1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
//! Environment and database flags.
use bitflags::bitflags;
use ffi::*;
use std::str::FromStr;
/// MDBX sync mode
#[derive(PartialEq, Eq, Clone, Copy, Debug, Default)]
pub enum SyncMode {
/// Default robust and durable sync mode.
/// Metadata is written and flushed to disk after a data is written and flushed, which
/// guarantees the integrity of the database in the event of a crash at any time.
#[default]
Durable,
/// Don't sync the meta-page after commit.
///
/// Flush system buffers to disk only once per transaction commit, omit the metadata flush.
/// Defer that until the system flushes files to disk, or next non-read-only commit or
/// [`Environment::sync()`](crate::Environment::sync). Depending on the platform and
/// hardware, with [`SyncMode::NoMetaSync`] you may get a doubling of write performance.
///
/// This trade-off maintains database integrity, but a system crash may undo the last committed
/// transaction. I.e. it preserves the ACPI (atomicity, consistency, isolation) but not D
/// (durability) database property.
NoMetaSync,
/// Don't sync anything but keep previous steady commits.
///
/// [`SyncMode::UtterlyNoSync`] the [`SyncMode::SafeNoSync`] flag disable similarly flush
/// system buffers to disk when committing a transaction. But there is a huge difference in
/// how are recycled the MVCC snapshots corresponding to previous "steady" transactions
/// (see below).
///
/// With [`crate::sys::EnvironmentKind::WriteMap`] the [`SyncMode::SafeNoSync`] instructs MDBX to
/// use asynchronous mmap-flushes to disk. Asynchronous mmap-flushes means that actually
/// all writes will scheduled and performed by operation system on it own manner, i.e.
/// unordered. MDBX itself just notify operating system that it would be nice to write data
/// to disk, but no more.
///
/// Depending on the platform and hardware, with [`SyncMode::SafeNoSync`] you may get a
/// multiple increase of write performance, even 10 times or more.
///
/// In contrast to [`SyncMode::UtterlyNoSync`] mode, with [`SyncMode::SafeNoSync`] flag MDBX
/// will keeps untouched pages within B-tree of the last transaction "steady" which was
/// synced to disk completely. This has big implications for both data durability and
/// (unfortunately) performance:
/// - A system crash can't corrupt the database, but you will lose the last transactions;
/// because MDBX will rollback to last steady commit since it kept explicitly.
/// - The last steady transaction makes an effect similar to "long-lived" read transaction
/// since prevents reuse of pages freed by newer write transactions, thus the any data
/// changes will be placed in newly allocated pages.
/// - To avoid rapid database growth, the system will sync data and issue a steady commit-point
/// to resume reuse pages, each time there is insufficient space and before increasing the
/// size of the file on disk.
///
/// In other words, with
/// [`SyncMode::SafeNoSync`] flag MDBX protects you from the whole database corruption, at the
/// cost increasing database size and/or number of disk IOPs. So, [`SyncMode::SafeNoSync`]
/// flag could be used with [`Environment::sync()`](crate::Environment::sync) as alternatively
/// for batch committing or nested transaction (in some cases).
///
/// The number and volume of disk IOPs with [`SyncMode::SafeNoSync`] flag will exactly the
/// as without any no-sync flags. However, you should expect a larger process's work set
/// and significantly worse a locality of reference, due to the more intensive allocation
/// of previously unused pages and increase the size of the database.
SafeNoSync,
/// Don't sync anything and wipe previous steady commits.
///
/// Don't flush system buffers to disk when committing a transaction.
/// This optimization means a system crash can corrupt the database, if buffers are not yet
/// flushed to disk. Depending on the platform and hardware, with [`SyncMode::UtterlyNoSync`]
/// you may get a multiple increase of write performance, even 100 times or more.
///
/// If the filesystem preserves write order (which is rare and never provided unless explicitly
/// noted) and the [`WriteMap`](crate::sys::EnvironmentKind::WriteMap) and
/// [`EnvironmentFlags::liforeclaim`] flags are not used, then a system crash can't corrupt
/// the database, but you can lose the last transactions, if at least one buffer is not yet
/// flushed to disk. The risk is governed by how often the system flushes dirty buffers to
/// disk and how often [`Environment::sync()`](crate::Environment::sync) is called. So,
/// transactions exhibit ACPI (atomicity, consistency, isolation) properties and only lose D
/// (durability). I.e. database integrity is maintained, but a system crash may undo the
/// final transactions.
///
/// Otherwise, if the filesystem not preserves write order (which is typically) or
/// [`WriteMap`](crate::sys::EnvironmentKind::WriteMap) or [`EnvironmentFlags::liforeclaim`] flags
/// are used, you should expect the corrupted database after a system crash.
///
/// So, most important thing about [`SyncMode::UtterlyNoSync`]:
/// - A system crash immediately after commit the write transaction high likely lead to
/// database corruption.
/// - Successful completion of [`Environment::sync(force=true`)](crate::Environment::sync)
/// after one or more committed transactions guarantees consistency and durability.
/// - BUT by committing two or more transactions you back database into a weak state, in which
/// a system crash may lead to database corruption! In case single transaction after
/// [`Environment::sync()`](crate::Environment::sync), you may lose transaction itself, but
/// not a whole database.
///
/// Nevertheless, [`SyncMode::UtterlyNoSync`] provides "weak" durability in
/// case of an application crash (but no durability on system failure), and therefore may
/// be very useful in scenarios where data durability is not required over a system failure
/// (e.g for short-lived data), or if you can take such risk.
UtterlyNoSync,
}
/// Environment mode (read-only or read-write).
#[derive(Clone, Copy, Debug)]
pub enum Mode {
/// Read-only mode.
ReadOnly,
/// Read-write mode with specified sync mode.
ReadWrite {
/// Sync mode for write transactions.
sync_mode: SyncMode,
},
}
impl Default for Mode {
fn default() -> Self {
Self::ReadWrite { sync_mode: SyncMode::default() }
}
}
impl From<Mode> for EnvironmentFlags {
fn from(mode: Mode) -> Self {
Self { mode, ..Default::default() }
}
}
impl FromStr for SyncMode {
type Err = String;
fn from_str(s: &str) -> Result<Self, Self::Err> {
let val = s.trim().to_ascii_lowercase();
match val.as_str() {
"durable" => Ok(Self::Durable),
"safe-no-sync" | "safenosync" | "safe_no_sync" => Ok(Self::SafeNoSync),
_ => Err(format!(
"invalid value '{s}' for sync mode. valid values: durable, safe-no-sync"
)),
}
}
}
/// Environment opening flags.
#[derive(Clone, Copy, Debug, Default)]
pub struct EnvironmentFlags {
/// Do not create a subdirectory for the environment.
pub no_sub_dir: bool,
/// Open the environment in exclusive mode.
pub exclusive: bool,
/// Flag is intended to open an existing sub-database which was created with unknown flags
/// In such cases, instead of returning the `MDBX_INCOMPATIBLE` error, the sub-database will be
/// opened with flags which it was created, and then an application could determine the actual
/// flags.
pub accede: bool,
/// Environment mode (RO/RW).
pub mode: Mode,
/// Disable read-ahead.
pub no_rdahead: bool,
/// Do not initialize memory before writing to it.
pub no_meminit: bool,
/// Enable page coalescing.
pub coalesce: bool,
/// Enable LIFO reclamation of pages.
pub liforeclaim: bool,
}
impl EnvironmentFlags {
/// Configures the mdbx flags to use when opening the environment.
pub(crate) const fn make_flags(&self) -> ffi::MDBX_env_flags_t {
let mut flags = 0;
if self.no_sub_dir {
flags |= ffi::MDBX_NOSUBDIR;
}
if self.exclusive {
flags |= ffi::MDBX_EXCLUSIVE;
}
if self.accede {
flags |= ffi::MDBX_ACCEDE;
}
match self.mode {
Mode::ReadOnly => {
flags |= ffi::MDBX_RDONLY;
}
Mode::ReadWrite { sync_mode } => {
flags |= match sync_mode {
SyncMode::Durable => ffi::MDBX_SYNC_DURABLE,
SyncMode::NoMetaSync => ffi::MDBX_NOMETASYNC,
SyncMode::SafeNoSync => ffi::MDBX_SAFE_NOSYNC,
SyncMode::UtterlyNoSync => ffi::MDBX_UTTERLY_NOSYNC,
};
}
}
if self.no_rdahead {
flags |= ffi::MDBX_NORDAHEAD;
}
if self.no_meminit {
flags |= ffi::MDBX_NOMEMINIT;
}
if self.coalesce {
flags |= ffi::MDBX_COALESCE;
}
if self.liforeclaim {
flags |= ffi::MDBX_LIFORECLAIM;
}
flags |= ffi::MDBX_NOTLS;
flags
}
}
bitflags! {
#[doc="Database options."]
#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
#[repr(transparent)]
pub struct DatabaseFlags: MDBX_env_flags_t {
/// Use reverse string comparison for keys.
const REVERSE_KEY = MDBX_REVERSEKEY;
/// Use sorted duplicates, i.e. allow multi-values for a keys.
const DUP_SORT = MDBX_DUPSORT;
/// Numeric keys in native byte order either uint32_t or uint64_t (must
/// be one of uint32_t or uint64_t, other integer types, for example,
/// signed integer or uint16_t will not work). The keys must all be of
/// the same size and must be aligned while passing as arguments.
const INTEGER_KEY = MDBX_INTEGERKEY;
/// With MDBX_DUPSORT; sorted dup items have fixed size. The data
/// values must all be of the same size.
const DUP_FIXED = MDBX_DUPFIXED;
/// With MDBX_DUPSORT and with MDBX_DUPFIXED; dups are fixed size like
/// MDBX_INTEGERKEY -style integers. The data values must all be of the
/// same size and must be aligned while passing as arguments.
const INTEGER_DUP = MDBX_INTEGERDUP;
/// With MDBX_DUPSORT; use reverse string comparison for data values.
const REVERSE_DUP = MDBX_REVERSEDUP;
/// Create DB if not already existing.
const CREATE = MDBX_CREATE;
/// Open an existing table which may be open by another process
/// with unknown mode/flags. In such cases, instead of returning the
/// `MDBX_INCOMPATIBLE` error, the database will be opened with flags
/// which it was created, and then an application could determine the
/// actual flags.
const ACCEDE = MDBX_DB_ACCEDE;
}
}
bitflags! {
#[doc="Write options."]
#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
#[repr(transparent)]
pub struct WriteFlags: MDBX_env_flags_t {
/// Upsertion by default (without any other flags)
const UPSERT = MDBX_UPSERT;
/// For insertion: Don't write if the key already exists.
const NO_OVERWRITE = MDBX_NOOVERWRITE;
/// Has effect only for MDBX_DUPSORT tables. For upsertion: don't write
/// if the key-value pair already exist.
const NO_DUP_DATA = MDBX_NODUPDATA;
/// For upsertion: overwrite the current key/data pair. MDBX allows
/// this flag for mdbx_put() for explicit overwrite/update without
/// insertion. For deletion: remove only single entry at the current
/// cursor position.
const CURRENT = MDBX_CURRENT;
/// Has effect only for MDBX_DUPSORT tables. For deletion: remove all
/// multi-values (aka duplicates) for given key. For upsertion: replace
/// all multi-values for given key with a new one.
const ALLDUPS = MDBX_ALLDUPS;
/// For upsertion: Just reserve space for data, don't copy it. Return a
/// pointer to the reserved space.
const RESERVE = MDBX_RESERVE;
/// Data is being appended. Don't split full pages, continue on a new
/// instead.
const APPEND = MDBX_APPEND;
/// Has effect only for MDBX_DUPSORT tables. Duplicate data is being
/// appended. Don't split full pages, continue on a new instead.
const APPEND_DUP = MDBX_APPENDDUP;
/// Only for MDBX_DUPFIXED. Store multiple data items in one call.
const MULTIPLE = MDBX_MULTIPLE;
}
}