synor/crates/synor-storage/src/erasure.rs

//! Erasure Coding for fault-tolerant storage
//!
//! Uses Reed-Solomon coding to add redundancy to chunks.
//! Allows recovery of data even if some shards are lost.

use reed_solomon_erasure::galois_8::ReedSolomon;
use serde::{Deserialize, Serialize};
use crate::cid::ContentId;
use crate::error::{Error, Result};

/// Default number of data shards
pub const DEFAULT_DATA_SHARDS: usize = 10;
/// Default number of parity shards
pub const DEFAULT_PARITY_SHARDS: usize = 4;

/// Erasure coding configuration
#[derive(Debug, Clone)]
pub struct ErasureConfig {
    /// Number of data shards (original data pieces)
    pub data_shards: usize,
    /// Number of parity shards (redundancy pieces)
    pub parity_shards: usize,
}

impl Default for ErasureConfig {
    fn default() -> Self {
        Self {
            data_shards: DEFAULT_DATA_SHARDS,
            parity_shards: DEFAULT_PARITY_SHARDS,
        }
    }
}

impl ErasureConfig {
    /// Total number of shards
    pub fn total_shards(&self) -> usize {
        self.data_shards + self.parity_shards
    }

    /// Maximum shards that can be lost while still recovering
    pub fn fault_tolerance(&self) -> usize {
        self.parity_shards
    }
}

/// A single shard of encoded data
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Shard {
    /// Shard index (0..total_shards)
    pub index: usize,
    /// Whether this is a data shard (vs parity)
    pub is_data: bool,
    /// Shard content
    #[serde(with = "serde_bytes")]
    pub data: Vec<u8>,
    /// Hash of shard data
    pub hash: [u8; 32],
}

impl Shard {
    /// Create a new shard
    pub fn new(index: usize, is_data: bool, data: Vec<u8>) -> Self {
        let hash = *blake3::hash(&data).as_bytes();
        Self {
            index,
            is_data,
            data,
            hash,
        }
    }

    /// Verify shard integrity
    pub fn verify(&self) -> bool {
        let computed = *blake3::hash(&self.data).as_bytes();
        computed == self.hash
    }
}

/// Erasure-coded chunk
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct EncodedChunk {
    /// Original chunk CID
    pub chunk_cid: ContentId,
    /// Original data size (before padding)
    pub original_size: usize,
    /// Shard size (padded to be divisible by data_shards)
    pub shard_size: usize,
    /// Configuration used for encoding
    pub data_shards: usize,
    pub parity_shards: usize,
    /// All shards
    pub shards: Vec<Shard>,
}

/// Erasure encoder/decoder
pub struct ErasureCoder {
    config: ErasureConfig,
    rs: ReedSolomon,
}

impl ErasureCoder {
    /// Create a new erasure coder with default config
    pub fn new() -> Result<Self> {
        Self::with_config(ErasureConfig::default())
    }

    /// Create a new erasure coder with custom config
    pub fn with_config(config: ErasureConfig) -> Result<Self> {
        let rs = ReedSolomon::new(config.data_shards, config.parity_shards)
            .map_err(|e| Error::ErasureCoding(format!("Failed to create RS coder: {}", e)))?;
        
        Ok(Self { config, rs })
    }

    /// Encode data into shards with parity
    pub fn encode(&self, data: &[u8], chunk_cid: ContentId) -> Result<EncodedChunk> {
        let original_size = data.len();
        
        // Pad data to be divisible by data_shards
        let shard_size = data.len().div_ceil(self.config.data_shards);
        let padded_size = shard_size * self.config.data_shards;
        
        let mut padded_data = data.to_vec();
        padded_data.resize(padded_size, 0);

        // Split into data shards
        let mut shards: Vec<Vec<u8>> = padded_data
            .chunks(shard_size)
            .map(|c| c.to_vec())
            .collect();

        // Add parity shards (initially empty)
        for _ in 0..self.config.parity_shards {
            shards.push(vec![0u8; shard_size]);
        }

        // Encode parity
        self.rs.encode(&mut shards)
            .map_err(|e| Error::ErasureCoding(format!("Encoding failed: {}", e)))?;

        // Create shard structs
        let shard_structs: Vec<Shard> = shards
            .into_iter()
            .enumerate()
            .map(|(i, data)| {
                Shard::new(i, i < self.config.data_shards, data)
            })
            .collect();

        Ok(EncodedChunk {
            chunk_cid,
            original_size,
            shard_size,
            data_shards: self.config.data_shards,
            parity_shards: self.config.parity_shards,
            shards: shard_structs,
        })
    }

    /// Decode shards back to original data
    /// Some shards can be None (missing) as long as enough remain
    pub fn decode(&self, encoded: &EncodedChunk) -> Result<Vec<u8>> {
        let total = encoded.data_shards + encoded.parity_shards;
        
        // Prepare shards (Some for present, None for missing)
        let mut shards: Vec<Option<Vec<u8>>> = vec![None; total];
        let mut present_count = 0;

        for shard in &encoded.shards {
            if shard.index < total && shard.verify() {
                shards[shard.index] = Some(shard.data.clone());
                present_count += 1;
            }
        }

        // Check if we have enough shards
        if present_count < encoded.data_shards {
            return Err(Error::ErasureCoding(format!(
                "Not enough shards: have {}, need {}",
                present_count, encoded.data_shards
            )));
        }

        // Reconstruct missing shards
        self.rs.reconstruct(&mut shards)
            .map_err(|e| Error::ErasureCoding(format!("Reconstruction failed: {}", e)))?;

        // Combine data shards
        let mut result = Vec::with_capacity(encoded.original_size);
        for i in 0..encoded.data_shards {
            if let Some(ref shard_data) = shards[i] {
                result.extend_from_slice(shard_data);
            } else {
                return Err(Error::ErasureCoding("Reconstruction incomplete".into()));
            }
        }

        // Trim padding
        result.truncate(encoded.original_size);

        Ok(result)
    }
}

impl Default for ErasureCoder {
    fn default() -> Self {
        Self::new().expect("Default erasure config should work")
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_encode_decode() {
        let coder = ErasureCoder::new().unwrap();
        let data = b"Hello, erasure coding!";
        let cid = ContentId::from_content(data);
        
        let encoded = coder.encode(data, cid).unwrap();
        assert_eq!(encoded.shards.len(), 14); // 10 data + 4 parity
        
        let decoded = coder.decode(&encoded).unwrap();
        assert_eq!(decoded, data);
    }

    #[test]
    fn test_recovery_with_missing_shards() {
        let coder = ErasureCoder::new().unwrap();
        let data = b"Test data for recovery with some missing shards";
        let cid = ContentId::from_content(data);
        
        let mut encoded = coder.encode(data, cid).unwrap();
        
        // Remove 4 shards (max we can lose with 4 parity)
        encoded.shards.remove(0);
        encoded.shards.remove(2);
        encoded.shards.remove(4);
        encoded.shards.remove(6);
        
        let decoded = coder.decode(&encoded).unwrap();
        assert_eq!(decoded, data);
    }

    #[test]
    fn test_too_many_missing_shards() {
        let coder = ErasureCoder::new().unwrap();
        let data = b"Test data";
        let cid = ContentId::from_content(data);
        
        let mut encoded = coder.encode(data, cid).unwrap();
        
        // Remove 5 shards (more than 4 parity can handle)
        for _ in 0..5 {
            encoded.shards.remove(0);
        }
        
        let result = coder.decode(&encoded);
        assert!(result.is_err());
    }

    #[test]
    fn test_shard_verification() {
        let shard = Shard::new(0, true, b"test data".to_vec());
        assert!(shard.verify());
    }

    #[test]
    fn test_small_config() {
        let config = ErasureConfig {
            data_shards: 2,
            parity_shards: 1,
        };
        let coder = ErasureCoder::with_config(config).unwrap();
        
        let data = b"Small test";
        let cid = ContentId::from_content(data);
        
        let encoded = coder.encode(data, cid).unwrap();
        assert_eq!(encoded.shards.len(), 3);
        
        let decoded = coder.decode(&encoded).unwrap();
        assert_eq!(decoded, data);
    }
}