mito2/worker/

handle_compaction.rs

// Copyright 2023 Greptime Team
//
// 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 api::v1::region::compact_request;
use common_telemetry::{debug, error, info};
use store_api::logstore::LogStore;
use store_api::region_request::RegionCompactRequest;
use store_api::storage::RegionId;

use crate::compaction::{CompactionPickFinished, CompactionTransition};
use crate::config::IndexBuildMode;
use crate::error::{RegionNotFoundSnafu, StaleCompactionExecutionSnafu};
use crate::metrics::COMPACTION_REQUEST_COUNT;
use crate::region::MitoRegionRef;
use crate::request::{
    BuildIndexRequest, CompactionFailed, CompactionFinished, CompactionUnitNotification, OnFailure,
    OptionOutputTx,
};
use crate::sst::index::IndexBuildType;
use crate::worker::RegionWorkerLoop;

fn made_progress(files_to_add: usize, files_to_remove: usize) -> bool {
    files_to_add > 0 || files_to_remove > files_to_add
}

impl<S> RegionWorkerLoop<S> {
    /// Drives local unit publication and applies successful edits before acknowledging completion.
    pub(crate) async fn handle_compaction_unit(
        &mut self,
        region_id: RegionId,
        notify: CompactionUnitNotification,
    ) where
        S: LogStore,
    {
        match notify {
            CompactionUnitNotification::Finished {
                plan_id,
                result,
                applied,
            } => {
                let Some(region) = self.regions.get_region(region_id) else {
                    return;
                };
                match result {
                    Ok(edit) => {
                        if !self.compaction_scheduler.can_apply_unit(region_id, plan_id) {
                            let _ = applied.send(());
                            return;
                        }
                        let progress =
                            made_progress(edit.files_to_add.len(), edit.files_to_remove.len());
                        region.version_control.apply_edit(
                            Some(edit.clone()),
                            &[],
                            region.file_purger.clone(),
                        );
                        self.listener.on_compaction_unit_applied(region_id, plan_id);
                        self.compaction_scheduler.on_unit_finished(
                            region_id,
                            plan_id,
                            Ok(progress),
                        );
                        if self.config.index.build_mode == IndexBuildMode::Async
                            && !edit.files_to_add.is_empty()
                        {
                            self.handle_rebuild_index(
                                BuildIndexRequest {
                                    region_id,
                                    build_type: IndexBuildType::Compact,
                                    file_metas: edit.files_to_add,
                                    result_senders: Vec::new(),
                                },
                                OptionOutputTx::new(None),
                            )
                            .await;
                        }
                    }
                    Err(err) => {
                        self.compaction_scheduler
                            .on_unit_finished(region_id, plan_id, Err(err));
                    }
                }
                let _ = applied.send(());
            }
            CompactionUnitNotification::Released { plan_id } => {
                let Some(region) = self.regions.get_region(region_id) else {
                    return;
                };
                let transition = self
                    .compaction_scheduler
                    .on_unit_released(
                        region_id,
                        plan_id,
                        &region.manifest_ctx,
                        self.schema_metadata_manager.clone(),
                    )
                    .await;
                self.listener.on_compaction_result_notified(region_id).await;
                match transition {
                    CompactionTransition::AutomaticFollowupScheduled => {
                        region.update_schedule_compaction_millis()
                    }
                    CompactionTransition::DdlReady(mut ddls) => {
                        self.handle_ddl_requests(&mut ddls).await
                    }
                    CompactionTransition::NoAction => {}
                }
            }
        }
    }

    pub(crate) async fn handle_compaction_pick_finished(
        &mut self,
        region_id: RegionId,
        request: CompactionPickFinished,
    ) where
        S: LogStore,
    {
        let Some(region) = self.regions.get_region(region_id) else {
            return;
        };
        // A terminal pick (canceled, no plan, or failed) may return the region
        // to idle and release DDLs fenced behind its picking phase.
        // Such a pick never produces an execution callback, so the worker must
        // execute the returned DDLs from this notification.
        let transition = self
            .compaction_scheduler
            .handle_compaction_pick_finished(
                request,
                &region.manifest_ctx,
                self.schema_metadata_manager.clone(),
            )
            .await;
        match transition {
            CompactionTransition::AutomaticFollowupScheduled => {
                // There will be a followup compaction so we need to update the
                // last schedule time to avoid frequent compactions.
                region.update_schedule_compaction_millis();
            }
            CompactionTransition::NoAction => {}
            CompactionTransition::DdlReady(mut pending_ddls) => {
                if !pending_ddls.is_empty() {
                    // Preserve the lifecycle order observed by listeners: compaction
                    // termination is visible before its dependent DDLs are dispatched.
                    self.listener.on_compaction_result_notified(region_id).await;
                    self.handle_ddl_requests(&mut pending_ddls).await;
                }
            }
        }
    }

    /// Handles compaction request submitted to region worker.
    pub(crate) async fn handle_compaction_request(
        &mut self,
        region_id: RegionId,
        req: RegionCompactRequest,
        mut sender: OptionOutputTx,
    ) {
        let Some(region) = self.regions.writable_region_or(region_id, &mut sender) else {
            return;
        };
        COMPACTION_REQUEST_COUNT.inc();
        let parallelism = req.parallelism.unwrap_or(1) as usize;
        match self.compaction_scheduler.schedule_manual_compaction(
            req.options,
            &region.version_control,
            &region.access_layer,
            sender,
            &region.manifest_ctx,
            self.schema_metadata_manager.clone(),
            parallelism,
            req.time_range,
        ) {
            // Ok(false) means the request was merged, queued or rejected; the
            // waiter was already notified or will be completed by the queued
            // cycle, so only a newly scheduled task is logged as success.
            Ok(true) => info!(
                "Successfully scheduled compaction task for region: {}",
                region_id
            ),
            Ok(false) => {}
            Err(e) => {
                error!(e; "Failed to schedule compaction task for region: {}", region_id);
            }
        }
    }

    /// Handles compaction finished, update region version and manifest, deleted compacted files.
    pub(crate) async fn handle_compaction_finished(
        &mut self,
        region_id: RegionId,
        mut request: CompactionFinished,
    ) where
        S: LogStore,
    {
        let region = match self.regions.get_region(region_id) {
            Some(region) => region,
            None => {
                request.on_failure(RegionNotFoundSnafu { region_id }.build());
                return;
            }
        };
        // Reject stale terminal results before applying their manifest edit.
        if !self
            .compaction_scheduler
            .is_current_execution(region_id, &request.execution)
        {
            request.on_failure(StaleCompactionExecutionSnafu { region_id }.build());
            return;
        }
        let execution = request.execution.clone();
        let made_progress = made_progress(
            request.edit.files_to_add.len(),
            request.edit.files_to_remove.len(),
        );

        region.version_control.apply_edit(
            Some(request.edit.clone()),
            &[],
            region.file_purger.clone(),
        );

        let index_build_file_metas = std::mem::take(&mut request.edit.files_to_add);

        // compaction finished.
        request.on_success();
        self.listener.on_compaction_result_notified(region_id).await;

        // In async mode, create indexes after compact if new files are created.
        if self.config.index.build_mode == IndexBuildMode::Async
            && !index_build_file_metas.is_empty()
        {
            self.handle_rebuild_index(
                BuildIndexRequest {
                    region_id,
                    build_type: IndexBuildType::Compact,
                    file_metas: index_build_file_metas,
                    result_senders: Vec::new(),
                },
                OptionOutputTx::new(None),
            )
            .await;
        }

        // Schedule next compaction if necessary.
        let transition = self
            .compaction_scheduler
            .on_execution_finished(
                region_id,
                &execution,
                &region.manifest_ctx,
                self.schema_metadata_manager.clone(),
                made_progress,
            )
            .await;
        match transition {
            CompactionTransition::AutomaticFollowupScheduled => {
                region.update_schedule_compaction_millis();
            }
            CompactionTransition::NoAction => {}
            CompactionTransition::DdlReady(mut pending_ddls) => {
                self.handle_ddl_requests(&mut pending_ddls).await;
            }
        }
    }

    /// When compaction fails, we simply log the error.
    pub(crate) async fn handle_compaction_failure(&mut self, req: CompactionFailed) {
        if self.regions.get_region(req.region_id).is_none() {
            return;
        }
        if !self
            .compaction_scheduler
            .is_current_execution(req.region_id, &req.execution)
        {
            debug!(
                "Ignores stale compaction failure for region {}: {:?}",
                req.region_id, req.err
            );
            return;
        }

        error!(req.err; "Failed to compact region: {}", req.region_id);
        self.compaction_scheduler
            .on_execution_failed(req.region_id, &req.execution, req.err);
    }

    /// Schedule compaction for the region if necessary.
    pub(crate) async fn schedule_compaction(&mut self, region: &MitoRegionRef) {
        if region.is_staging() || region.is_enter_staging() || region.is_truncating() {
            info!(
                "Region {} is staging, entering staging, or truncating, skip compaction",
                region.region_id
            );
            return;
        }
        let now = self.time_provider.current_time_millis();
        if now - region.last_schedule_compaction_millis()
            >= self.config.min_compaction_interval.as_millis() as i64
        {
            debug!(
                "minimal compaction interval time {:?} has passed, scheduling next compaction",
                self.config.min_compaction_interval
            );
            match self.compaction_scheduler.schedule_automatic_compaction(
                compact_request::Options::Regular(Default::default()),
                &region.version_control,
                &region.access_layer,
                &region.manifest_ctx,
                self.schema_metadata_manager.clone(),
            ) {
                Ok(true) => region.update_schedule_compaction_millis(),
                Ok(false) => {}
                Err(e) => {
                    error!(e; "Failed to schedule compaction for region: {}", region.region_id)
                }
            }
        }
    }
}

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

    #[test]
    fn test_nonempty_output_or_file_reduction_is_progress() {
        for (files_to_add, files_to_remove, expected) in [
            (3, 3, true),  // Equal-count rewrite.
            (1, 3, true),  // Ordinary reduction.
            (0, 3, true),  // Zero-output reduction.
            (0, 0, false), // Empty edit.
            (3, 2, true),  // Growth rewrite with output.
        ] {
            assert_eq!(
                expected,
                made_progress(files_to_add, files_to_remove),
                "files_to_remove: {files_to_remove}, files_to_add: {files_to_add}"
            );
        }
    }
}