf8_err| {
DataCorruption::new(
META_FILEPATH.to_path_buf(),
format!("Meta file is not valid utf-8. {:?}", utf8_err)
)
})?;
IndexMeta::deserialize(&meta_string, &inventory)
.map_err(|e| {
DataCorruption::new(
META_FILEPATH.to_path_buf(),
format!("Meta file cannot be deserialized. {:?}.", e),
format!("Meta file cannot be deserialized. {:?}. content = {}", e, meta_string),
)
})
.map_err(From::from)

View File

@@ -20,6 +20,7 @@ fn check_index_content(searcher: &Searcher, vals: &[u64]) -> crate::Result<()> {
#[test]
#[ignore]
fn test_functional_store() -> crate::Result<()> {
env_logger::init();
let mut schema_builder = Schema::builder();
let id_field = schema_builder.add_u64_field("id", INDEXED | STORED);
@@ -35,8 +36,7 @@ fn test_functional_store() -> crate::Result<()> {
let mut doc_set: Vec<u64> = Vec::new();
let mut doc_id = 0u64;
for iteration in 0..500 {
dbg!(iteration);
for iteration in 0.. {
let num_docs: usize = rng.gen_range(0..4);
if doc_set.len() >= 1 {
let doc_to_remove_id = rng.gen_range(0..doc_set.len());
@@ -51,6 +51,7 @@ fn test_functional_store() -> crate::Result<()> {
index_writer.commit()?;
reader.reload()?;
let searcher = reader.searcher();
println!("#{} - {}", iteration, searcher.segment_readers().len());
check_index_content(&searcher, &doc_set)?;
}
Ok(())

View File

@@ -43,6 +43,9 @@ impl CheckpointBlock {
/// Adding another checkpoint in the block.
pub fn push(&mut self, checkpoint: Checkpoint) {
if let Some(prev_checkpoint) = self.checkpoints.last() {
assert!(checkpoint.follows(prev_checkpoint));
}
self.checkpoints.push(checkpoint);
}

View File

@@ -1,4 +1,4 @@
const CHECKPOINT_PERIOD: usize = 8;
const CHECKPOINT_PERIOD: usize = 2;
use std::fmt;
mod block;
@@ -26,6 +26,13 @@ pub struct Checkpoint {
pub end_offset: u64,
}
impl Checkpoint {
pub(crate) fn follows(&self, other: &Checkpoint) -> bool {
(self.start_doc == other.end_doc) &&
(self.start_offset == other.end_offset)
}
}
impl fmt::Debug for Checkpoint {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(
@@ -137,7 +144,7 @@ mod tests {
}
#[test]
fn test_merge_store_with_stacking() -> crate::Result<()> {
fn test_merge_store_with_stacking_reproducing_issue969() -> crate::Result<()> {
let mut schema_builder = SchemaBuilder::default();
let text = schema_builder.add_text_field("text", STORED | STRING);
let body = schema_builder.add_text_field("body", STORED);

View File

@@ -28,18 +28,20 @@ impl LayerBuilder {
///
/// If the block was empty to begin with, simply return None.
fn flush_block(&mut self) -> Option<Checkpoint> {
self.block.doc_interval().map(|(start_doc, end_doc)| {
if let Some((start_doc, end_doc)) = self.block.doc_interval() {
let start_offset = self.buffer.len() as u64;
self.block.serialize(&mut self.buffer);
let end_offset = self.buffer.len() as u64;
self.block.clear();
Checkpoint {
Some(Checkpoint {
start_doc,
end_doc,
start_offset,
end_offset,
}
})
})
} else {
None
}
}
fn push(&mut self, checkpoint: Checkpoint) {
@@ -48,7 +50,7 @@ impl LayerBuilder {
fn insert(&mut self, checkpoint: Checkpoint) -> Option<Checkpoint> {
self.push(checkpoint);
let emit_skip_info = (self.block.len() % CHECKPOINT_PERIOD) == 0;
let emit_skip_info = self.block.len() >= CHECKPOINT_PERIOD;
if emit_skip_info {
self.flush_block()
} else {

50
src/store/tests_store.rs Normal file
View File

@@ -0,0 +1,50 @@
use std::path::Path;
use crate::HasLen;
use crate::directory::{Directory, ManagedDirectory, MmapDirectory, RAMDirectory};
use crate::fastfield::DeleteBitSet;
use super::{StoreReader, StoreWriter};
#[test]
fn test_toto2() -> crate::Result<()> {
let directory = ManagedDirectory::wrap(MmapDirectory::open("src/store/broken_seg")?)?;
let path = Path::new("b6029ade1b954ea1acad15b432eaacb9.store");
assert!(directory.validate_checksum(path)?);
let store_file = directory.open_read(path)?;
let store = StoreReader::open(store_file)?;
let documents = store.documents();
// for doc in documents {
// println!("{:?}", doc);
// }
let doc= store.get(15_086)?;
Ok(())
}
#[test]
fn test_toto() -> crate::Result<()> {
let directory = ManagedDirectory::wrap(MmapDirectory::open("src/store/broken_seg")?)?;
assert!(directory.validate_checksum(Path::new("e6ece22e5bca4e0dbe7ce3e4dcbd5bbf.store"))?);
let store_file = directory.open_read(Path::new("e6ece22e5bca4e0dbe7ce3e4dcbd5bbf.store.patched"))?;
let store = StoreReader::open(store_file)?;
let doc= store.get(53)?;
println!("{:?}", doc);
// let documents = store.documents();
// let ram_directory = RAMDirectory::create();
// let path = Path::new("store");
// let store_wrt = ram_directory.open_write(path)?;
// let mut store_writer = StoreWriter::new(store_wrt);
// for doc in &documents {
// store_writer.store(doc)?;
// }
// store_writer.close()?;
// let store_data = ram_directory.open_read(path)?;
// let new_store = StoreReader::open(store_data)?;
// for doc in 0..59 {
// println!("{}", doc);
// let doc = new_store.get(doc)?;
// println!("{:?}", doc);
// }
Ok(())
}

View File

@@ -10,7 +10,7 @@ use crate::store::index::Checkpoint;
use crate::DocId;
use std::io::{self, Write};
const BLOCK_SIZE: usize = 16_384;
const BLOCK_SIZE: usize = 30;
/// Write tantivy's [`Store`](./index.html)
///
@@ -72,6 +72,7 @@ impl StoreWriter {
if !self.current_block.is_empty() {
self.write_and_compress_block()?;
}
assert_eq!(self.first_doc_in_block, self.doc);
let doc_shift = self.doc;
let start_shift = self.writer.written_bytes() as u64;
@@ -86,12 +87,17 @@ impl StoreWriter {
checkpoint.end_doc += doc_shift;
checkpoint.start_offset += start_shift;
checkpoint.end_offset += start_shift;
self.offset_index_writer.insert(checkpoint);
self.doc = checkpoint.end_doc;
self.register_checkpoint(checkpoint);
}
Ok(())
}
fn register_checkpoint(&mut self, checkpoint: Checkpoint) {
self.offset_index_writer.insert(checkpoint);
self.first_doc_in_block = checkpoint.end_doc;
self.doc = checkpoint.end_doc;
}
fn write_and_compress_block(&mut self) -> io::Result<()> {
assert!(self.doc > 0);
self.intermediary_buffer.clear();
@@ -100,14 +106,13 @@ impl StoreWriter {
self.writer.write_all(&self.intermediary_buffer)?;
let end_offset = self.writer.written_bytes();
let end_doc = self.doc;
self.offset_index_writer.insert(Checkpoint {
self.register_checkpoint(Checkpoint {
start_doc: self.first_doc_in_block,
end_doc,
start_offset,
end_offset,
});
self.current_block.clear();
self.first_doc_in_block = self.doc;
Ok(())
}