lz4+zstd compression

CHANGELOG.md

@@ -7,14 +7,18 @@
 
 * Enable access key nonce range for implicit accounts to prevent tx hash collisions [#5482](https://github.com/near/nearcore/pull/5482)
 
+### Non-protocol Changes
+
+* Switch to LZ4+ZSTD compression from Snappy in RocksDB [#6365](https://github.com/near/nearcore/pull/6365)
+
 ## `1.23.0` [13-12-2021]
 
 ### Protocol Changes
 
 * Further lower regular_op_cost from 2_207_874 to 822_756.
 * Limit number of wasm functions in one contract to 10_000. [#4954](https://github.com/near/nearcore/pull/4954)
 * Add block header v3, required by new validator selection algorithm
-* Move to new validator selection and sampling algorithm. Now we would be able to use all available seats. First step to enable chunk only producers. 
+* Move to new validator selection and sampling algorithm. Now we would be able to use all available seats. First step to enable chunk only producers.
 
 ### Non-protocol Changes
 

core/store/src/db.rs

@@ -759,10 +759,32 @@ impl Database for TestDB {
     }
 }
 
+fn set_compression_options(opts: &mut Options) {
+    opts.set_compression_type(rocksdb::DBCompressionType::Lz4);
+    opts.set_bottommost_compression_type(rocksdb::DBCompressionType::Zstd);
+    // RocksDB documenation says that 16KB is a typical dictionary size.
+    // We've empirically tuned the dicionary size to twice of that 'typical' size.
+    // Having train data size x100 from dictionary size is a recommendation from RocksDB.
+    // See: https://rocksdb.org/blog/2021/05/31/dictionary-compression.html?utm_source=dbplatz
+    let dict_size = 2 * 16384;
+    let max_train_bytes = dict_size * 100;
+    // We use default parameters of RocksDB here:
+    //      window_bits is -14 and is unused (Zlib-specific parameter),
+    //      compression_level is 32767 meaning the default compression level for ZSTD,
+    //      compression_strategy is 0 and is unused (Zlib-specific parameter).
+    // See: https://github.com/facebook/rocksdb/blob/main/include/rocksdb/advanced_options.h#L176:
+    opts.set_bottommost_compression_options(
+        /*window_bits */ -14, /*compression_level */ 32767,
+        /*compression_strategy */ 0, dict_size, /*enabled */ true,
+    );
+    opts.set_bottommost_zstd_max_train_bytes(max_train_bytes, true);
+}
+
 /// DB level options
 fn rocksdb_options() -> Options {
     let mut opts = Options::default();
 
+    set_compression_options(&mut opts);
     opts.create_missing_column_families(true);
     opts.create_if_missing(true);
     opts.set_use_fsync(false);
@@ -817,12 +839,12 @@ fn choose_cache_size(col: DBCol) -> usize {
 
 fn rocksdb_column_options(col: DBCol) -> Options {
     let mut opts = Options::default();
+    set_compression_options(&mut opts);
     opts.set_level_compaction_dynamic_level_bytes(true);
     let cache_size = choose_cache_size(col);
     opts.set_block_based_table_factory(&rocksdb_block_based_options(cache_size));
     opts.optimize_level_style_compaction(128 * bytesize::MIB as usize);
     opts.set_target_file_size_base(64 * bytesize::MIB);
-    opts.set_compression_per_level(&[]);
     if col.is_rc() {
         opts.set_merge_operator("refcount merge", RocksDB::refcount_merge, RocksDB::refcount_merge);
         opts.set_compaction_filter("empty value filter", RocksDB::empty_value_compaction_filter);