mito2/region/
version.rs

1// Copyright 2023 Greptime Team
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15//! Version control of mito engine.
16//!
17//! Version is an immutable snapshot of region's metadata.
18//!
19//! To read latest data from `VersionControl`, we should
20//! 1. Acquire `Version` from `VersionControl`.
21//! 2. Then acquire last sequence.
22//!
23//! Reason: data may be flushed/compacted and some data with old sequence may be removed
24//! and became invisible between step 1 and 2, so need to acquire version at first.
25
26use std::sync::{Arc, RwLock};
27use std::time::Duration;
28
29use common_telemetry::info;
30use store_api::metadata::RegionMetadataRef;
31use store_api::storage::SequenceNumber;
32
33use crate::error::Result;
34use crate::manifest::action::RegionEdit;
35use crate::memtable::time_partition::{TimePartitions, TimePartitionsRef};
36use crate::memtable::version::{MemtableVersion, MemtableVersionRef};
37use crate::memtable::{MemtableBuilderRef, MemtableId};
38use crate::region::options::RegionOptions;
39use crate::sst::file::FileMeta;
40use crate::sst::file_purger::FilePurgerRef;
41use crate::sst::version::{SstVersion, SstVersionRef};
42use crate::wal::EntryId;
43
44/// Controls metadata and sequence numbers for a region.
45///
46/// It manages metadata in a copy-on-write fashion. Any modification to a region's metadata
47/// will generate a new [Version].
48#[derive(Debug)]
49pub(crate) struct VersionControl {
50    data: RwLock<VersionControlData>,
51}
52
53impl VersionControl {
54    /// Returns a new [VersionControl] with specific `version`.
55    pub(crate) fn new(version: Version) -> VersionControl {
56        // Initialize sequence and entry id from flushed sequence and entry id.
57        let (flushed_sequence, flushed_entry_id) =
58            (version.flushed_sequence, version.flushed_entry_id);
59        VersionControl {
60            data: RwLock::new(VersionControlData {
61                version: Arc::new(version),
62                committed_sequence: flushed_sequence,
63                last_entry_id: flushed_entry_id,
64                is_dropped: false,
65            }),
66        }
67    }
68
69    /// Returns current copy of data.
70    pub(crate) fn current(&self) -> VersionControlData {
71        self.data.read().unwrap().clone()
72    }
73
74    /// Updates committed sequence and entry id.
75    pub(crate) fn set_sequence_and_entry_id(&self, seq: SequenceNumber, entry_id: EntryId) {
76        let mut data = self.data.write().unwrap();
77        data.committed_sequence = seq;
78        data.last_entry_id = entry_id;
79    }
80
81    /// Updates last entry id.
82    pub(crate) fn set_entry_id(&self, entry_id: EntryId) {
83        let mut data = self.data.write().unwrap();
84        data.last_entry_id = entry_id;
85    }
86
87    /// Sequence number of last committed data.
88    pub(crate) fn committed_sequence(&self) -> SequenceNumber {
89        self.data.read().unwrap().committed_sequence
90    }
91
92    /// Freezes the mutable memtable if it is not empty.
93    pub(crate) fn freeze_mutable(&self) -> Result<()> {
94        let version = self.current().version;
95        let time_window = version.compaction_time_window;
96
97        let Some(new_memtables) = version
98            .memtables
99            .freeze_mutable(&version.metadata, time_window)?
100        else {
101            return Ok(());
102        };
103
104        // Create a new version with memtable switched.
105        let new_version = Arc::new(
106            VersionBuilder::from_version(version)
107                .memtables(new_memtables)
108                .build(),
109        );
110
111        let mut version_data = self.data.write().unwrap();
112        version_data.version = new_version;
113
114        Ok(())
115    }
116
117    /// Applies region option changes and generates a new version.
118    pub(crate) fn alter_options(&self, options: RegionOptions) {
119        let version = self.current().version;
120        let new_version = Arc::new(
121            VersionBuilder::from_version(version)
122                .options(options)
123                .build(),
124        );
125        let mut version_data = self.data.write().unwrap();
126        version_data.version = new_version;
127    }
128
129    /// Apply edit to current version.
130    pub(crate) fn apply_edit(
131        &self,
132        edit: RegionEdit,
133        memtables_to_remove: &[MemtableId],
134        purger: FilePurgerRef,
135    ) {
136        let version = self.current().version;
137        let new_version = Arc::new(
138            VersionBuilder::from_version(version)
139                .apply_edit(edit, purger)
140                .remove_memtables(memtables_to_remove)
141                .build(),
142        );
143
144        let mut version_data = self.data.write().unwrap();
145        version_data.version = new_version;
146    }
147
148    /// Mark all opened files as deleted and set the delete marker in [VersionControlData]
149    pub(crate) fn mark_dropped(&self, memtable_builder: &MemtableBuilderRef) {
150        let version = self.current().version;
151        let part_duration = Some(version.memtables.mutable.part_duration());
152        let next_memtable_id = version.memtables.mutable.next_memtable_id();
153        let new_mutable = Arc::new(TimePartitions::new(
154            version.metadata.clone(),
155            memtable_builder.clone(),
156            next_memtable_id,
157            part_duration,
158        ));
159
160        let mut data = self.data.write().unwrap();
161        data.is_dropped = true;
162        data.version.ssts.mark_all_deleted();
163        // Reset version so we can release the reference to memtables and SSTs.
164        let new_version =
165            Arc::new(VersionBuilder::new(version.metadata.clone(), new_mutable).build());
166        data.version = new_version;
167    }
168
169    /// Alter schema of the region.
170    ///
171    /// It replaces existing mutable memtable with a memtable that uses the
172    /// new schema. Memtables of the version must be empty.
173    pub(crate) fn alter_schema(&self, metadata: RegionMetadataRef, builder: &MemtableBuilderRef) {
174        let version = self.current().version;
175        let part_duration = Some(version.memtables.mutable.part_duration());
176        let next_memtable_id = version.memtables.mutable.next_memtable_id();
177        let new_mutable = Arc::new(TimePartitions::new(
178            metadata.clone(),
179            builder.clone(),
180            next_memtable_id,
181            part_duration,
182        ));
183        debug_assert!(version.memtables.mutable.is_empty());
184        debug_assert!(version.memtables.immutables().is_empty());
185        let new_version = Arc::new(
186            VersionBuilder::from_version(version)
187                .metadata(metadata)
188                .memtables(MemtableVersion::new(new_mutable))
189                .build(),
190        );
191
192        let mut version_data = self.data.write().unwrap();
193        version_data.version = new_version;
194    }
195
196    /// Truncate current version.
197    pub(crate) fn truncate(
198        &self,
199        truncated_entry_id: EntryId,
200        truncated_sequence: SequenceNumber,
201        memtable_builder: &MemtableBuilderRef,
202    ) {
203        let version = self.current().version;
204
205        let part_duration = version.memtables.mutable.part_duration();
206        let next_memtable_id = version.memtables.mutable.next_memtable_id();
207        let new_mutable = Arc::new(TimePartitions::new(
208            version.metadata.clone(),
209            memtable_builder.clone(),
210            next_memtable_id,
211            Some(part_duration),
212        ));
213        let new_version = Arc::new(
214            VersionBuilder::new(version.metadata.clone(), new_mutable)
215                .flushed_entry_id(truncated_entry_id)
216                .flushed_sequence(truncated_sequence)
217                .truncated_entry_id(Some(truncated_entry_id))
218                .build(),
219        );
220
221        let mut version_data = self.data.write().unwrap();
222        version_data.version.ssts.mark_all_deleted();
223        version_data.version = new_version;
224    }
225
226    /// Overwrites the current version with a new version.
227    pub(crate) fn overwrite_current(&self, version: VersionRef) {
228        let mut version_data = self.data.write().unwrap();
229        version_data.version = version;
230    }
231}
232
233pub(crate) type VersionControlRef = Arc<VersionControl>;
234
235/// Data of [VersionControl].
236#[derive(Debug, Clone)]
237pub(crate) struct VersionControlData {
238    /// Latest version.
239    pub(crate) version: VersionRef,
240    /// Sequence number of last committed data.
241    ///
242    /// Starts from 1 (zero means no data).
243    pub(crate) committed_sequence: SequenceNumber,
244    /// Last WAL entry Id.
245    ///
246    /// Starts from 1 (zero means no data).
247    pub(crate) last_entry_id: EntryId,
248    /// Marker of whether this region is dropped/dropping
249    pub(crate) is_dropped: bool,
250}
251
252impl VersionControlData {
253    /// Approximate timeseries count in current version.
254    pub(crate) fn series_count(&self) -> usize {
255        self.version.memtables.mutable.series_count()
256    }
257}
258
259/// Static metadata of a region.
260#[derive(Clone, Debug)]
261pub(crate) struct Version {
262    /// Metadata of the region.
263    ///
264    /// Altering metadata isn't frequent, storing metadata in Arc to allow sharing
265    /// metadata and reuse metadata when creating a new `Version`.
266    pub(crate) metadata: RegionMetadataRef,
267    /// Mutable and immutable memtables.
268    ///
269    /// Wrapped in Arc to make clone of `Version` much cheaper.
270    pub(crate) memtables: MemtableVersionRef,
271    /// SSTs of the region.
272    pub(crate) ssts: SstVersionRef,
273    /// Inclusive max WAL entry id of flushed data.
274    pub(crate) flushed_entry_id: EntryId,
275    /// Inclusive max sequence of flushed data.
276    pub(crate) flushed_sequence: SequenceNumber,
277    /// Latest entry id during the truncating table.
278    ///
279    /// Used to check if it is a flush task during the truncating table.
280    pub(crate) truncated_entry_id: Option<EntryId>,
281    /// Inferred compaction time window from flush.
282    ///
283    /// If compaction options contain a time window, it will overwrite this value
284    /// when creating a new version from the [VersionBuilder].
285    pub(crate) compaction_time_window: Option<Duration>,
286    /// Options of the region.
287    pub(crate) options: RegionOptions,
288}
289
290pub(crate) type VersionRef = Arc<Version>;
291
292/// Version builder.
293pub(crate) struct VersionBuilder {
294    metadata: RegionMetadataRef,
295    memtables: MemtableVersionRef,
296    ssts: SstVersionRef,
297    flushed_entry_id: EntryId,
298    flushed_sequence: SequenceNumber,
299    truncated_entry_id: Option<EntryId>,
300    compaction_time_window: Option<Duration>,
301    options: RegionOptions,
302}
303
304impl VersionBuilder {
305    /// Returns a new builder.
306    pub(crate) fn new(metadata: RegionMetadataRef, mutable: TimePartitionsRef) -> Self {
307        VersionBuilder {
308            metadata,
309            memtables: Arc::new(MemtableVersion::new(mutable)),
310            ssts: Arc::new(SstVersion::new()),
311            flushed_entry_id: 0,
312            flushed_sequence: 0,
313            truncated_entry_id: None,
314            compaction_time_window: None,
315            options: RegionOptions::default(),
316        }
317    }
318
319    /// Returns a new builder from an existing version.
320    pub(crate) fn from_version(version: VersionRef) -> Self {
321        VersionBuilder {
322            metadata: version.metadata.clone(),
323            memtables: version.memtables.clone(),
324            ssts: version.ssts.clone(),
325            flushed_entry_id: version.flushed_entry_id,
326            flushed_sequence: version.flushed_sequence,
327            truncated_entry_id: version.truncated_entry_id,
328            compaction_time_window: version.compaction_time_window,
329            options: version.options.clone(),
330        }
331    }
332
333    /// Sets memtables.
334    pub(crate) fn memtables(mut self, memtables: MemtableVersion) -> Self {
335        self.memtables = Arc::new(memtables);
336        self
337    }
338
339    /// Sets metadata.
340    pub(crate) fn metadata(mut self, metadata: RegionMetadataRef) -> Self {
341        self.metadata = metadata;
342        self
343    }
344
345    /// Sets flushed entry id.
346    pub(crate) fn flushed_entry_id(mut self, entry_id: EntryId) -> Self {
347        self.flushed_entry_id = entry_id;
348        self
349    }
350
351    /// Sets flushed sequence.
352    pub(crate) fn flushed_sequence(mut self, sequence: SequenceNumber) -> Self {
353        self.flushed_sequence = sequence;
354        self
355    }
356
357    /// Sets truncated entty id.
358    pub(crate) fn truncated_entry_id(mut self, entry_id: Option<EntryId>) -> Self {
359        self.truncated_entry_id = entry_id;
360        self
361    }
362
363    /// Sets compaction time window.
364    pub(crate) fn compaction_time_window(mut self, window: Option<Duration>) -> Self {
365        self.compaction_time_window = window;
366        self
367    }
368
369    /// Sets options.
370    pub(crate) fn options(mut self, options: RegionOptions) -> Self {
371        self.options = options;
372        self
373    }
374
375    /// Apply edit to the builder.
376    pub(crate) fn apply_edit(mut self, edit: RegionEdit, file_purger: FilePurgerRef) -> Self {
377        if let Some(entry_id) = edit.flushed_entry_id {
378            self.flushed_entry_id = self.flushed_entry_id.max(entry_id);
379        }
380        if let Some(sequence) = edit.flushed_sequence {
381            self.flushed_sequence = self.flushed_sequence.max(sequence);
382        }
383        if let Some(window) = edit.compaction_time_window {
384            self.compaction_time_window = Some(window);
385        }
386        if !edit.files_to_add.is_empty() || !edit.files_to_remove.is_empty() {
387            let mut ssts = (*self.ssts).clone();
388            ssts.add_files(file_purger, edit.files_to_add.into_iter());
389            ssts.remove_files(edit.files_to_remove.into_iter());
390            self.ssts = Arc::new(ssts);
391        }
392
393        self
394    }
395
396    /// Remove memtables from the builder.
397    pub(crate) fn remove_memtables(mut self, ids: &[MemtableId]) -> Self {
398        if !ids.is_empty() {
399            let mut memtables = (*self.memtables).clone();
400            memtables.remove_memtables(ids);
401            self.memtables = Arc::new(memtables);
402        }
403        self
404    }
405
406    /// Add files to the builder.
407    pub(crate) fn add_files(
408        mut self,
409        file_purger: FilePurgerRef,
410        files: impl Iterator<Item = FileMeta>,
411    ) -> Self {
412        let mut ssts = (*self.ssts).clone();
413        ssts.add_files(file_purger, files);
414        self.ssts = Arc::new(ssts);
415
416        self
417    }
418
419    /// Builds a new [Version] from the builder.
420    /// It overwrites the window size by compaction option.
421    pub(crate) fn build(self) -> Version {
422        let compaction_time_window = self
423            .options
424            .compaction
425            .time_window()
426            .or(self.compaction_time_window);
427        if self.compaction_time_window.is_some()
428            && compaction_time_window != self.compaction_time_window
429        {
430            info!(
431                "VersionBuilder overwrites region compaction time window from {:?} to {:?}, region: {}",
432                self.compaction_time_window,
433                compaction_time_window,
434                self.metadata.region_id
435            );
436        }
437
438        Version {
439            metadata: self.metadata,
440            memtables: self.memtables,
441            ssts: self.ssts,
442            flushed_entry_id: self.flushed_entry_id,
443            flushed_sequence: self.flushed_sequence,
444            truncated_entry_id: self.truncated_entry_id,
445            compaction_time_window,
446            options: self.options,
447        }
448    }
449}