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
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.

//! Implementation of [Consensus] backed by Postgres.

use std::fmt::Formatter;
use std::sync::Arc;
use std::time::Duration;

use anyhow::anyhow;
use async_stream::try_stream;
use async_trait::async_trait;
use bytes::Bytes;
use deadpool_postgres::tokio_postgres::types::{to_sql_checked, FromSql, IsNull, ToSql, Type};
use deadpool_postgres::{Object, PoolError};
use futures_util::StreamExt;
use mz_ore::cast::CastFrom;
use mz_ore::metrics::MetricsRegistry;
use mz_postgres_client::metrics::PostgresClientMetrics;
use mz_postgres_client::{PostgresClient, PostgresClientConfig, PostgresClientKnobs};
use tokio_postgres::error::SqlState;
use tracing::warn;

use crate::error::Error;
use crate::location::{CaSResult, Consensus, ExternalError, ResultStream, SeqNo, VersionedData};

// These `sql_stats_automatic_collection_enabled` are for the cost-based
// optimizer but all the queries against this table are single-table and very
// carefully tuned to hit the primary index, so the cost-based optimizer doesn't
// really get us anything. OTOH, the background jobs that crdb creates to
// collect these stats fill up the jobs table (slowing down all sorts of
// things).
const SCHEMA: &str = "
CREATE TABLE IF NOT EXISTS consensus (
    shard text NOT NULL,
    sequence_number bigint NOT NULL,
    data bytea NOT NULL,
    PRIMARY KEY(shard, sequence_number)
) WITH (sql_stats_automatic_collection_enabled = false);
";

impl ToSql for SeqNo {
    fn to_sql(
        &self,
        ty: &Type,
        w: &mut bytes::BytesMut,
    ) -> Result<IsNull, Box<dyn std::error::Error + Sync + Send>> {
        // We can only represent sequence numbers in the range [0, i64::MAX].
        let value = i64::try_from(self.0)?;
        <i64 as ToSql>::to_sql(&value, ty, w)
    }

    fn accepts(ty: &Type) -> bool {
        <i64 as ToSql>::accepts(ty)
    }

    to_sql_checked!();
}

impl<'a> FromSql<'a> for SeqNo {
    fn from_sql(
        ty: &Type,
        raw: &'a [u8],
    ) -> Result<SeqNo, Box<dyn std::error::Error + Sync + Send>> {
        let sequence_number = <i64 as FromSql>::from_sql(ty, raw)?;

        // Sanity check that the sequence number we received falls in the
        // [0, i64::MAX] range.
        let sequence_number = u64::try_from(sequence_number)?;
        Ok(SeqNo(sequence_number))
    }

    fn accepts(ty: &Type) -> bool {
        <i64 as FromSql>::accepts(ty)
    }
}

/// Configuration to connect to a Postgres backed implementation of [Consensus].
#[derive(Clone, Debug)]
pub struct PostgresConsensusConfig {
    url: String,
    knobs: Arc<dyn PostgresClientKnobs>,
    metrics: PostgresClientMetrics,
}

impl From<PostgresConsensusConfig> for PostgresClientConfig {
    fn from(config: PostgresConsensusConfig) -> Self {
        PostgresClientConfig::new(config.url, config.knobs, config.metrics)
    }
}

impl PostgresConsensusConfig {
    const EXTERNAL_TESTS_POSTGRES_URL: &'static str =
        "MZ_PERSIST_EXTERNAL_STORAGE_TEST_POSTGRES_URL";

    /// Returns a new [PostgresConsensusConfig] for use in production.
    pub fn new(
        url: &str,
        knobs: Box<dyn PostgresClientKnobs>,
        metrics: PostgresClientMetrics,
    ) -> Result<Self, Error> {
        Ok(PostgresConsensusConfig {
            url: url.to_string(),
            knobs: Arc::from(knobs),
            metrics,
        })
    }

    /// Returns a new [PostgresConsensusConfig] for use in unit tests.
    ///
    /// By default, persist tests that use external storage (like Postgres) are
    /// no-ops so that `cargo test` works on new environments without any
    /// configuration. To activate the tests for [PostgresConsensus] set the
    /// `MZ_PERSIST_EXTERNAL_STORAGE_TEST_POSTGRES_URL` environment variable
    /// with a valid connection url [1].
    ///
    /// [1]: https://docs.rs/tokio-postgres/latest/tokio_postgres/config/struct.Config.html#url
    pub fn new_for_test() -> Result<Option<Self>, Error> {
        let url = match std::env::var(Self::EXTERNAL_TESTS_POSTGRES_URL) {
            Ok(url) => url,
            Err(_) => {
                if mz_ore::env::is_var_truthy("CI") {
                    panic!("CI is supposed to run this test but something has gone wrong!");
                }
                return Ok(None);
            }
        };

        struct TestConsensusKnobs;
        impl std::fmt::Debug for TestConsensusKnobs {
            fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
                f.debug_struct("TestConsensusKnobs").finish_non_exhaustive()
            }
        }
        impl PostgresClientKnobs for TestConsensusKnobs {
            fn connection_pool_max_size(&self) -> usize {
                2
            }

            fn connection_pool_max_wait(&self) -> Option<Duration> {
                Some(Duration::from_secs(1))
            }

            fn connection_pool_ttl(&self) -> Duration {
                Duration::MAX
            }
            fn connection_pool_ttl_stagger(&self) -> Duration {
                Duration::MAX
            }
            fn connect_timeout(&self) -> Duration {
                Duration::MAX
            }
            fn tcp_user_timeout(&self) -> Duration {
                Duration::ZERO
            }
        }

        let config = PostgresConsensusConfig::new(
            &url,
            Box::new(TestConsensusKnobs),
            PostgresClientMetrics::new(&MetricsRegistry::new(), "mz_persist"),
        )?;
        Ok(Some(config))
    }
}

/// Implementation of [Consensus] over a Postgres database.
pub struct PostgresConsensus {
    postgres_client: PostgresClient,
}

impl std::fmt::Debug for PostgresConsensus {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("PostgresConsensus").finish_non_exhaustive()
    }
}

impl PostgresConsensus {
    /// Open a Postgres [Consensus] instance with `config`, for the collection
    /// named `shard`.
    pub async fn open(config: PostgresConsensusConfig) -> Result<Self, ExternalError> {
        let postgres_client = PostgresClient::open(config.into())?;

        let client = postgres_client.get_connection().await?;

        // The `consensus` table creates and deletes rows at a high frequency, generating many
        // tombstoned rows. If Cockroach's GC interval is set high (the default is 25h) and
        // these tombstones accumulate, scanning over the table will take increasingly and
        // prohibitively long.
        //
        // See: https://github.com/MaterializeInc/materialize/issues/13975
        // See: https://www.cockroachlabs.com/docs/stable/configure-zone.html#variables
        match client
            .batch_execute(&format!(
                "{} {}",
                SCHEMA, "ALTER TABLE consensus CONFIGURE ZONE USING gc.ttlseconds = 600;",
            ))
            .await
        {
            Ok(()) => {}
            Err(e) if e.code() == Some(&SqlState::INSUFFICIENT_PRIVILEGE) => {
                warn!("unable to ALTER TABLE consensus, this is expected and OK when connecting with a read-only user");
            }
            Err(e) => return Err(e.into()),
        }

        Ok(PostgresConsensus { postgres_client })
    }

    /// Drops and recreates the `consensus` table in Postgres
    ///
    /// ONLY FOR TESTING
    pub async fn drop_and_recreate(&self) -> Result<(), ExternalError> {
        // this could be a TRUNCATE if we're confident the db won't reuse any state
        let client = self.get_connection().await?;
        client.execute("DROP TABLE consensus", &[]).await?;
        client.execute(SCHEMA, &[]).await?;
        Ok(())
    }

    async fn get_connection(&self) -> Result<Object, PoolError> {
        self.postgres_client.get_connection().await
    }
}

#[async_trait]
impl Consensus for PostgresConsensus {
    fn list_keys(&self) -> ResultStream<String> {
        let q = "SELECT DISTINCT shard FROM consensus";

        Box::pin(try_stream! {
            // NB: it's important that we hang on to this client for the lifetime of the stream,
            // to avoid returning it to the pool prematurely.
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            let params: &[String] = &[];
            let mut rows = Box::pin(client.query_raw(&statement, params).await?);
            while let Some(row) = rows.next().await {
                let shard: String = row?.try_get("shard")?;
                yield shard;
            }
        })
    }

    async fn head(&self, key: &str) -> Result<Option<VersionedData>, ExternalError> {
        let q = "SELECT sequence_number, data FROM consensus
             WHERE shard = $1 ORDER BY sequence_number DESC LIMIT 1";
        let row = {
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            client.query_opt(&statement, &[&key]).await?
        };
        let row = match row {
            None => return Ok(None),
            Some(row) => row,
        };

        let seqno: SeqNo = row.try_get("sequence_number")?;

        let data: Vec<u8> = row.try_get("data")?;
        Ok(Some(VersionedData {
            seqno,
            data: Bytes::from(data),
        }))
    }

    async fn compare_and_set(
        &self,
        key: &str,
        expected: Option<SeqNo>,
        new: VersionedData,
    ) -> Result<CaSResult, ExternalError> {
        if let Some(expected) = expected {
            if new.seqno <= expected {
                return Err(Error::from(
                        format!("new seqno must be strictly greater than expected. Got new: {:?} expected: {:?}",
                                 new.seqno, expected)).into());
            }
        }

        let result = if let Some(expected) = expected {
            // This query has been written to execute within a single
            // network round-trip. The insert performance has been tuned
            // against CockroachDB, ensuring it goes through the fast-path
            // 1-phase commit of CRDB. Any changes to this query should
            // confirm an EXPLAIN ANALYZE (VERBOSE) query plan contains
            // `auto commit`
            let q = r#"
                INSERT INTO consensus (shard, sequence_number, data)
                SELECT $1, $2, $3
                WHERE (SELECT sequence_number FROM consensus
                       WHERE shard = $1
                       ORDER BY sequence_number DESC LIMIT 1) = $4;
            "#;
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            client
                .execute(
                    &statement,
                    &[&key, &new.seqno, &new.data.as_ref(), &expected],
                )
                .await?
        } else {
            // Insert the new row as long as no other row exists for the same shard.
            let q = "INSERT INTO consensus SELECT $1, $2, $3 WHERE
                     NOT EXISTS (
                         SELECT * FROM consensus WHERE shard = $1
                     )
                     ON CONFLICT DO NOTHING";
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            client
                .execute(&statement, &[&key, &new.seqno, &new.data.as_ref()])
                .await?
        };

        if result == 1 {
            Ok(CaSResult::Committed)
        } else {
            Ok(CaSResult::ExpectationMismatch)
        }
    }

    async fn scan(
        &self,
        key: &str,
        from: SeqNo,
        limit: usize,
    ) -> Result<Vec<VersionedData>, ExternalError> {
        let q = "SELECT sequence_number, data FROM consensus
             WHERE shard = $1 AND sequence_number >= $2
             ORDER BY sequence_number ASC LIMIT $3";
        let Ok(limit) = i64::try_from(limit) else {
            return Err(ExternalError::from(anyhow!(
                "limit must be [0, i64::MAX]. was: {:?}",
                limit
            )));
        };
        let rows = {
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            client.query(&statement, &[&key, &from, &limit]).await?
        };
        let mut results = Vec::with_capacity(rows.len());

        for row in rows {
            let seqno: SeqNo = row.try_get("sequence_number")?;
            let data: Vec<u8> = row.try_get("data")?;
            results.push(VersionedData {
                seqno,
                data: Bytes::from(data),
            });
        }
        Ok(results)
    }

    async fn truncate(&self, key: &str, seqno: SeqNo) -> Result<usize, ExternalError> {
        let q = "DELETE FROM consensus
                WHERE shard = $1 AND sequence_number < $2 AND
                EXISTS(
                    SELECT * FROM consensus WHERE shard = $1 AND sequence_number >= $2
                )";

        let result = {
            let client = self.get_connection().await?;
            let statement = client.prepare_cached(q).await?;
            client.execute(&statement, &[&key, &seqno]).await?
        };
        if result == 0 {
            // We weren't able to successfully truncate any rows inspect head to
            // determine whether the request was valid and there were no records in
            // the provided range, or the request was invalid because it would have
            // also deleted head.

            // It's safe to call head in a subsequent transaction rather than doing
            // so directly in the same transaction because, once a given (seqno, data)
            // pair exists for our shard, we enforce the invariants that
            // 1. Our shard will always have _some_ data mapped to it.
            // 2. All operations that modify the (seqno, data) can only increase
            //    the sequence number.
            let current = self.head(key).await?;
            if current.map_or(true, |data| data.seqno < seqno) {
                return Err(ExternalError::from(anyhow!(
                    "upper bound too high for truncate: {:?}",
                    seqno
                )));
            }
        }

        Ok(usize::cast_from(result))
    }
}

#[cfg(test)]
mod tests {
    use tracing::info;
    use uuid::Uuid;

    use crate::location::tests::consensus_impl_test;

    use super::*;

    #[mz_ore::test(tokio::test(flavor = "multi_thread"))]
    #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
    async fn postgres_consensus() -> Result<(), ExternalError> {
        let config = match PostgresConsensusConfig::new_for_test()? {
            Some(config) => config,
            None => {
                info!(
                    "{} env not set: skipping test that uses external service",
                    PostgresConsensusConfig::EXTERNAL_TESTS_POSTGRES_URL
                );
                return Ok(());
            }
        };

        consensus_impl_test(|| PostgresConsensus::open(config.clone())).await?;

        // and now verify the implementation-specific `drop_and_recreate` works as intended
        let consensus = PostgresConsensus::open(config.clone()).await?;
        let key = Uuid::new_v4().to_string();
        let state = VersionedData {
            seqno: SeqNo(5),
            data: Bytes::from("abc"),
        };

        assert_eq!(
            consensus.compare_and_set(&key, None, state.clone()).await,
            Ok(CaSResult::Committed),
        );

        assert_eq!(consensus.head(&key).await, Ok(Some(state.clone())));

        consensus.drop_and_recreate().await?;

        assert_eq!(consensus.head(&key).await, Ok(None));

        Ok(())
    }

    #[mz_ore::test(tokio::test(flavor = "multi_thread"))]
    #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
    async fn postgres_consensus_blocking() -> Result<(), ExternalError> {
        let config = match PostgresConsensusConfig::new_for_test()? {
            Some(config) => config,
            None => {
                info!(
                    "{} env not set: skipping test that uses external service",
                    PostgresConsensusConfig::EXTERNAL_TESTS_POSTGRES_URL
                );
                return Ok(());
            }
        };

        let consensus: PostgresConsensus = PostgresConsensus::open(config.clone()).await?;
        // Max size in test is 2... let's saturate the pool.
        let _conn1 = consensus.get_connection().await?;
        let _conn2 = consensus.get_connection().await?;

        // And finally, we should see the next connect time out.
        let conn3 = consensus.get_connection().await;

        assert!(conn3.is_err());

        Ok(())
    }
}