mirror of
https://github.com/quickwit-oss/tantivy.git
synced 2026-05-18 17:20:41 +00:00
513 lines
17 KiB
Rust
513 lines
17 KiB
Rust
use std::mem;
|
|
|
|
use common::serialize_vint_u32;
|
|
|
|
use crate::fastcpy::fast_short_slice_copy;
|
|
use crate::{Addr, MemoryArena};
|
|
|
|
const FIRST_BLOCK_NUM: u32 = 2;
|
|
|
|
/// An exponential unrolled link.
|
|
///
|
|
/// The use case is as follows. Tantivy's indexer conceptually acts like a
|
|
/// `HashMap<Term, Vec<u32>>`. As we come across a given term in document
|
|
/// `D`, we lookup the term in the map and append the document id to its vector.
|
|
///
|
|
/// The vector is then only read when it is serialized.
|
|
///
|
|
/// The `ExpUnrolledLinkedList` offers a more efficient solution to this
|
|
/// problem.
|
|
///
|
|
/// It combines the idea of the unrolled linked list and tries to address the
|
|
/// problem of selecting an adequate block size using a strategy similar to
|
|
/// that of the `Vec` amortized resize strategy.
|
|
///
|
|
/// Data is stored in a linked list of blocks. The first block has a size of `8`
|
|
/// and each block has a length of twice that of the previous block up to
|
|
/// `MAX_BLOCK_LEN = 1<<15`.
|
|
///
|
|
/// This strategy is a good trade off to handle numerous very rare terms
|
|
/// and avoid wasting half of the memory for very frequent terms.
|
|
#[derive(Debug, Clone, Copy)]
|
|
pub struct ExpUnrolledLinkedList {
|
|
// u16, since the max size of each block is (1<<next_cap_pow_2)
|
|
// Limited to 15, so we don't overflow remaining_cap.
|
|
remaining_cap: u16,
|
|
// Tracks the number of blocks allocated: block_num - FIRST_BLOCK_NUM
|
|
block_num: u32,
|
|
head: Addr,
|
|
tail: Addr,
|
|
}
|
|
|
|
impl Default for ExpUnrolledLinkedList {
|
|
fn default() -> Self {
|
|
Self {
|
|
// 0 to trigger an initial allocation. Init with MemoryArena would be better.
|
|
remaining_cap: 0,
|
|
block_num: FIRST_BLOCK_NUM,
|
|
head: Addr::null_pointer(),
|
|
tail: Addr::null_pointer(),
|
|
}
|
|
}
|
|
}
|
|
|
|
pub struct ExpUnrolledLinkedListWriter<'a> {
|
|
eull: &'a mut ExpUnrolledLinkedList,
|
|
arena: &'a mut MemoryArena,
|
|
}
|
|
|
|
#[inline]
|
|
fn ensure_capacity<'a>(
|
|
eull: &'a mut ExpUnrolledLinkedList,
|
|
arena: &'a mut MemoryArena,
|
|
allocate: u32,
|
|
) {
|
|
let new_block_addr: Addr = arena.allocate_space(allocate as usize + mem::size_of::<Addr>());
|
|
// Check first write
|
|
if eull.head.is_null() {
|
|
eull.head = new_block_addr;
|
|
} else {
|
|
arena.write_at(eull.tail, new_block_addr);
|
|
}
|
|
|
|
eull.tail = new_block_addr;
|
|
eull.remaining_cap = allocate as u16;
|
|
}
|
|
|
|
impl ExpUnrolledLinkedListWriter<'_> {
|
|
#[inline]
|
|
pub fn write_u32_vint(&mut self, val: u32) {
|
|
let mut buf = [0u8; 8];
|
|
let data = serialize_vint_u32(val, &mut buf);
|
|
self.extend_from_slice(data);
|
|
}
|
|
|
|
#[inline]
|
|
pub fn extend_from_slice(&mut self, mut buf: &[u8]) {
|
|
while !buf.is_empty() {
|
|
let add_len: usize;
|
|
{
|
|
if self.eull.remaining_cap == 0 {
|
|
// Double the next cap
|
|
self.eull.increment_num_blocks();
|
|
let block_size = get_block_size(self.eull.block_num);
|
|
ensure_capacity(self.eull, self.arena, block_size as u32);
|
|
}
|
|
|
|
let output_buf = self
|
|
.arena
|
|
.slice_mut(self.eull.tail, self.eull.remaining_cap as usize);
|
|
add_len = buf.len().min(output_buf.len());
|
|
let output_buf = &mut output_buf[..add_len];
|
|
let buf = &buf[..add_len];
|
|
|
|
fast_short_slice_copy(buf, output_buf);
|
|
}
|
|
self.eull.remaining_cap -= add_len as u16;
|
|
self.eull.tail = self.eull.tail.offset(add_len as u32);
|
|
buf = &buf[add_len..];
|
|
}
|
|
}
|
|
}
|
|
|
|
// The block size is 2^block_num, but max 2^15 = 32KB
|
|
// Initial size is 8 bytes (2^3), for the first block => block_num == 2
|
|
// Block size caps at 32KB (2^15) regardless of how high block_num goes
|
|
#[inline]
|
|
fn get_block_size(block_num: u32) -> u16 {
|
|
// Cap at 15 to prevent block sizes > 32KB
|
|
// block_num can now be much larger than 15, but block size maxes out
|
|
let exp: u32 = block_num.min(15u32);
|
|
(1u32 << exp) as u16
|
|
}
|
|
|
|
impl ExpUnrolledLinkedList {
|
|
#[inline(always)]
|
|
pub fn increment_num_blocks(&mut self) {
|
|
// Add overflow check as a safety measure
|
|
// With u32, we can handle up to ~4 billion blocks before overflow
|
|
// At 32KB per block (max size), that's 128 TB of data
|
|
self.block_num = self
|
|
.block_num
|
|
.checked_add(1)
|
|
.expect("ExpUnrolledLinkedList block count overflow - exceeded 4 billion blocks");
|
|
}
|
|
|
|
#[inline]
|
|
pub fn writer<'a>(&'a mut self, arena: &'a mut MemoryArena) -> ExpUnrolledLinkedListWriter<'a> {
|
|
ExpUnrolledLinkedListWriter { eull: self, arena }
|
|
}
|
|
|
|
pub fn read_to_end(&self, arena: &MemoryArena, output: &mut Vec<u8>) {
|
|
let mut addr = self.head;
|
|
if addr.is_null() {
|
|
return;
|
|
}
|
|
|
|
// Calculate last block length with bounds checking to prevent underflow
|
|
let block_size = get_block_size(self.block_num) as usize;
|
|
let last_block_len = block_size.saturating_sub(self.remaining_cap as usize);
|
|
|
|
// Safety check: if remaining_cap > block_size, the metadata is corrupted
|
|
assert!(
|
|
self.remaining_cap as usize <= block_size,
|
|
"ExpUnrolledLinkedList metadata corruption detected: remaining_cap ({}) > block_size \
|
|
({}). This indicates a serious bug, please report! (block_num={}, head={:?}, \
|
|
tail={:?})",
|
|
self.remaining_cap,
|
|
block_size,
|
|
self.block_num,
|
|
self.head,
|
|
self.tail
|
|
);
|
|
|
|
// Full Blocks (iterate through all blocks except the last one)
|
|
// Note: Blocks are numbered starting from FIRST_BLOCK_NUM+1 (=3) after first allocation
|
|
for block_num in FIRST_BLOCK_NUM + 1..self.block_num {
|
|
let cap = get_block_size(block_num) as usize;
|
|
let data = arena.slice(addr, cap);
|
|
output.extend_from_slice(data);
|
|
addr = arena.read(addr.offset(cap as u32));
|
|
}
|
|
// Last Block
|
|
let data = arena.slice(addr, last_block_len);
|
|
output.extend_from_slice(data);
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use common::{read_u32_vint, write_u32_vint};
|
|
|
|
use super::*;
|
|
|
|
#[test]
|
|
fn test_eull_empty() {
|
|
let arena = MemoryArena::default();
|
|
let stack = ExpUnrolledLinkedList::default();
|
|
{
|
|
let mut buffer = Vec::new();
|
|
stack.read_to_end(&arena, &mut buffer);
|
|
assert_eq!(&buffer[..], &[] as &[u8]);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull1() {
|
|
let mut arena = MemoryArena::default();
|
|
let mut stack = ExpUnrolledLinkedList::default();
|
|
stack.writer(&mut arena).extend_from_slice(&[1u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[2u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[3u8, 4u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[5u8]);
|
|
{
|
|
let mut buffer = Vec::new();
|
|
stack.read_to_end(&arena, &mut buffer);
|
|
assert_eq!(&buffer[..], &[1u8, 2u8, 3u8, 4u8, 5u8]);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull_vint1() {
|
|
let mut arena = MemoryArena::default();
|
|
let mut stack = ExpUnrolledLinkedList::default();
|
|
stack.writer(&mut arena).extend_from_slice(&[1u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[2u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[3u8, 4u8]);
|
|
stack.writer(&mut arena).extend_from_slice(&[5u8]);
|
|
{
|
|
let mut buffer = Vec::new();
|
|
stack.read_to_end(&arena, &mut buffer);
|
|
assert_eq!(&buffer[..], &[1u8, 2u8, 3u8, 4u8, 5u8]);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull_first_write_extends_cap() {
|
|
let mut arena = MemoryArena::default();
|
|
let mut stack = ExpUnrolledLinkedList::default();
|
|
stack
|
|
.writer(&mut arena)
|
|
.extend_from_slice(&[1u8, 2, 3, 4, 5, 6, 7, 8, 9]);
|
|
{
|
|
let mut buffer = Vec::new();
|
|
stack.read_to_end(&arena, &mut buffer);
|
|
assert_eq!(&buffer[..], &[1u8, 2, 3, 4, 5, 6, 7, 8, 9]);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull_long() {
|
|
let mut arena = MemoryArena::default();
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
let data: Vec<u32> = (0..100).collect();
|
|
for &el in &data {
|
|
eull.writer(&mut arena).write_u32_vint(el);
|
|
}
|
|
let mut buffer = Vec::new();
|
|
eull.read_to_end(&arena, &mut buffer);
|
|
let mut result = vec![];
|
|
let mut remaining = &buffer[..];
|
|
while !remaining.is_empty() {
|
|
result.push(read_u32_vint(&mut remaining));
|
|
}
|
|
assert_eq!(&result[..], &data[..]);
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull_limit() {
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
for _ in 0..100 {
|
|
eull.increment_num_blocks();
|
|
}
|
|
assert_eq!(get_block_size(eull.block_num), 1 << 15);
|
|
}
|
|
|
|
#[test]
|
|
fn test_eull_interlaced() {
|
|
let mut arena = MemoryArena::default();
|
|
let mut stack = ExpUnrolledLinkedList::default();
|
|
let mut stack2 = ExpUnrolledLinkedList::default();
|
|
|
|
let mut vec1: Vec<u8> = vec![];
|
|
let mut vec2: Vec<u8> = vec![];
|
|
|
|
for i in 0..9 {
|
|
stack.writer(&mut arena).write_u32_vint(i);
|
|
assert!(write_u32_vint(i, &mut vec1).is_ok());
|
|
if i % 2 == 0 {
|
|
stack2.writer(&mut arena).write_u32_vint(i);
|
|
assert!(write_u32_vint(i, &mut vec2).is_ok());
|
|
}
|
|
}
|
|
let mut res1 = vec![];
|
|
let mut res2 = vec![];
|
|
stack.read_to_end(&arena, &mut res1);
|
|
stack2.read_to_end(&arena, &mut res2);
|
|
assert_eq!(&vec1[..], &res1[..]);
|
|
assert_eq!(&vec2[..], &res2[..]);
|
|
}
|
|
|
|
// Tests for u32 block_num fix (issue with large arrays)
|
|
|
|
#[test]
|
|
fn test_block_num_exceeds_u16_max() {
|
|
// Test that we can handle more than 65,535 blocks (old u16 limit)
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
|
|
// Simulate allocating 70,000 blocks (exceeds u16::MAX of 65,535)
|
|
for _ in 0..70_000 {
|
|
eull.increment_num_blocks();
|
|
}
|
|
|
|
// Verify block_num is correct
|
|
assert_eq!(eull.block_num, FIRST_BLOCK_NUM + 70_000);
|
|
|
|
// Verify we can still get block size (should be capped at 32KB)
|
|
let block_size = get_block_size(eull.block_num);
|
|
assert_eq!(block_size, 1 << 15); // 32KB max
|
|
}
|
|
|
|
#[test]
|
|
#[allow(clippy::needless_range_loop)]
|
|
fn test_large_dataset_simulation() {
|
|
// Simulate the scenario: large arrays requiring many blocks
|
|
// We write enough data to require thousands of blocks
|
|
let mut arena = MemoryArena::default();
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
|
|
// Write 100 MB of data (this will require ~3,200 blocks at 32KB each)
|
|
// This is enough to validate the system works with large datasets
|
|
// but not so much that the test is slow
|
|
let bytes_per_write = 10_000;
|
|
let num_writes = 10_000; // 10k * 10k = 100 MB
|
|
|
|
let data: Vec<u8> = (0..bytes_per_write).map(|i| (i % 256) as u8).collect();
|
|
for _ in 0..num_writes {
|
|
eull.writer(&mut arena).extend_from_slice(&data);
|
|
}
|
|
|
|
// Verify we allocated many blocks (should be in the thousands)
|
|
assert!(
|
|
eull.block_num > 1000,
|
|
"block_num ({}) should be > 1000 for this much data",
|
|
eull.block_num
|
|
);
|
|
|
|
// Verify we can read back correctly
|
|
let mut buffer = Vec::new();
|
|
eull.read_to_end(&arena, &mut buffer);
|
|
assert_eq!(buffer.len(), bytes_per_write * num_writes);
|
|
|
|
// Verify data integrity on a sample
|
|
for i in 0..bytes_per_write {
|
|
assert_eq!(buffer[i], (i % 256) as u8);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_get_block_size_with_large_block_num() {
|
|
// Test that get_block_size handles large u32 values correctly
|
|
|
|
// Small block numbers (under 15)
|
|
assert_eq!(get_block_size(2), 4); // 2^2 = 4
|
|
assert_eq!(get_block_size(3), 8); // 2^3 = 8
|
|
assert_eq!(get_block_size(10), 1024); // 2^10 = 1KB
|
|
|
|
// At the cap (15)
|
|
assert_eq!(get_block_size(15), 32768); // 2^15 = 32KB
|
|
|
|
// Beyond the cap (should stay at 32KB)
|
|
assert_eq!(get_block_size(16), 32768);
|
|
assert_eq!(get_block_size(100), 32768);
|
|
assert_eq!(get_block_size(65_536), 32768); // Old u16::MAX + 1
|
|
assert_eq!(get_block_size(100_000), 32768);
|
|
assert_eq!(get_block_size(1_000_000), 32768);
|
|
}
|
|
|
|
#[test]
|
|
fn test_increment_blocks_near_u16_boundary() {
|
|
// Test incrementing around the old u16::MAX boundary
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
|
|
// Set to just before old limit
|
|
for _ in 0..65_533 {
|
|
eull.increment_num_blocks();
|
|
}
|
|
assert_eq!(eull.block_num, FIRST_BLOCK_NUM + 65_533);
|
|
|
|
// Cross the old u16::MAX boundary (this would have overflowed before)
|
|
eull.increment_num_blocks(); // 65,534
|
|
eull.increment_num_blocks(); // 65,535 (old max)
|
|
eull.increment_num_blocks(); // 65,536 (would overflow u16)
|
|
eull.increment_num_blocks(); // 65,537
|
|
|
|
// Verify we're past the old limit
|
|
assert_eq!(eull.block_num, FIRST_BLOCK_NUM + 65_537);
|
|
}
|
|
|
|
#[test]
|
|
fn test_write_and_read_with_many_blocks() {
|
|
// Test that write/read works correctly with many blocks
|
|
let mut arena = MemoryArena::default();
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
|
|
// Write data that will span many blocks
|
|
let test_data: Vec<u8> = (0..50_000).map(|i| (i % 256) as u8).collect();
|
|
eull.writer(&mut arena).extend_from_slice(&test_data);
|
|
|
|
// Read it back
|
|
let mut buffer = Vec::new();
|
|
eull.read_to_end(&arena, &mut buffer);
|
|
|
|
// Verify data integrity
|
|
assert_eq!(buffer.len(), test_data.len());
|
|
assert_eq!(&buffer[..], &test_data[..]);
|
|
}
|
|
|
|
#[test]
|
|
fn test_multiple_eull_with_large_block_counts() {
|
|
// Test multiple ExpUnrolledLinkedLists with high block counts
|
|
// (simulates parallel columnar writes)
|
|
let mut arena = MemoryArena::default();
|
|
let mut eull1 = ExpUnrolledLinkedList::default();
|
|
let mut eull2 = ExpUnrolledLinkedList::default();
|
|
|
|
// Write different data to each
|
|
for i in 0..10_000u32 {
|
|
eull1.writer(&mut arena).write_u32_vint(i);
|
|
eull2.writer(&mut arena).write_u32_vint(i * 2);
|
|
}
|
|
|
|
// Read back and verify
|
|
let mut buf1 = Vec::new();
|
|
let mut buf2 = Vec::new();
|
|
eull1.read_to_end(&arena, &mut buf1);
|
|
eull2.read_to_end(&arena, &mut buf2);
|
|
|
|
// Deserialize and check
|
|
let mut cursor1 = &buf1[..];
|
|
let mut cursor2 = &buf2[..];
|
|
for i in 0..10_000u32 {
|
|
assert_eq!(read_u32_vint(&mut cursor1), i);
|
|
assert_eq!(read_u32_vint(&mut cursor2), i * 2);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_block_size_stays_capped() {
|
|
// Verify that even with massive block numbers, size stays at 32KB
|
|
let mut eull = ExpUnrolledLinkedList::default();
|
|
|
|
// Increment to a very large number
|
|
for _ in 0..200_000 {
|
|
eull.increment_num_blocks();
|
|
}
|
|
|
|
let block_size = get_block_size(eull.block_num);
|
|
assert_eq!(block_size, 32768, "Block size should be capped at 32KB");
|
|
}
|
|
|
|
#[test]
|
|
#[should_panic(expected = "ExpUnrolledLinkedList block count overflow")]
|
|
fn test_increment_overflow_protection() {
|
|
// Test that we panic gracefully if we somehow hit u32::MAX
|
|
// This is extremely unlikely in practice (would require 128TB of data)
|
|
let mut eull = ExpUnrolledLinkedList {
|
|
block_num: u32::MAX,
|
|
..Default::default()
|
|
};
|
|
|
|
// This should panic with our custom error message
|
|
eull.increment_num_blocks();
|
|
}
|
|
}
|
|
|
|
#[cfg(all(test, feature = "unstable"))]
|
|
mod bench {
|
|
use std::iter;
|
|
|
|
use test::Bencher;
|
|
|
|
use super::super::MemoryArena;
|
|
use super::ExpUnrolledLinkedList;
|
|
|
|
const NUM_STACK: usize = 10_000;
|
|
const STACK_SIZE: u32 = 1000;
|
|
|
|
#[bench]
|
|
fn bench_push_vec(bench: &mut Bencher) {
|
|
bench.iter(|| {
|
|
let mut vecs = Vec::with_capacity(100);
|
|
for _ in 0..NUM_STACK {
|
|
vecs.push(Vec::new());
|
|
}
|
|
for s in 0..NUM_STACK {
|
|
for i in 0u32..STACK_SIZE {
|
|
let t = s * 392017 % NUM_STACK;
|
|
vecs[t].push(i);
|
|
}
|
|
}
|
|
});
|
|
}
|
|
|
|
#[bench]
|
|
fn bench_push_stack(bench: &mut Bencher) {
|
|
bench.iter(|| {
|
|
let mut arena = MemoryArena::default();
|
|
let mut stacks: Vec<ExpUnrolledLinkedList> =
|
|
iter::repeat_with(ExpUnrolledLinkedList::default)
|
|
.take(NUM_STACK)
|
|
.collect();
|
|
for s in 0..NUM_STACK {
|
|
for i in 0u32..STACK_SIZE {
|
|
let t = s * 392017 % NUM_STACK;
|
|
stacks[t]
|
|
.writer(&mut arena)
|
|
.extend_from_slice(&i.to_ne_bytes());
|
|
}
|
|
}
|
|
});
|
|
}
|
|
}
|