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
474
475
476
477
478
479
480
481
482
// Copyright Materialize, Inc. 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.

//! Symbiosis mode.
//!
//! In symbiosis mode, Materialize will conjoin with an OLTP database to
//! masquerade as a HTAP system. All DDL statements and writes will be routed to
//! the OLTP database (like `CREATE TABLE`, `INSERT`, etc.), while reads will be
//! routed through Materialize. Changes to the tables in the OLTP database are
//! automatically streamed through Materialize.
//!
//! The only supported OLTP database at the moment is PostgreSQL. Supporting
//! other databases is complicated by the fact that we roughly followe
//! Postgres's SQL semantics; using, say, MySQL, would be rather confusing,
//! because `INSERT`, `UPDATE`, and `DELETE` statements would be subject to a
//! wildly different set of SQL semantics than `SELECT` statements.
//!
//! Symbiosis mode is only suitable for development. It is likely to be
//! extremely slow and inefficient on large data sets.

use std::cell::RefCell;
use std::collections::HashSet;
use std::collections::{BTreeMap, HashMap};
use std::convert::TryInto;
use std::env;
use std::rc::Rc;

use anyhow::{anyhow, bail};
use chrono::Utc;
use tokio_postgres::types::FromSql;
use uuid::Uuid;

use pgrepr::Jsonb;
use repr::adt::decimal::Significand;
use repr::{Datum, RelationDesc, RelationType, Row};
use sql::ast::{
    ColumnOption, CreateTableStatement, DataType, DeleteStatement, DropObjectsStatement, Expr,
    InsertStatement, ObjectType, Raw, Statement, TableConstraint, UpdateStatement,
};
use sql::catalog::Catalog;
use sql::names::FullName;
use sql::normalize;
use sql::plan::{
    plan_default_expr, resolve_names_data_type, Aug, MutationKind, Plan, PlanContext,
    StatementContext, Table,
};

pub struct Postgres {
    client: tokio_postgres::Client,
    table_types: HashMap<FullName, (Vec<DataType<Aug>>, RelationDesc)>,
}

impl Postgres {
    pub async fn open_and_erase(url: &str) -> Result<Self, anyhow::Error> {
        let mut config: tokio_postgres::Config = url.parse()?;
        let username = whoami::username();
        if config.get_user().is_none() {
            config.user(env::var("PGUSER").ok().as_deref().unwrap_or(&username));
        }
        if config.get_password().is_none() {
            if let Ok(password) = env::var("PGPASSWORD") {
                config.password(password);
            }
        }
        if config.get_dbname().is_none() {
            if let Ok(dbname) = env::var("PGDATABASE") {
                config.dbname(&dbname);
            }
        }
        if config.get_hosts().is_empty() {
            config.host(env::var("PGHOST").ok().as_deref().unwrap_or("localhost"));
        }
        let (client, conn) = config
            .connect(tokio_postgres::NoTls)
            .await
            .map_err(|err| anyhow!("Postgres connection failed: {}", err))?;

        tokio::spawn(async move {
            if let Err(e) = conn.await {
                panic!("connection error: {}", e);
            }
        });

        // drop all tables
        client
            .execute(
                r#"
DO $$ DECLARE
    r RECORD;
BEGIN
    FOR r IN (SELECT tablename FROM pg_tables WHERE schemaname = current_schema()) LOOP
        EXECUTE 'DROP TABLE IF EXISTS ' || quote_ident(r.tablename) || ' CASCADE';
    END LOOP;
END $$;
"#,
                &[],
            )
            .await?;
        Ok(Self {
            client,
            table_types: HashMap::new(),
        })
    }

    pub fn can_handle(&self, stmt: &Statement<Raw>) -> bool {
        matches!(
            stmt,
            Statement::CreateTable { .. }
                | Statement::DropObjects { .. }
                | Statement::Delete { .. }
                | Statement::Insert { .. }
                | Statement::Update { .. }
        )
    }

    pub async fn execute(
        &mut self,
        pcx: &PlanContext,
        catalog: &dyn Catalog,
        stmt: &Statement<Raw>,
    ) -> Result<Plan, anyhow::Error> {
        let scx = StatementContext {
            pcx,
            catalog,
            ids: HashSet::new(),
            param_types: Rc::new(RefCell::new(BTreeMap::new())),
        };
        Ok(match stmt {
            Statement::CreateTable(CreateTableStatement {
                name,
                columns,
                constraints,
                if_not_exists,
                temporary,
                ..
            }) => {
                let names: Vec<_> = columns
                    .iter()
                    .map(|c| Some(sql::normalize::column_name(c.name.clone())))
                    .collect();

                // Build initial relation type that handles declared data types
                // and NOT NULL, UNIQUE, and PRIMARY KEY constraints.
                let mut column_types = Vec::with_capacity(columns.len());
                let mut defaults = Vec::with_capacity(columns.len());
                let mut depends_on = Vec::new();
                let mut keys = vec![];

                let mut sql_types: Vec<_> = Vec::with_capacity(columns.len());
                for (index, column) in columns.iter().enumerate() {
                    let (aug_data_type, ids) =
                        resolve_names_data_type(&scx, column.data_type.clone())?;
                    let ty = sql::plan::scalar_type_from_sql(&scx, &aug_data_type)?;
                    sql_types.push(aug_data_type);
                    let mut nullable = true;
                    let mut default = Expr::null();

                    for option in &column.options {
                        match &option.option {
                            ColumnOption::NotNull => nullable = false,
                            ColumnOption::Default(expr) => {
                                let (_, expr_depends_on) = plan_default_expr(&scx, expr, &ty)?;
                                depends_on.extend(expr_depends_on);
                                default = expr.clone();
                            }
                            // PRIMARY KEY implies UNIQUE and NOT NULL.
                            ColumnOption::Unique { is_primary } => {
                                keys.push(vec![index]);
                                if *is_primary {
                                    nullable = false;
                                }
                            }
                            other => bail!("unsupported column constraint: {}", other),
                        }
                    }
                    column_types.push(ty.nullable(nullable));
                    defaults.push(default);
                    depends_on.extend(ids);
                }
                let mut typ = RelationType { column_types, keys };

                // Handle table-level UNIQUE and PRIMARY KEY constraints.
                // PRIMARY KEY implies UNIQUE and NOT NULL.
                for constraint in constraints {
                    if let TableConstraint::Unique {
                        name: _,
                        columns,
                        is_primary,
                    } = constraint
                    {
                        let mut key = vec![];
                        for column in columns {
                            let name = normalize::column_name(column.clone());
                            match names.iter().position(|n| n.as_ref() == Some(&name)) {
                                None => bail!("unknown column {} in unique constraint", name),
                                Some(i) => key.push(i),
                            }
                        }
                        if *is_primary {
                            for i in key.iter() {
                                typ.column_types[*i].nullable = false;
                            }
                        }
                        typ = typ.with_key(key);
                    }
                }

                self.client.execute(&*stmt.to_string(), &[]).await?;
                let name = scx.allocate_name(normalize::unresolved_object_name(name.clone())?);
                let desc = RelationDesc::new(typ, names);
                self.table_types
                    .insert(name.clone(), (sql_types, desc.clone()));

                let temporary = *temporary;
                let table = Table {
                    create_sql: stmt.to_string(),
                    desc,
                    defaults,
                    temporary,
                };
                Plan::CreateTable {
                    name,
                    table,
                    if_not_exists: *if_not_exists,
                    depends_on,
                }
            }
            Statement::DropObjects(DropObjectsStatement {
                names,
                object_type: ObjectType::Table,
                if_exists,
                ..
            }) => {
                self.client.execute(&*stmt.to_string(), &[]).await?;
                let mut items = vec![];
                for name in names {
                    match scx.resolve_item(name.clone()) {
                        Ok(item) => items.push(item.id()),
                        Err(err) => {
                            if *if_exists {
                                continue;
                            } else {
                                return Err(err.into());
                            }
                        }
                    }
                }
                Plan::DropItems {
                    items,
                    ty: ObjectType::Table,
                }
            }
            Statement::Delete(DeleteStatement { table_name, .. }) => {
                let mut updates = vec![];
                let table = scx.resolve_item(table_name.clone())?;
                let sql = format!("{} RETURNING *", stmt.to_string());
                for row in self.run_query(table.name(), sql, 0).await? {
                    updates.push((row, -1));
                }
                let affected_rows = updates.len();
                Plan::SendDiffs {
                    id: table.id(),
                    updates,
                    affected_rows,
                    kind: MutationKind::Delete,
                }
            }
            Statement::Insert(InsertStatement { table_name, .. }) => {
                let mut updates = vec![];
                let table = scx.resolve_item(table_name.clone())?;
                // RETURNING cannot return zero columns, but we might be
                // executing INSERT INTO t DEFAULT VALUES where t is a zero
                // arity table. So use a time-honored trick of always including
                // a junk column in RETURNING, then stripping that column out.
                let sql = format!("{} RETURNING *, 1", stmt.to_string());
                for row in self.run_query(table.name(), sql, 1).await? {
                    updates.push((row, 1));
                }
                let affected_rows = updates.len();
                Plan::SendDiffs {
                    id: table.id(),
                    updates,
                    affected_rows,
                    kind: MutationKind::Insert,
                }
            }
            Statement::Update(UpdateStatement {
                table_name,
                selection,
                ..
            }) => {
                let mut updates = vec![];
                let mut sql = format!("SELECT * FROM {}", table_name);
                let table = scx.resolve_item(table_name.clone())?;
                if let Some(selection) = selection {
                    sql += &format!(" WHERE {}", selection);
                }
                for row in self.run_query(table.name(), sql, 0).await? {
                    updates.push((row, -1))
                }
                let affected_rows = updates.len();
                let sql = format!("{} RETURNING *", stmt.to_string());
                for row in self.run_query(table.name(), sql, 0).await? {
                    updates.push((row, 1));
                }
                assert_eq!(affected_rows * 2, updates.len());
                Plan::SendDiffs {
                    id: table.id(),
                    updates,
                    affected_rows,
                    kind: MutationKind::Update,
                }
            }
            _ => bail!("Unsupported symbiosis statement: {:?}", stmt),
        })
    }

    async fn run_query(
        &mut self,
        table_name: &FullName,
        query: String,
        junk: usize,
    ) -> Result<Vec<Row>, anyhow::Error> {
        let (sql_types, desc) = self
            .table_types
            .get(table_name)
            .ok_or_else(|| anyhow!("Unknown table: {:?}", table_name))?
            .clone();
        let mut rows = vec![];
        let postgres_rows = self.client.query(&*query, &[]).await?;
        let mut row = Row::default();
        for postgres_row in postgres_rows.iter() {
            for c in 0..postgres_row.len() - junk {
                row = push_column(
                    row,
                    &postgres_row,
                    c,
                    &sql_types[c],
                    desc.typ().column_types[c].nullable,
                )?;
            }
            rows.push(row.finish_and_reuse());
        }
        Ok(rows)
    }
}

fn push_column(
    mut row: Row,
    postgres_row: &tokio_postgres::Row,
    i: usize,
    sql_type: &DataType<Aug>,
    nullable: bool,
) -> Result<Row, anyhow::Error> {
    // NOTE this needs to stay in sync with materialize::sql::scalar_type_from_sql
    // in some cases, we use slightly different representations than postgres does for the same sql types, so we have to be careful about conversions
    match sql_type {
        DataType::Other { name, typ_mod } => {
            match name.raw_name().to_string().as_str() {
                "pg_catalog.bool" => {
                    let bool = get_column_inner::<bool>(postgres_row, i, nullable)?;
                    row.push(Datum::from(bool));
                }
                "pg_catalog.bytea" => {
                    let bytes = get_column_inner::<Vec<u8>>(postgres_row, i, nullable)?;
                    row.push(Datum::from(bytes.as_deref()));
                }
                "pg_catalog.char" | "pg_catalog.text" | "pg_catalog.varchar" => {
                    let string = get_column_inner::<String>(postgres_row, i, nullable)?;
                    row.push(Datum::from(string.as_deref()));
                }
                "pg_catalog.date" => {
                    let d: chrono::NaiveDate =
                        get_column_inner::<chrono::NaiveDate>(postgres_row, i, nullable)?.unwrap();
                    row.push(Datum::Date(d));
                }
                "pg_catalog.float4" => {
                    let f = get_column_inner::<f32>(postgres_row, i, nullable)?.map(f32::from);
                    row.push(Datum::from(f));
                }
                "pg_catalog.float8" => {
                    let f = get_column_inner::<f64>(postgres_row, i, nullable)?;
                    row.push(Datum::from(f));
                }
                "pg_catalog.int4" => {
                    let i = get_column_inner::<i32>(postgres_row, i, nullable)?;
                    row.push(Datum::from(i));
                }
                "pg_catalog.int8" => {
                    let i = get_column_inner::<i64>(postgres_row, i, nullable)?;
                    row.push(Datum::from(i));
                }
                "pg_catalog.interval" => {
                    let iv =
                        get_column_inner::<pgrepr::Interval>(postgres_row, i, nullable)?.unwrap();
                    row.push(Datum::Interval(iv.0));
                }
                "pg_catalog.jsonb" => {
                    let jsonb = get_column_inner::<Jsonb>(postgres_row, i, nullable)?;
                    if let Some(jsonb) = jsonb {
                        row.extend_by_row(&jsonb.0.into_row())
                    } else {
                        row.push(Datum::Null)
                    }
                }
                "pg_catalog.numeric" => {
                    let (_, desired_scale) = sql::plan::unwrap_numeric_typ_mod(typ_mod)?;
                    match get_column_inner::<pgrepr::Numeric>(postgres_row, i, nullable)? {
                        None => row.push(Datum::Null),
                        Some(d) => {
                            let mut significand = d.0.significand();
                            // TODO(jamii) lots of potential for unchecked edge cases here eg 10^scale_correction could overflow
                            // current representation is `significand * 10^current_scale`
                            // want to get to `significand2 * 10^desired_scale`
                            // so `significand2 = significand * 10^(current_scale - desired_scale)`
                            let scale_correction =
                                (d.0.scale() as isize) - (desired_scale as isize);
                            if scale_correction > 0 {
                                significand /= 10i128.pow(scale_correction.try_into()?);
                            } else {
                                significand *= 10i128.pow((-scale_correction).try_into()?);
                            };
                            row.push(Datum::from(Significand::new(significand)));
                        }
                    }
                }
                "pg_catalog.int2" | "pg_catalog.smallint" => {
                    let i = get_column_inner::<i16>(postgres_row, i, nullable)?.map(i32::from);
                    row.push(Datum::from(i));
                }
                "pg_catalog.timestamp" => {
                    let d: chrono::NaiveDateTime =
                        get_column_inner::<chrono::NaiveDateTime>(postgres_row, i, nullable)?
                            .unwrap();
                    row.push(Datum::Timestamp(d));
                }
                "pg_catalog.timestamptz" => {
                    let d: chrono::DateTime<Utc> =
                        get_column_inner::<chrono::DateTime<Utc>>(postgres_row, i, nullable)?
                            .unwrap();
                    row.push(Datum::TimestampTz(d));
                }
                "pg_catalog.uuid" => {
                    let u = get_column_inner::<Uuid>(postgres_row, i, nullable)?.unwrap();
                    row.push(Datum::Uuid(u));
                }
                _ => bail!(
                    "Postgres to materialize conversion not yet supported for {:?}",
                    sql_type
                ),
            }
        }
        _ => bail!(
            "Postgres to materialize conversion not yet supported for {:?}",
            sql_type
        ),
    }
    Ok(row)
}

fn get_column_inner<'a, T>(
    postgres_row: &'a tokio_postgres::Row,
    i: usize,
    nullable: bool,
) -> Result<Option<T>, anyhow::Error>
where
    T: FromSql<'a>,
{
    if nullable {
        let value: Option<T> = postgres_row.try_get(i)?;
        Ok(value)
    } else {
        let value: T = postgres_row.try_get(i)?;
        Ok(Some(value))
    }
}