feat(meta): do not split small table from default group

src/common/src/config.rs

@@ -299,6 +299,9 @@ pub struct MetaConfig {
     #[serde(default = "default::meta::split_group_size_limit")]
     pub split_group_size_limit: u64,
 
+    #[serde(default = "default::meta::max_group_size")]
+    pub max_group_size: u64,
+
     #[serde(default = "default::meta::cut_table_size_limit")]
     pub cut_table_size_limit: u64,
 
@@ -1262,11 +1265,15 @@ pub mod default {
         }
 
         pub fn move_table_size_limit() -> u64 {
-            10 * 1024 * 1024 * 1024 // 10GB
+            192 * 1024 * 1024 * 1024 // 192GB
         }
 
         pub fn split_group_size_limit() -> u64 {
-            64 * 1024 * 1024 * 1024 // 64GB
+            256 * 1024 * 1024 * 1024 // 256GB
+        }
+
+        pub fn max_group_size() -> u64 {
+            1024 * 1024 * 1024 * 1024 // 1TB
         }
 
         pub fn partition_vnode_count() -> u32 {
@@ -1796,7 +1803,7 @@ pub mod default {
         const DEFAULT_MAX_SPACE_RECLAIM_BYTES: u64 = 512 * 1024 * 1024; // 512MB;
         const DEFAULT_LEVEL0_STOP_WRITE_THRESHOLD_SUB_LEVEL_NUMBER: u64 = 300;
         const DEFAULT_MAX_COMPACTION_FILE_COUNT: u64 = 100;
-        const DEFAULT_MIN_SUB_LEVEL_COMPACT_LEVEL_COUNT: u32 = 3;
+        const DEFAULT_MIN_SUB_LEVEL_COMPACT_LEVEL_COUNT: u32 = 4;
         const DEFAULT_MIN_OVERLAPPING_SUB_LEVEL_COMPACT_LEVEL_COUNT: u32 = 12;
         const DEFAULT_TOMBSTONE_RATIO_PERCENT: u32 = 40;
         const DEFAULT_EMERGENCY_PICKER: bool = true;

src/config/docs.md

@@ -41,12 +41,13 @@ This page is automatically generated by `./risedev generate-example-config`
 | full_gc_interval_sec | Interval of automatic hummock full GC. | 86400 |
 | hummock_version_checkpoint_interval_sec | Interval of hummock version checkpoint. | 30 |
 | hybrid_partition_vnode_count | Count of partitions of tables in default group and materialized view group. The meta node will decide according to some strategy whether to cut the boundaries of the file according to the vnode alignment. Each partition contains aligned data of `VirtualNode::COUNT / hybrid_partition_vnode_count` consecutive virtual-nodes of one state table. Set it zero to disable this feature. | 4 |
+| max_group_size |  | 1099511627776 |
 | max_heartbeat_interval_secs | Maximum allowed heartbeat interval in seconds. | 60 |
 | meta_leader_lease_secs |  | 30 |
 | min_delta_log_num_for_hummock_version_checkpoint | The minimum delta log number a new checkpoint should compact, otherwise the checkpoint attempt is rejected. | 10 |
 | min_sst_retention_time_sec | Objects within `min_sst_retention_time_sec` won't be deleted by hummock full GC, even they are dangling. | 86400 |
 | min_table_split_write_throughput | If the size of one table is smaller than `min_table_split_write_throughput`, we would not split it to an single group. | 4194304 |
-| move_table_size_limit |  | 10737418240 |
+| move_table_size_limit |  | 137438953472 |
 | node_num_monitor_interval_sec |  | 10 |
 | parallelism_control_batch_size | The number of streaming jobs per scaling operation. | 10 |
 | parallelism_control_trigger_first_delay_sec | The first delay of parallelism control. | 30 |

src/config/example.toml

@@ -39,8 +39,9 @@ periodic_space_reclaim_compaction_interval_sec = 3600
 periodic_ttl_reclaim_compaction_interval_sec = 1800
 periodic_tombstone_reclaim_compaction_interval_sec = 600
 periodic_split_compact_group_interval_sec = 10
-move_table_size_limit = 10737418240
-split_group_size_limit = 68719476736
+move_table_size_limit = 137438953472
+split_group_size_limit = 274877906944
+max_group_size = 1099511627776
 cut_table_size_limit = 1073741824
 do_not_config_object_storage_lifecycle = false
 partition_vnode_count = 16

src/meta/node/src/lib.rs

@@ -339,6 +339,7 @@ pub fn start(opts: MetaNodeOpts) -> Pin<Box<dyn Future<Output = ()> + Send>> {
                     .meta
                     .periodic_split_compact_group_interval_sec,
                 split_group_size_limit: config.meta.split_group_size_limit,
+                max_group_size: config.meta.max_group_size,
                 min_table_split_size: config.meta.move_table_size_limit,
                 table_write_throughput_threshold: config.meta.table_write_throughput_threshold,
                 min_table_split_write_throughput: config.meta.min_table_split_write_throughput,

src/meta/src/hummock/compaction/picker/base_level_compaction_picker.rs

@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::cell::RefCell;
 use std::sync::Arc;
 
 use itertools::Itertools;
@@ -29,10 +28,6 @@ use crate::hummock::compaction::picker::TrivialMovePicker;
 use crate::hummock::compaction::{create_overlap_strategy, CompactionDeveloperConfig};
 use crate::hummock::level_handler::LevelHandler;
 
-std::thread_local! {
-    static LOG_COUNTER: RefCell<usize> = const { RefCell::new(0) };
-}
-
 pub struct LevelCompactionPicker {
     target_level: usize,
     config: Arc<CompactionConfig>,
@@ -159,7 +154,7 @@ impl LevelCompactionPicker {
             // half quota.
             std::cmp::max(
                 self.config.max_bytes_for_level_base,
-                self.config.max_compaction_bytes / 2,
+                self.config.max_compaction_bytes,
             ),
             1,
             // The maximum number of sub_level compact level per task
@@ -257,23 +252,6 @@ impl LevelCompactionPicker {
                 ValidationRuleType::ToBase,
                 stats,
             ) {
-                if l0.total_file_size > target_level.total_file_size * 8 {
-                    let log_counter = LOG_COUNTER.with_borrow_mut(|counter| {
-                        *counter += 1;
-                        *counter
-                    });
-
-                    // reduce log
-                    if log_counter % 100 == 0 {
-                        tracing::warn!("skip task with level count: {}, file count: {}, select size: {}, target size: {}, target level size: {}",
-                            result.input_levels.len(),
-                            result.total_file_count,
-                            result.select_input_size,
-                            result.target_input_size,
-                            target_level.total_file_size,
-                        );
-                    }
-                }
                 continue;
             }
 

src/meta/src/hummock/compaction/picker/compaction_task_validator.rs

@@ -179,17 +179,11 @@ impl CompactionTaskValidationRule for BaseCompactionTaskValidationRule {
     fn validate(&self, input: &CompactionInput, stats: &mut LocalPickerStatistic) -> bool {
         if input.total_file_count >= self.config.level0_max_compact_file_number
             || input.input_levels.len() >= MAX_COMPACT_LEVEL_COUNT
+            || input.select_input_size >= self.config.max_compaction_bytes
         {
             return true;
         }
 
-        // The size of target level may be too large, we shall skip this compact task and wait
-        //  the data in base level compact to lower level.
-        if input.target_input_size > self.config.max_compaction_bytes {
-            stats.skip_by_count_limit += 1;
-            return false;
-        }
-
         if input.select_input_size < input.target_input_size {
             stats.skip_by_write_amp_limit += 1;
             return false;

src/meta/src/hummock/compaction/picker/intra_compaction_picker.rs

@@ -159,16 +159,6 @@ impl IntraCompactionPicker {
             let mut select_input_size = 0;
             let mut total_file_count = 0;
             for input in l0_select_tables_vec {
-                let mut max_level_size = 0;
-                for level_select_table in &input.sstable_infos {
-                    let level_select_size = level_select_table
-                        .iter()
-                        .map(|sst| sst.file_size)
-                        .sum::<u64>();
-
-                    max_level_size = std::cmp::max(max_level_size, level_select_size);
-                }
-
                 let mut select_level_inputs = Vec::with_capacity(input.sstable_infos.len());
                 for level_select_sst in input.sstable_infos {
                     if level_select_sst.is_empty() {
@@ -239,6 +229,8 @@ impl IntraCompactionPicker {
                 continue;
             }
 
+            assert!(l0.sub_levels[idx + 1].sub_level_id != l0.sub_levels[idx].sub_level_id);
+
             let trivial_move_picker = TrivialMovePicker::new(0, 0, overlap_strategy.clone());
 
             let select_sst = trivial_move_picker.pick_trivial_move_sst(

src/meta/src/hummock/compaction/picker/tier_compaction_picker.rs

@@ -124,9 +124,9 @@ impl TierCompactionPicker {
                 target_level: 0,
                 target_sub_level_id: level.sub_level_id,
                 select_input_size: compaction_bytes,
-                target_input_size: 0,
                 total_file_count: compact_file_count,
                 vnode_partition_count,
+                ..Default::default()
             };
 
             if !self.compaction_task_validator.valid_compact_task(