Add RocksDB Preset System and WAL Directory Support for HDD Archive Nodes (and benefiting regular nodes)

consensus/src/consensus/factory.rs

@@ -9,7 +9,8 @@ use kaspa_consensusmanager::{ConsensusFactory, ConsensusInstance, DynConsensusCt
 use kaspa_core::{debug, time::unix_now, warn};
 use kaspa_database::{
     prelude::{
-        BatchDbWriter, CachePolicy, CachedDbAccess, CachedDbItem, DirectDbWriter, StoreError, StoreResult, StoreResultExtensions, DB,
+        BatchDbWriter, CachePolicy, CachedDbAccess, CachedDbItem, DirectDbWriter, RocksDbPreset, StoreError, StoreResult,
+        StoreResultExtensions, DB,
     },
     registry::DatabaseStorePrefixes,
 };
@@ -255,6 +256,8 @@ pub struct Factory {
     tx_script_cache_counters: Arc<TxScriptCacheCounters>,
     fd_budget: i32,
     mining_rules: Arc<MiningRules>,
+    rocksdb_preset: RocksDbPreset,
+    wal_dir: Option<PathBuf>,
 }
 
 impl Factory {
@@ -268,6 +271,8 @@ impl Factory {
         tx_script_cache_counters: Arc<TxScriptCacheCounters>,
         fd_budget: i32,
         mining_rules: Arc<MiningRules>,
+        rocksdb_preset: RocksDbPreset,
+        wal_dir: Option<PathBuf>,
     ) -> Self {
         assert!(fd_budget > 0, "fd_budget has to be positive");
         let mut config = config.clone();
@@ -286,6 +291,8 @@ impl Factory {
             tx_script_cache_counters,
             fd_budget,
             mining_rules,
+            rocksdb_preset,
+            wal_dir,
         };
         factory.delete_inactive_consensus_entries();
         factory
@@ -316,6 +323,8 @@ impl ConsensusFactory for Factory {
             .with_db_path(dir)
             .with_parallelism(self.db_parallelism)
             .with_files_limit(self.fd_budget / 2) // active and staging consensuses should have equal budgets
+            .with_preset(self.rocksdb_preset)
+            .with_wal_dir(self.wal_dir.clone())
             .build()
             .unwrap();
 
@@ -351,6 +360,8 @@ impl ConsensusFactory for Factory {
             .with_db_path(dir)
             .with_parallelism(self.db_parallelism)
             .with_files_limit(self.fd_budget / 2) // active and staging consensuses should have equal budgets
+            .with_preset(self.rocksdb_preset)
+            .with_wal_dir(self.wal_dir.clone())
             .build()
             .unwrap();
 

database/src/db.rs

@@ -4,8 +4,10 @@ use std::path::PathBuf;
 
 pub use conn_builder::ConnBuilder;
 use kaspa_utils::fd_budget::FDGuard;
+pub use rocksdb_preset::RocksDbPreset;
 
 mod conn_builder;
+mod rocksdb_preset;
 
 /// The DB type used for Kaspad stores
 pub struct DB {

database/src/db/conn_builder.rs

@@ -1,3 +1,4 @@
+use super::rocksdb_preset::RocksDbPreset;
 use crate::db::DB;
 use rocksdb::{DBWithThreadMode, MultiThreaded};
 use std::{path::PathBuf, sync::Arc};
@@ -13,6 +14,8 @@ pub struct ConnBuilder<Path, const STATS_ENABLED: bool, StatsPeriod, FDLimit> {
     files_limit: FDLimit,
     mem_budget: usize,
     stats_period: StatsPeriod,
+    preset: RocksDbPreset,
+    wal_dir: Option<PathBuf>,
 }
 
 impl Default for ConnBuilder<Unspecified, false, Unspecified, Unspecified> {
@@ -24,6 +27,8 @@ impl Default for ConnBuilder<Unspecified, false, Unspecified, Unspecified> {
             mem_budget: 64 * 1024 * 1024,
             stats_period: Unspecified,
             files_limit: Unspecified,
+            preset: RocksDbPreset::Default,
+            wal_dir: None,
         }
     }
 }
@@ -37,6 +42,8 @@ impl<Path, const STATS_ENABLED: bool, StatsPeriod, FDLimit> ConnBuilder<Path, ST
             parallelism: self.parallelism,
             mem_budget: self.mem_budget,
             stats_period: self.stats_period,
+            preset: self.preset,
+            wal_dir: self.wal_dir,
         }
     }
     pub fn with_create_if_missing(self, create_if_missing: bool) -> ConnBuilder<Path, STATS_ENABLED, StatsPeriod, FDLimit> {
@@ -56,8 +63,16 @@ impl<Path, const STATS_ENABLED: bool, StatsPeriod, FDLimit> ConnBuilder<Path, ST
             parallelism: self.parallelism,
             mem_budget: self.mem_budget,
             stats_period: self.stats_period,
+            preset: self.preset,
+            wal_dir: self.wal_dir,
         }
     }
+    pub fn with_preset(self, preset: RocksDbPreset) -> ConnBuilder<Path, STATS_ENABLED, StatsPeriod, FDLimit> {
+        ConnBuilder { preset, ..self }
+    }
+    pub fn with_wal_dir(self, wal_dir: Option<PathBuf>) -> ConnBuilder<Path, STATS_ENABLED, StatsPeriod, FDLimit> {
+        ConnBuilder { wal_dir, ..self }
+    }
 }
 
 impl<Path, FDLimit> ConnBuilder<Path, false, Unspecified, FDLimit> {
@@ -69,6 +84,8 @@ impl<Path, FDLimit> ConnBuilder<Path, false, Unspecified, FDLimit> {
             files_limit: self.files_limit,
             mem_budget: self.mem_budget,
             stats_period: self.stats_period,
+            preset: self.preset,
+            wal_dir: self.wal_dir,
         }
     }
 }
@@ -82,6 +99,8 @@ impl<Path, StatsPeriod, FDLimit> ConnBuilder<Path, true, StatsPeriod, FDLimit> {
             files_limit: self.files_limit,
             mem_budget: self.mem_budget,
             stats_period: Unspecified,
+            preset: self.preset,
+            wal_dir: self.wal_dir,
         }
     }
     pub fn with_stats_period(self, stats_period: impl Into<u32>) -> ConnBuilder<Path, true, u32, FDLimit> {
@@ -92,18 +111,39 @@ impl<Path, StatsPeriod, FDLimit> ConnBuilder<Path, true, StatsPeriod, FDLimit> {
             files_limit: self.files_limit,
             mem_budget: self.mem_budget,
             stats_period: stats_period.into(),
+            preset: self.preset,
+            wal_dir: self.wal_dir,
         }
     }
 }
 
 macro_rules! default_opts {
     ($self: expr) => {{
         let mut opts = rocksdb::Options::default();
-        if $self.parallelism > 1 {
-            opts.increase_parallelism($self.parallelism as i32);
+
+        // Apply the preset configuration (includes parallelism and compaction settings)
+        $self.preset.apply_to_options(&mut opts, $self.parallelism, $self.mem_budget);
+
+        // Configure WAL directory if specified (for RAM cache / tmpfs)
+        // Auto-generate unique subdirectory from database path to avoid conflicts
+        if let Some(ref wal_base) = $self.wal_dir {
+            let db_name = $self
+                .db_path
+                .file_name()
+                .and_then(|n| n.to_str())
+                .expect(&format!("Invalid database path: {}", $self.db_path.display()));
+            let wal_subdir = wal_base.join(db_name);
+
+            // Create subdirectory if needed (each DB gets its own WAL space)
+            std::fs::create_dir_all(&wal_subdir).expect(&format!(
+                "Failed to create WAL subdirectory {}: {}",
+                wal_subdir.display(),
+                "error"
+            ));
+
+            opts.set_wal_dir(&wal_subdir);
         }
 
-        opts.optimize_level_style_compaction($self.mem_budget);
         let guard = kaspa_utils::fd_budget::acquire_guard($self.files_limit)?;
         opts.set_max_open_files($self.files_limit);
         opts.create_if_missing($self.create_if_missing);

database/src/db/rocksdb_preset.rs

@@ -0,0 +1,223 @@
+//! RocksDB configuration presets for different use cases
+//!
+//! This module provides pre-configured RocksDB option sets optimized for different
+//! deployment scenarios. Based on Issue #681 and community testing.
+
+use rocksdb::Options;
+use std::str::FromStr;
+
+/// Available RocksDB configuration presets
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub enum RocksDbPreset {
+    /// Default configuration - balanced for general use on SSD/NVMe
+    /// - 64MB write buffer
+    /// - Standard compression
+    /// - Optimized for fast storage
+    #[default]
+    Default,
+
+    /// Archive configuration - optimized for HDD storage
+    /// - 256MB write buffer (4x default)
+    /// - Aggressive compression (LZ4 + ZSTD)
+    /// - BlobDB enabled for large values
+    /// - Rate limiting to prevent I/O spikes
+    /// - Optimized for sequential writes and reduced write amplification
+    ///
+    /// Based on Callidon's configuration from Issue #681.
+    /// Recommended for archival nodes on HDD storage.
+    Archive,
+}
+
+impl FromStr for RocksDbPreset {
+    type Err = String;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s.to_lowercase().as_str() {
+            "default" => Ok(Self::Default),
+            "archive" => Ok(Self::Archive),
+            _ => Err(format!("Unknown RocksDB preset: '{}'. Valid options: default, archive", s)),
+        }
+    }
+}
+
+impl std::fmt::Display for RocksDbPreset {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::Default => write!(f, "default"),
+            Self::Archive => write!(f, "archive"),
+        }
+    }
+}
+
+impl RocksDbPreset {
+    /// Apply the preset configuration to RocksDB options
+    ///
+    /// # Arguments
+    /// * `opts` - RocksDB options to configure
+    /// * `parallelism` - Number of background threads
+    /// * `mem_budget` - Memory budget (only used for Default preset, Archive uses fixed 256MB)
+    pub fn apply_to_options(&self, opts: &mut Options, parallelism: usize, mem_budget: usize) {
+        match self {
+            Self::Default => self.apply_default(opts, parallelism, mem_budget),
+            Self::Archive => self.apply_archive(opts, parallelism),
+        }
+    }
+
+    /// Apply default preset configuration
+    fn apply_default(&self, opts: &mut Options, parallelism: usize, mem_budget: usize) {
+        if parallelism > 1 {
+            opts.increase_parallelism(parallelism as i32);
+        }
+
+        // Use the provided memory budget (typically 64MB)
+        opts.optimize_level_style_compaction(mem_budget);
+    }
+
+    /// Apply archive preset configuration (Callidon's HDD-optimized settings)
+    fn apply_archive(&self, opts: &mut Options, parallelism: usize) {
+        if parallelism > 1 {
+            opts.increase_parallelism(parallelism as i32);
+        }
+
+        // Memory and write buffer settings (256MB for better batching on HDD)
+        let write_buffer_size = 256 * 1024 * 1024; // 256MB
+
+        // Optimize for level-style compaction with archive-appropriate memory
+        // This sets up LSM tree parameters
+        opts.optimize_level_style_compaction(write_buffer_size);
+
+        // Re-set write_buffer_size after optimize_level_style_compaction()
+        // because optimize_level_style_compaction() internally overrides it to size/4
+        opts.set_write_buffer_size(write_buffer_size);
+
+        // LSM Tree Structure - Optimized for large (4TB+) archives
+        // 256 MB SST files reduce file count dramatically (500K → 16K files for 4TB)
+        opts.set_target_file_size_base(256 * 1024 * 1024); // 256 MB SST files
+        opts.set_target_file_size_multiplier(1); // Same size across all levels
+        opts.set_max_bytes_for_level_base(1024 * 1024 * 1024); // 1 GB L1 base
+        opts.set_level_compaction_dynamic_level_bytes(true); // Minimize space amplification
+
+        // Compaction settings
+        // Trigger compaction when L0 has just 1 file (minimize write amplification)
+        opts.set_level_zero_file_num_compaction_trigger(1);
+
+        // Prioritize compacting older/smaller files first
+        use rocksdb::CompactionPri;
+        opts.set_compaction_pri(CompactionPri::OldestSmallestSeqFirst);
+
+        // Read-ahead for compactions (4MB - good for sequential HDD reads)
+        opts.set_compaction_readahead_size(4 * 1024 * 1024);
+
+        // Compression strategy: LZ4 for all levels, ZSTD for bottommost
+        use rocksdb::DBCompressionType;
+
+        // Set default compression to LZ4 (fast)
+        opts.set_compression_type(DBCompressionType::Lz4);
+
+        // Enable bottommost level compression with maximum ZSTD level
+        opts.set_bottommost_compression_type(DBCompressionType::Zstd);
+
+        // ZSTD compression options for bottommost level
+        // Larger dictionaries (64 KB) improve compression on large archives
+        opts.set_compression_options(
+            -1,        // window_bits (let ZSTD choose optimal)
+            22,        // level (maximum compression)
+            0,         // strategy (default)
+            64 * 1024, // dict_bytes (64 KB dictionary)
+        );
+
+        // Train ZSTD dictionaries on 8 MB of sample data (~125x dictionary size)
+        opts.set_zstd_max_train_bytes(8 * 1024 * 1024);
+
+        // Block-based table options for better caching
+        use rocksdb::{BlockBasedOptions, Cache};
+        let mut block_opts = BlockBasedOptions::default();
+
+        // Partitioned Bloom filters (18 bits per key for better false-positive rate)
+        block_opts.set_bloom_filter(18.0, false); // 18 bits per key
+        block_opts.set_partition_filters(true); // Partition for large databases
+        block_opts.set_format_version(5); // Latest format with optimizations
+        block_opts.set_index_type(rocksdb::BlockBasedIndexType::TwoLevelIndexSearch);
+
+        // Cache index and filter blocks in block cache for faster queries
+        block_opts.set_cache_index_and_filter_blocks(true);
+
+        // Block cache (2GB LRU cache for frequently accessed blocks)
+        let cache = Cache::new_lru_cache(2 * 1024 * 1024 * 1024); // 2GB
+        block_opts.set_block_cache(&cache);
+
+        // Set block size (256KB - better for sequential HDD reads)
+        block_opts.set_block_size(256 * 1024);
+
+        opts.set_block_based_table_factory(&block_opts);
+
+        // Rate limiting: prevent I/O spikes on HDD
+        // 12 MB/s rate limit for background writes
+        opts.set_ratelimiter(12 * 1024 * 1024, 100_000, 10);
+
+        // Enable BlobDB for large values (reduces write amplification)
+        opts.set_enable_blob_files(true);
+        opts.set_min_blob_size(512); // Only values >512 bytes go to blob files
+        opts.set_blob_file_size(256 * 1024 * 1024); // 256MB blob files
+        opts.set_blob_compression_type(DBCompressionType::Zstd); // Compress blobs
+        opts.set_enable_blob_gc(true); // Enable garbage collection
+        opts.set_blob_gc_age_cutoff(0.9); // GC blobs when 90% old
+        opts.set_blob_gc_force_threshold(0.1); // Force GC at 10% garbage
+        opts.set_blob_compaction_readahead_size(8 * 1024 * 1024); // 8 MB blob readahead
+    }
+
+    /// Get a human-readable description of the preset
+    pub fn description(&self) -> &'static str {
+        match self {
+            Self::Default => "Default preset - balanced for SSD/NVMe (64MB write buffer, standard compression)",
+            Self::Archive => "Archive preset - optimized for HDD (256MB write buffer, BlobDB, aggressive compression, rate limiting)",
+        }
+    }
+
+    /// Get the recommended use case for this preset
+    pub fn use_case(&self) -> &'static str {
+        match self {
+            Self::Default => "General purpose nodes on SSD/NVMe storage",
+            Self::Archive => "Archival nodes on HDD storage (--archival flag recommended)",
+        }
+    }
+
+    /// Get memory requirements for this preset
+    pub fn memory_requirements(&self) -> &'static str {
+        match self {
+            Self::Default => "~4GB minimum, scales with --ram-scale",
+            Self::Archive => "~8GB minimum (256MB write buffer + 2GB cache + overhead), 16GB+ recommended",
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_preset_from_str() {
+        assert_eq!(RocksDbPreset::from_str("default").unwrap(), RocksDbPreset::Default);
+        assert_eq!(RocksDbPreset::from_str("Default").unwrap(), RocksDbPreset::Default);
+        assert_eq!(RocksDbPreset::from_str("archive").unwrap(), RocksDbPreset::Archive);
+        assert_eq!(RocksDbPreset::from_str("ARCHIVE").unwrap(), RocksDbPreset::Archive);
+        assert!(RocksDbPreset::from_str("unknown").is_err());
+    }
+
+    #[test]
+    fn test_preset_display() {
+        assert_eq!(RocksDbPreset::Default.to_string(), "default");
+        assert_eq!(RocksDbPreset::Archive.to_string(), "archive");
+    }
+
+    #[test]
+    fn test_apply_presets() {
+        let mut opts = Options::default();
+
+        // Test default preset
+        RocksDbPreset::Default.apply_to_options(&mut opts, 4, 64 * 1024 * 1024);
+
+        // Test archive preset
+        RocksDbPreset::Archive.apply_to_options(&mut opts, 4, 64 * 1024 * 1024);
+    }
+}