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
|
use std::collections::HashMap;
use rand::Rng;
use crate::{
atuin_common::record::{EncryptedData, HostId, Record, RecordIdx, RecordStatus},
atuin_server::database::{DbError, DbResult, DbSettings, models::User},
};
use sqlx::postgres::PgPoolOptions;
use tracing::instrument;
use uuid::Uuid;
use wrappers::DbRecord;
mod wrappers;
const MIN_PG_VERSION: u32 = 14;
#[derive(Clone)]
pub(crate) struct ServerPostgres {
pool: sqlx::Pool<sqlx::postgres::Postgres>,
/// Optional read replica pool for read-only queries
read_pool: Option<sqlx::Pool<sqlx::postgres::Postgres>>,
}
impl ServerPostgres {
/// Returns the appropriate pool for read operations.
/// Uses `read_pool` if available, otherwise falls back to the primary pool.
fn read_pool(&self) -> &sqlx::Pool<sqlx::postgres::Postgres> {
self.read_pool.as_ref().unwrap_or(&self.pool)
}
}
impl ServerPostgres {
pub(crate) async fn new(settings: &DbSettings) -> DbResult<Self> {
let pool = PgPoolOptions::new()
.max_connections(100)
.connect(settings.db_uri.as_str())
.await?;
// Call server_version_num to get the DB server's major version number
// The call returns None for servers older than 8.x.
let pg_major_version: u32 =
pool.acquire()
.await?
.server_version_num()
.ok_or(DbError::Other(eyre::Report::msg(
"could not get PostgreSQL version",
)))?
/ 10000;
if pg_major_version < MIN_PG_VERSION {
return Err(DbError::Other(eyre::Report::msg(format!(
"unsupported PostgreSQL version {pg_major_version}, minimum required is {MIN_PG_VERSION}"
))));
}
sqlx::migrate!("./db/server-pg-migrations")
.run(&pool)
.await
.map_err(|error| DbError::Other(error.into()))?;
// Create read replica pool if configured
let read_pool = if let Some(read_db_uri) = &settings.read_db_uri {
tracing::info!("Connecting to read replica database");
let read_pool = PgPoolOptions::new()
.max_connections(100)
.connect(read_db_uri.as_str())
.await?;
// Verify the read replica is also a supported PostgreSQL version
let read_pg_major_version: u32 = read_pool
.acquire()
.await?
.server_version_num()
.ok_or(DbError::Other(eyre::Report::msg(
"could not get PostgreSQL version from read replica",
)))?
/ 10000;
if read_pg_major_version < MIN_PG_VERSION {
return Err(DbError::Other(eyre::Report::msg(format!(
"unsupported PostgreSQL version {read_pg_major_version} on read replica, minimum required is {MIN_PG_VERSION}"
))));
}
Some(read_pool)
} else {
None
};
Ok(Self { pool, read_pool })
}
#[instrument(skip_all)]
pub(crate) async fn add_records(
&self,
user: &User,
records: &[Record<EncryptedData>],
) -> DbResult<()> {
let mut tx = self.pool.begin().await?;
// We won't have uploaded this data if it wasn't the max. Therefore, we can deduce the max
// idx without having to make further database queries. Doing the query on this small
// amount of data should be much, much faster.
//
// Worst case, say we get this wrong. We end up caching data that isn't actually the max
// idx, so clients upload again. The cache logic can be verified with a sql query anyway :)
let mut heads = HashMap::<(HostId, &str), u64>::new();
for i in records {
let id = crate::atuin_common::utils::uuid_v7();
let result = sqlx::query(
"
INSERT INTO store (id, client_id, host, idx, timestamp, version, tag, data, cek, user_id)
VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10)
ON conflict DO nothing
",
)
.bind(id)
.bind(i.id)
.bind(i.host.id)
.bind(i.idx as i64)
.bind(i.timestamp as i64) // throwing away some data, but i64 is still big in terms of time
.bind(&i.version)
.bind(&i.tag)
.bind(&i.data.data)
.bind(&i.data.content_encryption_key)
.bind(user.id)
.execute(&mut *tx)
.await?;
// Only update heads if we actually inserted the record
if result.rows_affected() > 0 {
heads
.entry((i.host.id, &i.tag))
.and_modify(|e| {
if i.idx > *e {
*e = i.idx;
}
})
.or_insert(i.idx);
}
}
// we've built the map of heads for this push, so commit it to the database
for ((host, tag), idx) in heads {
sqlx::query(
"
INSERT INTO store_idx_cache (user_id, host, tag, idx)
VALUES ($1, $2, $3, $4)
ON conflict(user_id, host, tag) DO update
SET idx = greatest(store_idx_cache.idx, $4)
",
)
.bind(user.id)
.bind(host)
.bind(tag)
.bind(idx as i64)
.execute(&mut *tx)
.await?;
}
tx.commit().await?;
Ok(())
}
#[instrument(skip_all)]
pub(crate) async fn next_records(
&self,
user: &User,
host: HostId,
tag: String,
start: Option<RecordIdx>,
count: u64,
) -> DbResult<Vec<Record<EncryptedData>>> {
tracing::debug!("{:?} - {:?} - {:?}", host, tag, start);
let start = start.unwrap_or(0);
let records: Result<Vec<DbRecord>, DbError> = sqlx::query_as(
"
SELECT client_id, host, idx, timestamp, version, tag, data, cek FROM store
WHERE user_id = $1
AND tag = $2
AND host = $3
AND idx >= $4
ORDER BY idx asc
LIMIT $5
",
)
.bind(user.id)
.bind(tag.clone())
.bind(host)
.bind(start as i64)
.bind(count as i64)
.fetch_all(self.read_pool())
.await
.map_err(Into::into);
let ret = match records {
Ok(records) => {
let records: Vec<Record<EncryptedData>> = records
.into_iter()
.map(|f| {
let record: Record<EncryptedData> = f.into();
record
})
.collect();
records
}
Err(DbError::NotFound) => {
tracing::debug!("no records found in store: {:?}/{}", host, tag);
return Ok(vec![]);
}
Err(e) => return Err(e),
};
Ok(ret)
}
pub(crate) async fn status(&self, user: &User) -> DbResult<RecordStatus> {
// If IDX_CACHE_ROLLOUT is set, then we
// 1. Read the value of the var, use it as a % chance of using the cache
// 2. If we use the cache, just read from the cache table
// 3. If we don't use the cache, read from the store table
// IDX_CACHE_ROLLOUT should be between 0 and 100.
let idx_cache_rollout =
std::env::var("IDX_CACHE_ROLLOUT").unwrap_or_else(|_| "0".to_string());
let idx_cache_rollout = idx_cache_rollout.parse::<f64>().unwrap_or(0.0);
let use_idx_cache = rand::thread_rng().gen_bool(idx_cache_rollout / 100.0);
let mut res: Vec<(Uuid, String, i64)> = if use_idx_cache {
tracing::debug!("using idx cache for user {}", user.id);
sqlx::query_as(
"
SELECT host, tag, idx
FROM store_idx_cache
WHERE user_id = $1
",
)
.bind(user.id)
.fetch_all(self.read_pool())
.await?
} else {
tracing::debug!("using aggregate query for user {}", user.id);
sqlx::query_as(
"
SELECT host, tag, max(idx)
FROM store
WHERE user_id = $1
GROUP BY host, tag
",
)
.bind(user.id)
.fetch_all(self.read_pool())
.await?
};
res.sort();
let mut status = RecordStatus::new();
for i in &res {
status.set_raw(HostId(i.0), i.1.clone(), i.2 as u64);
}
Ok(status)
}
}
|