mini_lsm/mini-lsm/src/table.rs

mod builder;
mod iterator;

use std::fs::File;
use std::path::Path;
use std::sync::Arc;

use anyhow::{anyhow, Result};
pub use builder::SsTableBuilder;
use bytes::{Buf, BufMut, Bytes};
pub use iterator::SsTableIterator;

use crate::block::Block;
use crate::lsm_storage::BlockCache;

#[derive(Clone, Debug, PartialEq, Eq)]
pub struct BlockMeta {
    /// Offset of this data block.
    pub offset: usize,
    /// The first key of the data block.
    pub first_key: Bytes,
    /// The last key of the data block.
    pub last_key: Bytes,
}

impl BlockMeta {
    /// Encode block meta to a buffer.
    pub fn encode_block_meta(block_meta: &[BlockMeta], buf: &mut Vec<u8>) {
        let mut estimated_size = 0;
        for meta in block_meta {
            // The size of offset
            estimated_size += std::mem::size_of::<u32>();
            // The size of key length
            estimated_size += std::mem::size_of::<u16>();
            // The size of actual key
            estimated_size += meta.first_key.len();
            // The size of key length
            estimated_size += std::mem::size_of::<u16>();
            // The size of actual key
            estimated_size += meta.last_key.len();
        }
        // Reserve the space to improve performance, especially when the size of incoming data is
        // large
        buf.reserve(estimated_size);
        let original_len = buf.len();
        for meta in block_meta {
            buf.put_u32(meta.offset as u32);
            buf.put_u16(meta.first_key.len() as u16);
            buf.put_slice(&meta.first_key);
            buf.put_u16(meta.last_key.len() as u16);
            buf.put_slice(&meta.last_key);
        }
        assert_eq!(estimated_size, buf.len() - original_len);
    }

    /// Decode block meta from a buffer.
    pub fn decode_block_meta(mut buf: impl Buf) -> Vec<BlockMeta> {
        let mut block_meta = Vec::new();
        while buf.has_remaining() {
            let offset = buf.get_u32() as usize;
            let first_key_len = buf.get_u16() as usize;
            let first_key = buf.copy_to_bytes(first_key_len);
            let last_key_len = buf.get_u16() as usize;
            let last_key = buf.copy_to_bytes(last_key_len);
            block_meta.push(BlockMeta {
                offset,
                first_key,
                last_key,
            });
        }
        block_meta
    }
}

/// A file object.
///
/// Before day 4, it should look like:
pub struct FileObject(Option<File>, u64);

impl FileObject {
    pub fn read(&self, offset: u64, len: u64) -> Result<Vec<u8>> {
        use std::os::unix::fs::FileExt;
        let mut data = vec![0; len as usize];
        self.0
            .as_ref()
            .unwrap()
            .read_exact_at(&mut data[..], offset)?;
        Ok(data)
    }

    pub fn size(&self) -> u64 {
        self.1
    }

    /// Create a new file object (day 2) and write the file to the disk (day 4).
    pub fn create(path: &Path, data: Vec<u8>) -> Result<Self> {
        std::fs::write(path, &data)?;
        File::open(path)?.sync_all()?;
        Ok(FileObject(
            Some(File::options().read(true).write(false).open(path)?),
            data.len() as u64,
        ))
    }

    pub fn open(path: &Path) -> Result<Self> {
        let file = File::options().read(true).write(false).open(path)?;
        let size = file.metadata()?.len();
        Ok(FileObject(Some(file), size))
    }
}

pub struct SsTable {
    file: FileObject,
    block_metas: Vec<BlockMeta>,
    block_meta_offset: usize,
    id: usize,
    block_cache: Option<Arc<BlockCache>>,
    first_key: Bytes,
    last_key: Bytes,
}

impl SsTable {
    #[cfg(test)]
    pub(crate) fn open_for_test(file: FileObject) -> Result<Self> {
        Self::open(0, None, file)
    }

    /// Open SSTable from a file.
    pub fn open(id: usize, block_cache: Option<Arc<BlockCache>>, file: FileObject) -> Result<Self> {
        let len = file.size();
        let raw_meta_offset = file.read(len - 4, 4)?;
        let block_meta_offset = (&raw_meta_offset[..]).get_u32() as u64;
        let raw_meta = file.read(block_meta_offset, len - 4 - block_meta_offset)?;
        let block_metas = BlockMeta::decode_block_meta(&raw_meta[..]);
        Ok(Self {
            file,
            first_key: block_metas.first().unwrap().first_key.clone(),
            last_key: block_metas.last().unwrap().last_key.clone(),
            block_metas,
            block_meta_offset: block_meta_offset as usize,
            id,
            block_cache,
        })
    }

    /// Create a mock SST with only first key + last key metadata
    pub fn create_meta_only(id: usize, file_size: u64, first_key: Bytes, last_key: Bytes) -> Self {
        Self {
            file: FileObject(None, file_size),
            block_metas: vec![],
            block_meta_offset: 0,
            id,
            block_cache: None,
            first_key,
            last_key,
        }
    }

    /// Read a block from the disk.
    pub fn read_block(&self, block_idx: usize) -> Result<Arc<Block>> {
        let offset = self.block_metas[block_idx].offset;
        let offset_end = self
            .block_metas
            .get(block_idx + 1)
            .map_or(self.block_meta_offset, |x| x.offset);
        let block_data = self
            .file
            .read(offset as u64, (offset_end - offset) as u64)?;
        Ok(Arc::new(Block::decode(&block_data[..])))
    }

    /// Read a block from disk, with block cache.
    pub fn read_block_cached(&self, block_idx: usize) -> Result<Arc<Block>> {
        if let Some(ref block_cache) = self.block_cache {
            let blk = block_cache
                .try_get_with((self.id, block_idx), || self.read_block(block_idx))
                .map_err(|e| anyhow!("{}", e))?;
            Ok(blk)
        } else {
            self.read_block(block_idx)
        }
    }

    /// Find the block that may contain `key`.
    pub fn find_block_idx(&self, key: &[u8]) -> usize {
        self.block_metas
            .partition_point(|meta| meta.first_key <= key)
            .saturating_sub(1)
    }

    /// Get number of data blocks.
    pub fn num_of_blocks(&self) -> usize {
        self.block_metas.len()
    }

    pub fn first_key(&self) -> &Bytes {
        &self.first_key
    }

    pub fn last_key(&self) -> &Bytes {
        &self.last_key
    }

    pub fn table_size(&self) -> u64 {
        self.file.1
    }

    pub fn sst_id(&self) -> usize {
        self.id
    }
}

#[cfg(test)]
mod tests;