risingwave_meta/hummock/manager/compaction/

compaction_group_manager.rs

// Copyright 2024 RisingWave Labs
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use std::collections::{BTreeMap, HashMap, HashSet};
use std::ops::DerefMut;
use std::sync::Arc;

use itertools::Itertools;
use risingwave_common::catalog::TableId;
use risingwave_common::config::meta::default::compaction_config as default_compaction_config;
use risingwave_common::util::epoch::INVALID_EPOCH;
use risingwave_hummock_sdk::CompactionGroupId;
use risingwave_hummock_sdk::compaction_group::StaticCompactionGroupId;
use risingwave_hummock_sdk::compaction_group::hummock_version_ext::get_compaction_group_ids;
use risingwave_hummock_sdk::version::GroupDelta;
use risingwave_meta_model::compaction_config;
use risingwave_pb::hummock::rise_ctl_update_compaction_config_request::mutable_config::MutableConfig;
use risingwave_pb::hummock::write_limits::WriteLimit;
use risingwave_pb::hummock::{
    CompactionConfig, CompactionGroupInfo, PbGroupConstruct, PbGroupDestroy, PbStateTableInfoDelta,
};
use sea_orm::EntityTrait;
use tokio::sync::OnceCell;

use super::CompactionGroupStatistic;
use crate::hummock::compaction::compaction_config::{
    CompactionConfigBuilder, validate_compaction_config,
};
use crate::hummock::error::{Error, Result};
use crate::hummock::manager::transaction::HummockVersionTransaction;
use crate::hummock::manager::versioning::Versioning;
use crate::hummock::manager::{HummockManager, commit_multi_var};
use crate::hummock::metrics_utils::remove_compaction_group_metrics;
use crate::hummock::model::CompactionGroup;
use crate::hummock::sequence::next_compaction_group_id;
use crate::manager::MetaSrvEnv;
use crate::model::{
    BTreeMapTransaction, BTreeMapTransactionInner, DerefMutForward, MetadataModelError,
};

type CompactionGroupTransaction<'a> = BTreeMapTransaction<'a, CompactionGroupId, CompactionGroup>;

impl CompactionGroupManager {
    pub(crate) async fn new(env: &MetaSrvEnv) -> Result<CompactionGroupManager> {
        let default_config = match env.opts.compaction_config.as_ref() {
            None => CompactionConfigBuilder::new().build(),
            Some(opt) => CompactionConfigBuilder::with_opt(opt).build(),
        };
        Self::new_with_config(env, default_config).await
    }

    pub(crate) async fn new_with_config(
        env: &MetaSrvEnv,
        default_config: CompactionConfig,
    ) -> Result<CompactionGroupManager> {
        let mut compaction_group_manager = CompactionGroupManager {
            compaction_groups: BTreeMap::new(),
            default_config: Arc::new(default_config),
            write_limit: Default::default(),
        };

        let loaded_compaction_groups: BTreeMap<CompactionGroupId, CompactionGroup> =
            compaction_config::Entity::find()
                .all(&env.meta_store_ref().conn)
                .await
                .map_err(MetadataModelError::from)?
                .into_iter()
                .map(|m| (m.compaction_group_id as CompactionGroupId, m.into()))
                .collect();

        compaction_group_manager.init(loaded_compaction_groups);
        Ok(compaction_group_manager)
    }

    fn init(&mut self, loaded_compaction_groups: BTreeMap<CompactionGroupId, CompactionGroup>) {
        if !loaded_compaction_groups.is_empty() {
            self.compaction_groups = loaded_compaction_groups;
        }
    }
}

impl HummockManager {
    /// Should not be called inside [`HummockManager`], because it requests locks internally.
    /// The implementation acquires `versioning` lock.
    pub async fn compaction_group_ids(&self) -> Vec<CompactionGroupId> {
        get_compaction_group_ids(&self.versioning.read().await.current_version).collect_vec()
    }

    /// The implementation acquires `compaction_group_manager` lock.
    pub async fn get_compaction_group_map(&self) -> BTreeMap<CompactionGroupId, CompactionGroup> {
        self.compaction_group_manager
            .read()
            .await
            .compaction_groups
            .clone()
    }

    #[cfg(test)]
    /// Registers `table_fragments` to compaction groups.
    pub async fn register_table_fragments(
        &self,
        mv_table: Option<TableId>,
        mut internal_tables: Vec<TableId>,
    ) -> Result<()> {
        let mut pairs = vec![];
        if let Some(mv_table) = mv_table {
            if internal_tables.extract_if(.., |t| *t == mv_table).count() > 0 {
                tracing::warn!("`mv_table` {} found in `internal_tables`", mv_table);
            }
            // materialized_view
            pairs.push((mv_table, StaticCompactionGroupId::MaterializedView));
        }
        // internal states
        for table_id in internal_tables {
            pairs.push((table_id, StaticCompactionGroupId::StateDefault));
        }
        self.register_table_ids_for_test(&pairs).await?;
        Ok(())
    }

    #[cfg(test)]
    /// Unregisters `table_fragments` from compaction groups
    pub async fn unregister_table_fragments_vec(
        &self,
        table_fragments: &[crate::model::StreamJobFragments],
    ) {
        self.unregister_table_ids(table_fragments.iter().flat_map(|t| t.all_table_ids()))
            .await
            .unwrap();
    }

    /// Unregisters stale members and groups
    /// The caller should ensure `table_fragments_list` remain unchanged during `purge`.
    /// Currently `purge` is only called during meta service start ups.
    pub async fn purge(&self, valid_ids: &HashSet<TableId>) -> Result<()> {
        let to_unregister = self
            .versioning
            .read()
            .await
            .current_version
            .state_table_info
            .info()
            .keys()
            .cloned()
            .filter(|table_id| !valid_ids.contains(table_id))
            .collect_vec();

        // As we have released versioning lock, the version that `to_unregister` is calculated from
        // may not be the same as the one used in unregister_table_ids. It is OK.
        self.unregister_table_ids(to_unregister).await
    }

    /// The implementation acquires `versioning` lock.
    ///
    /// The method name is temporarily added with a `_for_test` prefix to mark
    /// that it's currently only used in test.
    pub async fn register_table_ids_for_test(
        &self,
        pairs: &[(impl Into<TableId> + Copy, CompactionGroupId)],
    ) -> Result<()> {
        if pairs.is_empty() {
            return Ok(());
        }
        let mut versioning_guard = self.versioning.write().await;
        let versioning = versioning_guard.deref_mut();
        let mut compaction_group_manager = self.compaction_group_manager.write().await;
        let current_version = &versioning.current_version;
        let default_config = compaction_group_manager.default_compaction_config();
        let mut compaction_groups_txn = compaction_group_manager.start_compaction_groups_txn();

        for (table_id, _) in pairs {
            let table_id = (*table_id).into();
            if let Some(info) = current_version.state_table_info.info().get(&table_id) {
                return Err(Error::CompactionGroup(format!(
                    "table {} already {:?}",
                    table_id, info
                )));
            }
        }
        // All NewCompactionGroup pairs are mapped to one new compaction group.
        let new_compaction_group_id: OnceCell<CompactionGroupId> = OnceCell::new();
        let mut version = HummockVersionTransaction::new(
            &mut versioning.current_version,
            &mut versioning.hummock_version_deltas,
            &mut versioning.table_change_log,
            self.env.notification_manager(),
            None,
            &self.metrics,
            &self.env.opts,
        );
        let mut new_version_delta = version.new_delta();

        let committed_epoch = new_version_delta
            .latest_version()
            .state_table_info
            .info()
            .values()
            .map(|info| info.committed_epoch)
            .max()
            .unwrap_or(INVALID_EPOCH);

        for (table_id, raw_group_id) in pairs {
            let table_id = (*table_id).into();
            let mut group_id = *raw_group_id;
            if group_id == StaticCompactionGroupId::NewCompactionGroup {
                let mut is_group_init = false;
                group_id = *new_compaction_group_id
                    .get_or_try_init(|| async {
                        next_compaction_group_id(&self.env).await.inspect(|_| {
                            is_group_init = true;
                        })
                    })
                    .await?;
                if is_group_init {
                    let group_deltas = &mut new_version_delta
                        .group_deltas
                        .entry(group_id)
                        .or_default()
                        .group_deltas;

                    let config =
                        match compaction_groups_txn.try_get_compaction_group_config(group_id) {
                            Some(config) => config.compaction_config.as_ref().clone(),
                            None => {
                                compaction_groups_txn
                                    .create_compaction_groups(group_id, default_config.clone());
                                default_config.as_ref().clone()
                            }
                        };

                    let group_delta = GroupDelta::GroupConstruct(Box::new(PbGroupConstruct {
                        group_config: Some(config),
                        group_id,
                        ..Default::default()
                    }));

                    group_deltas.push(group_delta);
                }
            }
            assert!(
                new_version_delta
                    .state_table_info_delta
                    .insert(
                        table_id,
                        PbStateTableInfoDelta {
                            committed_epoch,
                            compaction_group_id: *raw_group_id,
                        }
                    )
                    .is_none()
            );
        }
        new_version_delta.pre_apply();
        commit_multi_var!(self.meta_store_ref(), version, compaction_groups_txn)?;

        Ok(())
    }

    pub async fn unregister_table_ids(
        &self,
        table_ids: impl IntoIterator<Item = TableId>,
    ) -> Result<()> {
        let table_ids = table_ids.into_iter().collect_vec();
        if table_ids.is_empty() {
            return Ok(());
        }

        {
            // Remove table write throughput statistics
            // The Caller acquires `Send`, so we should safely use `write` lock before the await point.
            // The table write throughput statistic accepts data inconsistencies (unregister table ids fail), so we can clean it up in advance.
            let mut table_write_throughput_statistic_manager =
                self.table_write_throughput_statistic_manager.write();
            for &table_id in table_ids.iter().unique() {
                table_write_throughput_statistic_manager.remove_table(table_id);
            }
        }

        let mut versioning_guard = self.versioning.write().await;
        let versioning = versioning_guard.deref_mut();
        let mut version = HummockVersionTransaction::new(
            &mut versioning.current_version,
            &mut versioning.hummock_version_deltas,
            &mut versioning.table_change_log,
            self.env.notification_manager(),
            None,
            &self.metrics,
            &self.env.opts,
        );
        let mut new_version_delta = version.new_delta();
        struct UnregisterGroupChange {
            remaining_member_count: usize,
            removed_table_ids: HashSet<TableId>,
        }
        let mut group_changes: HashMap<CompactionGroupId, UnregisterGroupChange> = HashMap::new();
        // Remove member tables
        for table_id in table_ids.into_iter().unique() {
            let version = new_version_delta.latest_version();
            let Some(info) = version.state_table_info.info().get(&table_id) else {
                continue;
            };
            let compaction_group_id = info.compaction_group_id;

            let group_change =
                group_changes
                    .entry(compaction_group_id)
                    .or_insert_with(|| UnregisterGroupChange {
                        remaining_member_count: version
                            .state_table_info
                            .compaction_group_member_tables()
                            .get(&compaction_group_id)
                            .expect("should exist")
                            .len(),
                        removed_table_ids: HashSet::new(),
                    });
            group_change.remaining_member_count = group_change
                .remaining_member_count
                .checked_sub(1)
                .expect("member table count should be positive");
            assert!(group_change.removed_table_ids.insert(table_id));
            new_version_delta.removed_table_ids.insert(table_id);
        }

        for (group_id, change) in group_changes {
            if change.remaining_member_count == 0 && group_id > StaticCompactionGroupId::End {
                let max_level = new_version_delta
                    .latest_version()
                    .get_compaction_group_levels(group_id)
                    .levels
                    .len();
                new_version_delta
                    .group_deltas
                    .entry(group_id)
                    .or_default()
                    .group_deltas
                    .push(GroupDelta::GroupDestroy(PbGroupDestroy {}));
                remove_compaction_group_metrics(&self.metrics, group_id, max_level);
                // clean up compaction schedule state for the removed group
                self.compaction_state.remove_compaction_group(group_id);
            } else {
                new_version_delta
                    .group_deltas
                    .entry(group_id)
                    .or_default()
                    .group_deltas
                    .push(GroupDelta::PruneTableIdsFromSsts(change.removed_table_ids));
            }
        }

        new_version_delta.pre_apply();

        // Purge may cause write to meta store. If it hurts performance while holding versioning
        // lock, consider to make it in batch.
        let mut compaction_group_manager = self.compaction_group_manager.write().await;
        let mut compaction_groups_txn = compaction_group_manager.start_compaction_groups_txn();

        compaction_groups_txn.purge(HashSet::from_iter(get_compaction_group_ids(
            version.latest_version(),
        )));
        commit_multi_var!(self.meta_store_ref(), version, compaction_groups_txn)?;

        // No need to handle DeltaType::GroupDestroy during time travel.
        Ok(())
    }

    pub async fn update_compaction_config(
        &self,
        compaction_group_ids: &[CompactionGroupId],
        config_to_update: &[MutableConfig],
    ) -> Result<()> {
        {
            // Avoid lock conflicts with `try_update_write_limits``
            let mut compaction_group_manager = self.compaction_group_manager.write().await;
            let mut compaction_groups_txn = compaction_group_manager.start_compaction_groups_txn();
            compaction_groups_txn
                .update_compaction_config(compaction_group_ids, config_to_update)?;
            commit_multi_var!(self.meta_store_ref(), compaction_groups_txn)?;
        }

        if config_to_update
            .iter()
            .any(|c| matches!(c, MutableConfig::Level0StopWriteThresholdSubLevelNumber(_)))
        {
            // Update write limits with lock
            self.try_update_write_limits(compaction_group_ids).await;
        }

        Ok(())
    }

    /// Gets complete compaction group info.
    /// It is the aggregate of `HummockVersion` and `CompactionGroupConfig`
    pub async fn list_compaction_group(&self) -> Vec<CompactionGroupInfo> {
        let mut versioning_guard = self.versioning.write().await;
        let versioning = versioning_guard.deref_mut();
        let current_version = &versioning.current_version;
        let mut results = vec![];
        let compaction_group_manager = self.compaction_group_manager.read().await;

        for levels in current_version.levels.values() {
            let compaction_config = compaction_group_manager
                .try_get_compaction_group_config(levels.group_id)
                .unwrap()
                .compaction_config
                .as_ref()
                .clone();
            let group = CompactionGroupInfo {
                id: levels.group_id,
                parent_id: levels.parent_group_id,
                member_table_ids: current_version
                    .state_table_info
                    .compaction_group_member_table_ids(levels.group_id)
                    .iter()
                    .copied()
                    .collect_vec(),
                compaction_config: Some(compaction_config),
            };
            results.push(group);
        }
        results
    }

    pub async fn calculate_compaction_group_statistic(&self) -> Vec<CompactionGroupStatistic> {
        let mut infos = vec![];
        {
            let versioning_guard = self.versioning.read().await;
            let manager = self.compaction_group_manager.read().await;
            let version = &versioning_guard.current_version;
            for group_id in version.levels.keys() {
                let mut group_info = CompactionGroupStatistic {
                    group_id: *group_id,
                    ..Default::default()
                };
                for table_id in version
                    .state_table_info
                    .compaction_group_member_table_ids(*group_id)
                {
                    let stats_size = versioning_guard
                        .version_stats
                        .table_stats
                        .get(table_id)
                        .map(|stats| stats.total_key_size + stats.total_value_size)
                        .unwrap_or(0);
                    let table_size = std::cmp::max(stats_size, 0) as u64;
                    group_info.group_size += table_size;
                    group_info.table_statistic.insert(*table_id, table_size);
                    group_info.compaction_group_config =
                        manager.try_get_compaction_group_config(*group_id).unwrap();
                }
                infos.push(group_info);
            }
        };
        infos
    }

    pub(crate) async fn initial_compaction_group_config_after_load(
        &self,
        versioning_guard: &Versioning,
        compaction_group_manager: &mut CompactionGroupManager,
    ) -> Result<()> {
        // 1. Due to version compatibility, we fix some of the configuration of older versions after hummock starts.
        let current_version = &versioning_guard.current_version;
        let all_group_ids = get_compaction_group_ids(current_version).collect_vec();
        let default_config = compaction_group_manager.default_compaction_config();
        let mut compaction_groups_txn = compaction_group_manager.start_compaction_groups_txn();
        compaction_groups_txn.try_create_compaction_groups(&all_group_ids, default_config);
        commit_multi_var!(self.meta_store_ref(), compaction_groups_txn)?;

        Ok(())
    }
}

/// We muse ensure there is an entry exists in [`CompactionGroupManager`] for any
/// compaction group found in current hummock version. That's done by invoking
/// `get_or_insert_compaction_group_config` or `get_or_insert_compaction_group_configs` before
/// adding any group in current hummock version:
/// 1. initialize default static compaction group.
/// 2. register new table to new compaction group.
/// 3. move existent table to new compaction group.
pub(crate) struct CompactionGroupManager {
    compaction_groups: BTreeMap<CompactionGroupId, CompactionGroup>,
    default_config: Arc<CompactionConfig>,
    /// Tables that write limit is trigger for.
    pub write_limit: HashMap<CompactionGroupId, WriteLimit>,
}

impl CompactionGroupManager {
    /// Starts a transaction to update compaction group configs.
    pub fn start_compaction_groups_txn(&mut self) -> CompactionGroupTransaction<'_> {
        CompactionGroupTransaction::new(&mut self.compaction_groups)
    }

    #[expect(clippy::type_complexity)]
    pub fn start_owned_compaction_groups_txn<P: DerefMut<Target = Self>>(
        inner: P,
    ) -> BTreeMapTransactionInner<
        CompactionGroupId,
        CompactionGroup,
        DerefMutForward<
            Self,
            BTreeMap<CompactionGroupId, CompactionGroup>,
            P,
            impl Fn(&Self) -> &BTreeMap<CompactionGroupId, CompactionGroup>,
            impl Fn(&mut Self) -> &mut BTreeMap<CompactionGroupId, CompactionGroup>,
        >,
    > {
        BTreeMapTransactionInner::new(DerefMutForward::new(
            inner,
            |mgr| &mgr.compaction_groups,
            |mgr| &mut mgr.compaction_groups,
        ))
    }

    /// Tries to get compaction group config for `compaction_group_id`.
    pub(crate) fn try_get_compaction_group_config(
        &self,
        compaction_group_id: impl Into<CompactionGroupId>,
    ) -> Option<CompactionGroup> {
        self.compaction_groups
            .get(&compaction_group_id.into())
            .cloned()
    }

    /// Tries to get compaction group config for `compaction_group_id`.
    pub(crate) fn default_compaction_config(&self) -> Arc<CompactionConfig> {
        self.default_config.clone()
    }
}

fn update_compaction_config(target: &mut CompactionConfig, items: &[MutableConfig]) -> Result<()> {
    for item in items {
        match item {
            MutableConfig::MaxBytesForLevelBase(c) => {
                target.max_bytes_for_level_base = *c;
            }
            MutableConfig::MaxBytesForLevelMultiplier(c) => {
                target.max_bytes_for_level_multiplier = *c;
            }
            MutableConfig::MaxCompactionBytes(c) => {
                target.max_compaction_bytes = *c;
            }
            MutableConfig::SubLevelMaxCompactionBytes(c) => {
                target.sub_level_max_compaction_bytes = *c;
            }
            MutableConfig::Level0TierCompactFileNumber(c) => {
                target.level0_tier_compact_file_number = *c;
            }
            MutableConfig::TargetFileSizeBase(c) => {
                target.target_file_size_base = *c;
            }
            MutableConfig::CompactionFilterMask(c) => {
                target.compaction_filter_mask = *c;
            }
            MutableConfig::MaxSubCompaction(c) => {
                target.max_sub_compaction = *c;
            }
            MutableConfig::Level0StopWriteThresholdSubLevelNumber(c) => {
                target.level0_stop_write_threshold_sub_level_number = *c;
            }
            MutableConfig::Level0SubLevelCompactLevelCount(c) => {
                target.level0_sub_level_compact_level_count = *c;
            }
            MutableConfig::Level0OverlappingSubLevelCompactLevelCount(c) => {
                target.level0_overlapping_sub_level_compact_level_count = *c;
            }
            MutableConfig::MaxSpaceReclaimBytes(c) => {
                target.max_space_reclaim_bytes = *c;
            }
            MutableConfig::Level0MaxCompactFileNumber(c) => {
                target.level0_max_compact_file_number = *c;
            }
            MutableConfig::EnableEmergencyPicker(c) => {
                target.enable_emergency_picker = *c;
            }
            MutableConfig::TombstoneReclaimRatio(c) => {
                target.tombstone_reclaim_ratio = *c;
            }
            MutableConfig::CompressionAlgorithm(c) => {
                let level = c.get_level();
                let max_level = try_u32_max_level(target.max_level)?;
                if level > max_level {
                    return Err(Error::CompactionGroup(format!(
                        "invalid compression_algorithm level {}, max_level is {}",
                        level, target.max_level
                    )));
                }

                let Some(algorithm) = target.compression_algorithm.get_mut(level as usize) else {
                    return Err(Error::CompactionGroup(format!(
                        "invalid compression_algorithm level {}, compression_algorithm len is {}",
                        level,
                        target.compression_algorithm.len()
                    )));
                };
                algorithm.clone_from(&c.compression_algorithm);
            }
            MutableConfig::ResetCompressionAlgorithm(reset) => {
                if *reset {
                    target.compression_algorithm =
                        default_compaction_config::compression_algorithm_vec(try_u32_max_level(
                            target.max_level,
                        )?);
                }
            }
            MutableConfig::MaxL0CompactLevelCount(c) => {
                target.max_l0_compact_level_count = Some(*c);
            }
            MutableConfig::SstAllowedTrivialMoveMinSize(c) => {
                target.sst_allowed_trivial_move_min_size = Some(*c);
            }
            MutableConfig::SplitWeightByVnode(c) => {
                target.split_weight_by_vnode = *c;
            }
            MutableConfig::DisableAutoGroupScheduling(c) => {
                target.disable_auto_group_scheduling = Some(*c);
            }
            MutableConfig::MaxOverlappingLevelSize(c) => {
                target.max_overlapping_level_size = Some(*c);
            }
            MutableConfig::SstAllowedTrivialMoveMaxCount(c) => {
                target.sst_allowed_trivial_move_max_count = Some(*c);
            }
            MutableConfig::EmergencyLevel0SstFileCount(c) => {
                target.emergency_level0_sst_file_count = Some(*c);
            }
            MutableConfig::EmergencyLevel0SubLevelPartition(c) => {
                target.emergency_level0_sub_level_partition = Some(*c);
            }
            MutableConfig::Level0StopWriteThresholdMaxSstCount(c) => {
                target.level0_stop_write_threshold_max_sst_count = Some(*c);
            }
            MutableConfig::Level0StopWriteThresholdMaxSize(c) => {
                target.level0_stop_write_threshold_max_size = Some(*c);
            }
            MutableConfig::EnableOptimizeL0IntervalSelection(c) => {
                target.enable_optimize_l0_interval_selection = Some(*c);
            }
            #[expect(deprecated)]
            MutableConfig::VnodeAlignedLevelSizeThreshold(_) => {
                // Deprecated. Keep accepting the field for old clients but do not apply it.
            }
            MutableConfig::MaxKvCountForXor16(c) => {
                target.max_kv_count_for_xor16 = optional_non_sentinel_u64_config(*c);
            }
            MutableConfig::MaxVnodeKeyRangeBytes(c) => {
                target.max_vnode_key_range_bytes = optional_positive_u64_config(*c);
            }
            MutableConfig::SstableFilterKind(c) => {
                if target.sstable_filter_kind.is_empty() {
                    target.sstable_filter_kind =
                        vec!["xor16".to_owned(); target.max_level as usize + 1];
                }
                let idx = c.get_level() as usize;
                let level_entry = target.sstable_filter_kind.get_mut(idx).ok_or_else(|| {
                    Error::CompactionGroup(format!(
                        "sstable_filter_kind level {} is out of range",
                        idx
                    ))
                })?;
                level_entry.clone_from(&c.filter_kind);
            }
            MutableConfig::SstableFilterLayout(c) => {
                if target.sstable_filter_layout.is_empty() {
                    target.sstable_filter_layout =
                        vec!["auto".to_owned(); target.max_level as usize + 1];
                }
                let idx = c.get_level() as usize;
                let level_entry = target.sstable_filter_layout.get_mut(idx).ok_or_else(|| {
                    Error::CompactionGroup(format!(
                        "sstable_filter_layout level {} is out of range",
                        idx
                    ))
                })?;
                level_entry.clone_from(&c.layout);
            }
        }
    }
    Ok(())
}

fn optional_u64_config(value: u64) -> Option<u64> {
    (value != u64::MIN && value != u64::MAX).then_some(value)
}

fn optional_non_sentinel_u64_config(value: u64) -> Option<u64> {
    (value != u64::MAX).then_some(value)
}

fn optional_positive_u64_config(value: u64) -> Option<u64> {
    optional_u64_config(value).filter(|value| *value > 0)
}

fn try_u32_max_level(max_level: u64) -> Result<u32> {
    u32::try_from(max_level).map_err(|_| {
        Error::CompactionGroup(format!(
            "invalid max_level {}, expect <= {}",
            max_level,
            u32::MAX
        ))
    })
}

impl CompactionGroupTransaction<'_> {
    /// Inserts compaction group configs if they do not exist.
    pub fn try_create_compaction_groups(
        &mut self,
        compaction_group_ids: &[CompactionGroupId],
        config: Arc<CompactionConfig>,
    ) -> bool {
        let mut trivial = true;
        for id in compaction_group_ids {
            if self.contains_key(id) {
                continue;
            }
            let new_entry = CompactionGroup::new(*id, config.as_ref().clone());
            self.insert(*id, new_entry);

            trivial = false;
        }

        !trivial
    }

    pub fn create_compaction_groups(
        &mut self,
        compaction_group_id: CompactionGroupId,
        config: Arc<CompactionConfig>,
    ) {
        self.try_create_compaction_groups(&[compaction_group_id], config);
    }

    /// Tries to get compaction group config for `compaction_group_id`.
    pub(crate) fn try_get_compaction_group_config(
        &self,
        compaction_group_id: CompactionGroupId,
    ) -> Option<&CompactionGroup> {
        self.get(&compaction_group_id)
    }

    /// Removes stale group configs.
    pub fn purge(&mut self, existing_groups: HashSet<CompactionGroupId>) {
        let stale_group = self
            .tree_ref()
            .keys()
            .cloned()
            .filter(|k| !existing_groups.contains(k))
            .collect_vec();
        if stale_group.is_empty() {
            return;
        }
        for group in stale_group {
            self.remove(group);
        }
    }

    pub(crate) fn update_compaction_config(
        &mut self,
        compaction_group_ids: &[CompactionGroupId],
        config_to_update: &[MutableConfig],
    ) -> Result<HashMap<CompactionGroupId, CompactionGroup>> {
        let mut results = HashMap::default();
        for compaction_group_id in compaction_group_ids.iter().unique() {
            let group = self.get(compaction_group_id).ok_or_else(|| {
                Error::CompactionGroup(format!("invalid group {}", *compaction_group_id))
            })?;
            let mut config = group.compaction_config.as_ref().clone();
            update_compaction_config(&mut config, config_to_update)?;
            if let Err(reason) = validate_compaction_config(&config) {
                return Err(Error::CompactionGroup(reason));
            }
            let mut new_group = group.clone();
            new_group.compaction_config = Arc::new(config);
            self.insert(*compaction_group_id, new_group.clone());
            results.insert(new_group.group_id(), new_group);
        }

        Ok(results)
    }
}

#[cfg(test)]
mod tests {
    use std::collections::{BTreeMap, HashSet};
    use std::sync::Arc;

    use itertools::Itertools;
    use risingwave_common::id::JobId;
    use risingwave_hummock_sdk::CompactionGroupId;
    use risingwave_pb::hummock::rise_ctl_update_compaction_config_request::mutable_config::MutableConfig;
    use risingwave_pb::hummock::rise_ctl_update_compaction_config_request::{
        CompressionAlgorithm, SstableFilterKind, SstableFilterLayout,
    };

    use crate::controller::SqlMetaStore;
    use crate::hummock::commit_multi_var;
    use crate::hummock::compaction::compaction_config::CompactionConfigBuilder;
    use crate::hummock::error::Result;
    use crate::hummock::manager::compaction_group_manager::CompactionGroupManager;
    use crate::hummock::test_utils::setup_compute_env;
    use crate::model::{Fragment, StreamJobFragments};

    #[test]
    fn test_update_compaction_config_filter_kind_layout_backward_compat() {
        let mut config = CompactionConfigBuilder::new().build();
        config.sstable_filter_kind.clear();
        config.sstable_filter_layout.clear();

        super::update_compaction_config(
            &mut config,
            &[MutableConfig::SstableFilterKind(SstableFilterKind {
                level: 0,
                filter_kind: "xor8".to_owned(),
            })],
        )
        .unwrap();
        assert_eq!(
            config.sstable_filter_kind.len(),
            config.max_level as usize + 1
        );
        assert_eq!(config.sstable_filter_kind[0], "xor8");

        super::update_compaction_config(
            &mut config,
            &[MutableConfig::SstableFilterLayout(SstableFilterLayout {
                level: 1,
                layout: "plain".to_owned(),
            })],
        )
        .unwrap();
        assert_eq!(
            config.sstable_filter_layout.len(),
            config.max_level as usize + 1
        );
        assert_eq!(config.sstable_filter_layout[1], "plain");
    }

    #[test]
    fn test_update_compaction_config_rejects_out_of_range_level() {
        let mut config = CompactionConfigBuilder::new().build();
        let oob = config.max_level as u32 + 1;

        assert!(
            super::update_compaction_config(
                &mut config,
                &[MutableConfig::SstableFilterKind(SstableFilterKind {
                    level: oob,
                    filter_kind: "xor8".to_owned(),
                })],
            )
            .is_err()
        );

        assert!(
            super::update_compaction_config(
                &mut config,
                &[MutableConfig::SstableFilterLayout(SstableFilterLayout {
                    level: oob,
                    layout: "plain".to_owned(),
                })],
            )
            .is_err()
        );

        assert!(
            super::update_compaction_config(
                &mut config,
                &[MutableConfig::CompressionAlgorithm(CompressionAlgorithm {
                    level: oob,
                    compression_algorithm: "Zstd".to_owned(),
                })],
            )
            .is_err()
        );
    }

    #[test]
    fn test_reset_compression_algorithm_false_is_noop() {
        let mut config = CompactionConfigBuilder::new().build();
        config.compression_algorithm[3] = "Zstd".to_owned();

        super::update_compaction_config(
            &mut config,
            &[MutableConfig::ResetCompressionAlgorithm(false)],
        )
        .unwrap();

        assert_eq!(config.compression_algorithm[3], "Zstd");
    }

    #[tokio::test]
    async fn test_inner() {
        let (env, ..) = setup_compute_env(8080).await;
        let mut inner = CompactionGroupManager::new(&env).await.unwrap();
        assert_eq!(inner.compaction_groups.len(), 2);

        async fn update_compaction_config(
            meta: &SqlMetaStore,
            inner: &mut CompactionGroupManager,
            cg_ids: &[impl Into<CompactionGroupId> + Copy],
            config_to_update: &[MutableConfig],
        ) -> Result<()> {
            let cg_ids = cg_ids.iter().copied().map_into().collect_vec();
            let mut compaction_groups_txn = inner.start_compaction_groups_txn();
            compaction_groups_txn.update_compaction_config(&cg_ids, config_to_update)?;
            commit_multi_var!(meta, compaction_groups_txn)
        }

        async fn insert_compaction_group_configs(
            meta: &SqlMetaStore,
            inner: &mut CompactionGroupManager,
            cg_ids: &[u64],
        ) {
            let default_config = inner.default_compaction_config();
            let mut compaction_groups_txn = inner.start_compaction_groups_txn();
            if compaction_groups_txn.try_create_compaction_groups(
                &cg_ids.iter().copied().map_into().collect_vec(),
                default_config,
            ) {
                commit_multi_var!(meta, compaction_groups_txn).unwrap();
            }
        }

        async fn insert_compaction_group_config_with_max_level(
            meta: &SqlMetaStore,
            inner: &mut CompactionGroupManager,
            cg_id: u64,
            max_level: u64,
        ) {
            let mut config = inner.default_compaction_config().as_ref().clone();
            config.max_level = max_level;
            config.compression_algorithm =
                super::default_compaction_config::compression_algorithm_vec(
                    super::try_u32_max_level(max_level).expect("max_level should fit u32 in test"),
                );
            let mut compaction_groups_txn = inner.start_compaction_groups_txn();
            compaction_groups_txn.create_compaction_groups(cg_id.into(), Arc::new(config));
            commit_multi_var!(meta, compaction_groups_txn).unwrap();
        }

        update_compaction_config(env.meta_store_ref(), &mut inner, &[100, 200], &[])
            .await
            .unwrap_err();
        insert_compaction_group_configs(env.meta_store_ref(), &mut inner, &[100, 200]).await;
        assert_eq!(inner.compaction_groups.len(), 4);
        let mut inner = CompactionGroupManager::new(&env).await.unwrap();
        assert_eq!(inner.compaction_groups.len(), 4);

        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100, 200],
            &[MutableConfig::MaxSubCompaction(123)],
        )
        .await
        .unwrap();
        assert_eq!(inner.compaction_groups.len(), 4);
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_sub_compaction,
            123
        );
        assert_eq!(
            inner
                .try_get_compaction_group_config(200)
                .unwrap()
                .compaction_config
                .max_sub_compaction,
            123
        );

        insert_compaction_group_config_with_max_level(env.meta_store_ref(), &mut inner, 300, 4)
            .await;
        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[300],
            &[MutableConfig::ResetCompressionAlgorithm(true)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(300)
                .unwrap()
                .compaction_config
                .compression_algorithm,
            super::default_compaction_config::compression_algorithm_vec(4)
        );
        let err = update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[300],
            &[MutableConfig::CompressionAlgorithm(CompressionAlgorithm {
                level: 6,
                compression_algorithm: "Zstd".to_owned(),
            })],
        )
        .await
        .unwrap_err();
        assert!(
            err.to_string()
                .contains("invalid compression_algorithm level 6")
        );

        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100],
            &[MutableConfig::MaxKvCountForXor16(0)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_kv_count_for_xor16,
            Some(0)
        );
        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100],
            &[MutableConfig::MaxKvCountForXor16(1024)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_kv_count_for_xor16,
            Some(1024)
        );
        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100],
            &[MutableConfig::MaxKvCountForXor16(u64::MAX)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_kv_count_for_xor16,
            None
        );

        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100],
            &[MutableConfig::MaxVnodeKeyRangeBytes(0)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_vnode_key_range_bytes,
            None
        );
        update_compaction_config(
            env.meta_store_ref(),
            &mut inner,
            &[100],
            &[MutableConfig::MaxVnodeKeyRangeBytes(1024)],
        )
        .await
        .unwrap();
        assert_eq!(
            inner
                .try_get_compaction_group_config(100)
                .unwrap()
                .compaction_config
                .max_vnode_key_range_bytes,
            Some(1024)
        );
    }

    #[tokio::test]
    async fn test_manager() {
        let (_, compaction_group_manager, ..) = setup_compute_env(8080).await;
        let table_fragment_1 = StreamJobFragments::for_test(
            JobId::new(10),
            BTreeMap::from([(
                1.into(),
                Fragment {
                    fragment_id: 1.into(),
                    state_table_ids: vec![10.into(), 11.into(), 12.into(), 13.into()],
                    ..Default::default()
                },
            )]),
        );
        let table_fragment_2 = StreamJobFragments::for_test(
            JobId::new(20),
            BTreeMap::from([(
                2.into(),
                Fragment {
                    fragment_id: 2.into(),
                    state_table_ids: vec![20.into(), 21.into(), 22.into(), 23.into()],
                    ..Default::default()
                },
            )]),
        );

        // Test register_table_fragments
        let registered_number = || async {
            compaction_group_manager
                .list_compaction_group()
                .await
                .iter()
                .map(|cg| cg.member_table_ids.len())
                .sum::<usize>()
        };
        let group_number =
            || async { compaction_group_manager.list_compaction_group().await.len() };
        assert_eq!(registered_number().await, 0);

        compaction_group_manager
            .register_table_fragments(
                Some(table_fragment_1.stream_job_id().as_mv_table_id()),
                table_fragment_1
                    .internal_table_ids()
                    .into_iter()
                    .map_into()
                    .collect(),
            )
            .await
            .unwrap();
        assert_eq!(registered_number().await, 4);
        compaction_group_manager
            .register_table_fragments(
                Some(table_fragment_2.stream_job_id().as_mv_table_id()),
                table_fragment_2
                    .internal_table_ids()
                    .into_iter()
                    .map_into()
                    .collect(),
            )
            .await
            .unwrap();
        assert_eq!(registered_number().await, 8);

        // Test unregister_table_fragments
        compaction_group_manager
            .unregister_table_fragments_vec(std::slice::from_ref(&table_fragment_1))
            .await;
        assert_eq!(registered_number().await, 4);

        // Test purge_stale_members: table fragments
        compaction_group_manager
            .purge(&table_fragment_2.all_table_ids().collect())
            .await
            .unwrap();
        assert_eq!(registered_number().await, 4);
        compaction_group_manager
            .purge(&HashSet::new())
            .await
            .unwrap();
        assert_eq!(registered_number().await, 0);

        assert_eq!(group_number().await, 2);

        compaction_group_manager
            .register_table_fragments(
                Some(table_fragment_1.stream_job_id().as_mv_table_id()),
                table_fragment_1
                    .internal_table_ids()
                    .into_iter()
                    .map_into()
                    .collect(),
            )
            .await
            .unwrap();
        assert_eq!(registered_number().await, 4);
        assert_eq!(group_number().await, 2);

        compaction_group_manager
            .unregister_table_fragments_vec(&[table_fragment_1])
            .await;
        assert_eq!(registered_number().await, 0);
        assert_eq!(group_number().await, 2);
    }
}