linera_views/backends/
journaling.rs

1// Copyright (c) Zefchain Labs, Inc.
2// SPDX-License-Identifier: Apache-2.0
3
4//! Turns a `DirectKeyValueStore` into a `KeyValueStore` by adding journaling.
5//!
6//! Journaling aims to allow writing arbitrarily large batches of data in an atomic way.
7//! This is useful for database backends that limit the number of keys and/or the size of
8//! the data that can be written atomically (i.e. in the same database transaction).
9//!
10//! Journaling requires to set aside a range of keys to hold a possible "header" and an
11//! array of unwritten entries called "blocks".
12//!
13//! When a new batch to be written exceeds the capacity of the underlying storage, the
14//! "slow path" is taken: the batch of operations is first written into blocks, then the
15//! journal header is (atomically) updated to make the batch of updates persistent.
16//!
17//! Before any new read or write operation, if a journal is present, it must first be
18//! cleared. This is done by processing every block of the journal successively. Every
19//! time the data in a block are written, the journal header is updated in the same
20//! transaction to mark the block as processed.
21
22use serde::{de::DeserializeOwned, Deserialize, Serialize};
23use static_assertions as sa;
24use thiserror::Error;
25
26use crate::{
27    batch::{Batch, BatchValueWriter, DeletePrefixExpander, SimplifiedBatch},
28    store::{AdminKeyValueStore, ReadableKeyValueStore, WithError, WritableKeyValueStore},
29    views::MIN_VIEW_TAG,
30};
31
32/// The tag used for the journal stuff.
33const JOURNAL_TAG: u8 = 0;
34// To prevent collisions, the tag value 0 is reserved for journals.
35// The tags used by views must be greater or equal than `MIN_VIEW_TAG`.
36sa::const_assert!(JOURNAL_TAG < MIN_VIEW_TAG);
37
38/// Data type indicating that the database is not consistent
39#[derive(Error, Debug)]
40#[allow(missing_docs)]
41pub enum JournalConsistencyError {
42    #[error("The journal block could not be retrieved, it could be missing or corrupted.")]
43    FailureToRetrieveJournalBlock,
44
45    #[error("Refusing to use the journal without exclusive database access to the root object.")]
46    JournalRequiresExclusiveAccess,
47}
48
49#[repr(u8)]
50enum KeyTag {
51    /// Prefix for the storing of the header of the journal.
52    Journal = 1,
53    /// Prefix for the block entry.
54    Entry,
55}
56
57fn get_journaling_key(tag: u8, pos: u32) -> Result<Vec<u8>, bcs::Error> {
58    let mut key = vec![JOURNAL_TAG];
59    key.extend([tag]);
60    bcs::serialize_into(&mut key, &pos)?;
61    Ok(key)
62}
63
64/// Low-level, asynchronous direct write key-value operations with simplified batch
65#[cfg_attr(not(web), trait_variant::make(Send + Sync))]
66pub trait DirectWritableKeyValueStore: WithError {
67    /// The maximal number of items in a batch.
68    const MAX_BATCH_SIZE: usize;
69
70    /// The maximal number of bytes of a batch.
71    const MAX_BATCH_TOTAL_SIZE: usize;
72
73    /// The maximal size of values that can be stored.
74    const MAX_VALUE_SIZE: usize;
75
76    /// The batch type.
77    type Batch: SimplifiedBatch + Serialize + DeserializeOwned + Default;
78
79    /// Writes the batch to the database.
80    async fn write_batch(&self, batch: Self::Batch) -> Result<(), Self::Error>;
81}
82
83/// Low-level, asynchronous direct read/write key-value operations with simplified batch
84pub trait DirectKeyValueStore:
85    ReadableKeyValueStore + DirectWritableKeyValueStore + AdminKeyValueStore
86{
87}
88
89impl<T> DirectKeyValueStore for T where
90    T: ReadableKeyValueStore + DirectWritableKeyValueStore + AdminKeyValueStore
91{
92}
93
94/// The header that contains the current state of the journal.
95#[derive(Serialize, Deserialize, Debug, Default)]
96struct JournalHeader {
97    block_count: u32,
98}
99
100/// A journaling Key Value Store built from an inner [`DirectKeyValueStore`].
101#[derive(Clone)]
102pub struct JournalingKeyValueStore<K> {
103    /// The inner store.
104    store: K,
105    /// Whether we have exclusive R/W access to the keys under root key.
106    has_exclusive_access: bool,
107}
108
109impl<K> DeletePrefixExpander for &JournalingKeyValueStore<K>
110where
111    K: DirectKeyValueStore,
112{
113    type Error = K::Error;
114
115    async fn expand_delete_prefix(&self, key_prefix: &[u8]) -> Result<Vec<Vec<u8>>, Self::Error> {
116        self.store.find_keys_by_prefix(key_prefix).await
117    }
118}
119
120impl<K> WithError for JournalingKeyValueStore<K>
121where
122    K: WithError,
123{
124    type Error = K::Error;
125}
126
127impl<K> ReadableKeyValueStore for JournalingKeyValueStore<K>
128where
129    K: ReadableKeyValueStore,
130    K::Error: From<JournalConsistencyError>,
131{
132    /// The size constant do not change
133    const MAX_KEY_SIZE: usize = K::MAX_KEY_SIZE;
134
135    /// The read stuff does not change
136    fn max_stream_queries(&self) -> usize {
137        self.store.max_stream_queries()
138    }
139
140    async fn read_value_bytes(&self, key: &[u8]) -> Result<Option<Vec<u8>>, Self::Error> {
141        self.store.read_value_bytes(key).await
142    }
143
144    async fn contains_key(&self, key: &[u8]) -> Result<bool, Self::Error> {
145        self.store.contains_key(key).await
146    }
147
148    async fn contains_keys(&self, keys: Vec<Vec<u8>>) -> Result<Vec<bool>, Self::Error> {
149        self.store.contains_keys(keys).await
150    }
151
152    async fn read_multi_values_bytes(
153        &self,
154        keys: Vec<Vec<u8>>,
155    ) -> Result<Vec<Option<Vec<u8>>>, Self::Error> {
156        self.store.read_multi_values_bytes(keys).await
157    }
158
159    async fn find_keys_by_prefix(&self, key_prefix: &[u8]) -> Result<Vec<Vec<u8>>, Self::Error> {
160        self.store.find_keys_by_prefix(key_prefix).await
161    }
162
163    async fn find_key_values_by_prefix(
164        &self,
165        key_prefix: &[u8],
166    ) -> Result<Vec<(Vec<u8>, Vec<u8>)>, Self::Error> {
167        self.store.find_key_values_by_prefix(key_prefix).await
168    }
169}
170
171impl<K> AdminKeyValueStore for JournalingKeyValueStore<K>
172where
173    K: AdminKeyValueStore,
174{
175    type Config = K::Config;
176
177    fn get_name() -> String {
178        format!("journaling {}", K::get_name())
179    }
180
181    async fn connect(config: &Self::Config, namespace: &str) -> Result<Self, Self::Error> {
182        let store = K::connect(config, namespace).await?;
183        Ok(Self {
184            store,
185            has_exclusive_access: false,
186        })
187    }
188
189    fn open_exclusive(&self, root_key: &[u8]) -> Result<Self, Self::Error> {
190        let store = self.store.open_exclusive(root_key)?;
191        Ok(Self {
192            store,
193            has_exclusive_access: true,
194        })
195    }
196
197    async fn list_all(config: &Self::Config) -> Result<Vec<String>, Self::Error> {
198        K::list_all(config).await
199    }
200
201    async fn list_root_keys(
202        config: &Self::Config,
203        namespace: &str,
204    ) -> Result<Vec<Vec<u8>>, Self::Error> {
205        K::list_root_keys(config, namespace).await
206    }
207
208    async fn delete_all(config: &Self::Config) -> Result<(), Self::Error> {
209        K::delete_all(config).await
210    }
211
212    async fn exists(config: &Self::Config, namespace: &str) -> Result<bool, Self::Error> {
213        K::exists(config, namespace).await
214    }
215
216    async fn create(config: &Self::Config, namespace: &str) -> Result<(), Self::Error> {
217        K::create(config, namespace).await
218    }
219
220    async fn delete(config: &Self::Config, namespace: &str) -> Result<(), Self::Error> {
221        K::delete(config, namespace).await
222    }
223}
224
225impl<K> WritableKeyValueStore for JournalingKeyValueStore<K>
226where
227    K: DirectKeyValueStore,
228    K::Error: From<JournalConsistencyError>,
229{
230    /// The size constant do not change
231    const MAX_VALUE_SIZE: usize = K::MAX_VALUE_SIZE;
232
233    async fn write_batch(&self, batch: Batch) -> Result<(), Self::Error> {
234        let batch = K::Batch::from_batch(self, batch).await?;
235        if Self::is_fastpath_feasible(&batch) {
236            self.store.write_batch(batch).await
237        } else {
238            if !self.has_exclusive_access {
239                return Err(JournalConsistencyError::JournalRequiresExclusiveAccess.into());
240            }
241            let header = self.write_journal(batch).await?;
242            self.coherently_resolve_journal(header).await
243        }
244    }
245
246    async fn clear_journal(&self) -> Result<(), Self::Error> {
247        let key = get_journaling_key(KeyTag::Journal as u8, 0)?;
248        let value = self.read_value::<JournalHeader>(&key).await?;
249        if let Some(header) = value {
250            self.coherently_resolve_journal(header).await?;
251        }
252        Ok(())
253    }
254}
255
256impl<K> JournalingKeyValueStore<K>
257where
258    K: DirectKeyValueStore,
259    K::Error: From<JournalConsistencyError>,
260{
261    /// Resolves the pending operations that were previously stored in the database
262    /// journal.
263    ///
264    /// For each block processed, we atomically update the journal header as well. When
265    /// the last block is processed, this atomically clears the journal and make the store
266    /// finally available again (for the range of keys managed by the journal).
267    ///
268    /// This function respects the constraints of the underlying key-value store `K` if
269    /// the following conditions are met:
270    ///
271    /// (1) each block contains at most `K::MAX_BATCH_SIZE - 2` operations;
272    ///
273    /// (2) the total size of the all operations in a block doesn't exceed:
274    /// `K::MAX_BATCH_TOTAL_SIZE - sizeof(block_key) - sizeof(header_key) - sizeof(bcs_header)`
275    ///
276    /// (3) every operation in a block satisfies the constraints on individual database
277    /// operations represented by `K::MAX_KEY_SIZE` and `K::MAX_VALUE_SIZE`.
278    ///
279    /// (4) `block_key` and `header_key` don't exceed `K::MAX_KEY_SIZE` and `bcs_header`
280    /// doesn't exceed `K::MAX_VALUE_SIZE`.
281    async fn coherently_resolve_journal(&self, mut header: JournalHeader) -> Result<(), K::Error> {
282        let header_key = get_journaling_key(KeyTag::Journal as u8, 0)?;
283        while header.block_count > 0 {
284            let block_key = get_journaling_key(KeyTag::Entry as u8, header.block_count - 1)?;
285            // Read the batch of updates (aka. "block") previously saved in the journal.
286            let mut batch = self
287                .store
288                .read_value::<K::Batch>(&block_key)
289                .await?
290                .ok_or(JournalConsistencyError::FailureToRetrieveJournalBlock)?;
291            // Execute the block and delete it from the journal atomically.
292            batch.add_delete(block_key);
293            header.block_count -= 1;
294            if header.block_count > 0 {
295                let value = bcs::to_bytes(&header)?;
296                batch.add_insert(header_key.clone(), value);
297            } else {
298                batch.add_delete(header_key.clone());
299            }
300            self.store.write_batch(batch).await?;
301        }
302        Ok(())
303    }
304
305    /// Writes the content of `batch` to the journal as a succession of blocks that can be
306    /// interpreted later by `coherently_resolve_journal`.
307    ///
308    /// Starting with a batch of operations that is typically too large to be executed in
309    /// one go (see `is_fastpath_feasible()` below), the goal of this function is to split
310    /// the batch into smaller blocks so that `coherently_resolve_journal` respects the
311    /// constraints of the underlying key-value store (see analysis above).
312    ///
313    /// For efficiency reasons, we write as many blocks as possible in each "transaction"
314    /// batch, using one write-operation per block. Then we also update the journal header
315    /// with the final number of blocks.
316    ///
317    /// As a result, the constraints of the underlying database are respected if the
318    /// following conditions are met while a "transaction" batch is being built:
319    ///
320    /// (1) The number of blocks per transaction doesn't exceed `K::MAX_BATCH_SIZE`.
321    /// But it is perfectly possible to have `K::MAX_BATCH_SIZE = usize::MAX`.
322    ///
323    /// (2) The total size of BCS-serialized blocks together with their corresponding keys
324    /// does not exceed `K::MAX_BATCH_TOTAL_SIZE`.
325    ///
326    /// (3) The size of each BCS-serialized block doesn't exceed `K::MAX_VALUE_SIZE`.
327    ///
328    /// (4) When processing a journal block, we have to do two other operations.
329    ///   (a) removing the existing block. The cost is `key_len`.
330    ///   (b) updating or removing the journal. The cost is `key_len + header_value_len`
331    ///       or `key_len`. An upper bound is thus
332    ///       `journal_len_upper_bound = key_len + header_value_len`.
333    ///   Thus the following has to be taken as upper bound on the block size:
334    ///   `K::MAX_BATCH_TOTAL_SIZE - key_len - journal_len_upper_bound`.
335    ///
336    /// NOTE:
337    /// * Since a block must contain at least one operation and M bytes of the
338    ///   serialization overhead (typically M is 2 or 3 bytes of vector sizes), condition (3)
339    ///   requires that each operation in the original batch satisfies:
340    ///   `sizeof(key) + sizeof(value) + M <= K::MAX_VALUE_SIZE`
341    ///
342    /// * Similarly, a transaction must contain at least one block so it is desirable that
343    ///   the maximum size of a block insertion `1 + sizeof(block_key) + K::MAX_VALUE_SIZE`
344    ///   plus M bytes of overhead doesn't exceed the threshold of condition (2).
345    async fn write_journal(&self, batch: K::Batch) -> Result<JournalHeader, K::Error> {
346        let header_key = get_journaling_key(KeyTag::Journal as u8, 0)?;
347        let key_len = header_key.len();
348        let header_value_len = bcs::serialized_size(&JournalHeader::default())?;
349        let journal_len_upper_bound = key_len + header_value_len;
350        // Each block in a transaction comes with a key.
351        let max_transaction_size = K::MAX_BATCH_TOTAL_SIZE;
352        let max_block_size = std::cmp::min(
353            K::MAX_VALUE_SIZE,
354            K::MAX_BATCH_TOTAL_SIZE - key_len - journal_len_upper_bound,
355        );
356
357        let mut iter = batch.into_iter();
358        let mut block_batch = K::Batch::default();
359        let mut block_size = 0;
360        let mut block_count = 0;
361        let mut transaction_batch = K::Batch::default();
362        let mut transaction_size = 0;
363        while iter.write_next_value(&mut block_batch, &mut block_size)? {
364            let (block_flush, transaction_flush) = {
365                if iter.is_empty() || transaction_batch.len() == K::MAX_BATCH_SIZE - 1 {
366                    (true, true)
367                } else {
368                    let next_block_size = iter
369                        .next_batch_size(&block_batch, block_size)?
370                        .expect("iter is not empty");
371                    let next_transaction_size = transaction_size + next_block_size + key_len;
372                    let transaction_flush = next_transaction_size > max_transaction_size;
373                    let block_flush = transaction_flush
374                        || block_batch.len() == K::MAX_BATCH_SIZE - 2
375                        || next_block_size > max_block_size;
376                    (block_flush, transaction_flush)
377                }
378            };
379            if block_flush {
380                block_size += block_batch.overhead_size();
381                let value = bcs::to_bytes(&block_batch)?;
382                block_batch = K::Batch::default();
383                assert_eq!(value.len(), block_size);
384                let key = get_journaling_key(KeyTag::Entry as u8, block_count)?;
385                transaction_batch.add_insert(key, value);
386                block_count += 1;
387                transaction_size += block_size + key_len;
388                block_size = 0;
389            }
390            if transaction_flush {
391                let batch = std::mem::take(&mut transaction_batch);
392                self.store.write_batch(batch).await?;
393                transaction_size = 0;
394            }
395        }
396        let header = JournalHeader { block_count };
397        if block_count > 0 {
398            let value = bcs::to_bytes(&header)?;
399            let mut batch = K::Batch::default();
400            batch.add_insert(header_key, value);
401            self.store.write_batch(batch).await?;
402        }
403        Ok(header)
404    }
405
406    fn is_fastpath_feasible(batch: &K::Batch) -> bool {
407        batch.len() <= K::MAX_BATCH_SIZE && batch.num_bytes() <= K::MAX_BATCH_TOTAL_SIZE
408    }
409}
410
411impl<K> JournalingKeyValueStore<K> {
412    /// Creates a new journaling store.
413    pub fn new(store: K) -> Self {
414        Self {
415            store,
416            has_exclusive_access: false,
417        }
418    }
419}
linera_views/backends/journaling.rs

linera_views/backends/
journaling.rs
