Compare · 022

Batch and copyfrom: DuckDB and PostgreSQL

Every file of the DuckDB and PostgreSQL versions side by side, with the lines that differ marked. 2 parameter or column types differ in Go.

Compare DuckDB withSQLite PostgreSQL CockroachDB TimescaleDB MySQL MariaDB ClickHouse

Configuration

sgl.yamldiffers

DuckDB
PostgreSQL
1version: 1
1version: 1
2
2
3databases:
3databases:
4  - name: tides
4  - name: tides
5    dialect: duckdb
5    dialect: postgres
6    migrations: db/migrations
6    migrations: db/migrations
7    queries: db/queries
7    queries: db/queries
8    go:
8    go:
9      package: store
9      package: store
10      out: internal/store
10      out: internal/store
11      driver: database/sql
12    migrate:
13      dsn: ${env:DATABASE_URL}

Migrations

db/migrations/0001_readings.up.sqldiffers

DuckDB
PostgreSQL
1CREATE SEQUENCE reading_ids;
2
3CREATE TABLE readings (
1CREATE TABLE readings (
4    id        BIGINT  PRIMARY KEY DEFAULT nextval('reading_ids'),
2    station   TEXT   NOT NULL,
5    station   VARCHAR NOT NULL,
3    minute    BIGINT NOT NULL,
6    minute    INTEGER NOT NULL,
4    height_cm BIGINT NOT NULL,
7    height_cm INTEGER NOT NULL,
5    note      TEXT,
8    note      VARCHAR
6    PRIMARY KEY (station, minute)
9);
7);

db/migrations/0001_readings.down.sqldiffers

DuckDB
PostgreSQL
1DROP TABLE readings;
1DROP TABLE readings;
2DROP SEQUENCE reading_ids;

Queries

RecordReadingsidentical

DuckDB · db/queries/readings.sql
PostgreSQL · db/queries/readings.sql
1-- sgl.query(RecordReadings, batch)
1-- sgl.query(RecordReadings, batch)
2INSERT INTO readings (station, minute, height_cm, note) VALUES ($1, $2, $3, $4);
2INSERT INTO readings (station, minute, height_cm, note) VALUES ($1, $2, $3, $4);

SummariseReadingsidentical

DuckDB · db/queries/readings.sql
PostgreSQL · db/queries/readings.sql
1-- sgl.query(SummariseReadings, many)
1-- sgl.query(SummariseReadings, many)
2SELECT station, COUNT(*) AS readings, COUNT(note) AS noted
2SELECT station, COUNT(*) AS readings, COUNT(note) AS noted
3FROM readings
3FROM readings
4GROUP BY station
4GROUP BY station
5ORDER BY station;
5ORDER BY station;

What sgl generates

RecordReadingsdiffers

DuckDB
PostgreSQL
1func (queries *Queries) RecordReadings(ctx context.Context, params []RecordReadingsParams) error
1func (queries *Queries) RecordReadings(ctx context.Context, params []RecordReadingsParams) error
2
2
3type RecordReadingsParams struct {
3type RecordReadingsParams struct {
4	Station  string
4	Station  string
5	Minute   int32
5	Minute   int64
6	HeightCm int32
6	HeightCm int64
7	Note     *string
7	Note     *string
8}
8}

SummariseReadingsidentical

DuckDB
PostgreSQL
1func (queries *Queries) SummariseReadings(ctx context.Context) ([]SummariseReadingsRow, error)
1func (queries *Queries) SummariseReadings(ctx context.Context) ([]SummariseReadingsRow, error)
2
2
3type SummariseReadingsRow struct {
3type SummariseReadingsRow struct {
4	Station  string `json:"station"`
4	Station  string `json:"station"`
5	Readings int64  `json:"readings"`
5	Readings int64  `json:"readings"`
6	Noted    int64  `json:"noted"`
6	Noted    int64  `json:"noted"`
7}
7}

The program

example.godiffers

DuckDB
PostgreSQL
1package main
1package main
2
2
3import (
3import (
4	"context"
4	"context"
5	"database/sql"
5	"database/sql"
6	"fmt"
6	"fmt"
7	"io"
7	"io"
8
8
9	"seagull.sh/seagull/examples/duckdb/022_bulk_writes/internal/store"
9	"seagull.sh/seagull/examples/postgres/022_bulk_writes/internal/store"
10)
10)
11
11
12const (
12const (
13	// minutesPerWeek is how many readings each tide gauge takes in a week, one a minute.
13	// minutesPerWeek is how many readings each tide gauge takes in a week.
14	minutesPerWeek = 7 * 24 * 60
14	minutesPerWeek = 7 * 24 * 60
15)
15)
16
16
17// statementCounter is a store.DBTX that counts the statements sent through ExecContext.
17// statementCounter is a store.DBTX that counts the statements sent through ExecContext.
18// QueryContext and QueryRowContext come from the embedded *sql.DB.
18// QueryContext and QueryRowContext come from the embedded *sql.DB.
19type statementCounter struct {
19type statementCounter struct {
20	*sql.DB
20	*sql.DB
21
21
22	// statements is how many times ExecContext has been called.
22	// statements is how many times ExecContext has been called.
23	statements int
23	statements int
24}
24}
25
25
26// ExecContext counts the statement, then runs it on the database.
26// ExecContext counts the statement, then runs it on the database.
27//
27//
28// Takes query (string) which is the SQL to run.
28// Takes query (string) which is the SQL to run.
29// Takes arguments (...any) which are its bind values.
29// Takes arguments (...any) which are its bind values.
30//
30//
31// Returns sql.Result which is the database's result.
31// Returns sql.Result which is the database's result.
32// Returns error when the statement fails.
32// Returns error when the statement fails.
33func (counter *statementCounter) ExecContext(ctx context.Context, query string, arguments ...any) (sql.Result, error) {
33func (counter *statementCounter) ExecContext(ctx context.Context, query string, arguments ...any) (sql.Result, error) {
34	counter.statements++
34	counter.statements++
35	return counter.DB.ExecContext(ctx, query, arguments...)
35	return counter.DB.ExecContext(ctx, query, arguments...)
36}
36}
37
37
38// run records a week of readings from two tide gauges in one batch call, shows how many
38// run records a week of readings from two tide gauges in one batch call, shows how many
39// INSERT statements that took, and summarises what landed in the table.
39// INSERT statements that took, and summarises what landed in the table.
40//
40//
41// Takes database (*sql.DB) which is a migrated database.
41// Takes database (*sql.DB) which is a migrated database.
42// Takes output (io.Writer) which receives the statement counts and the summary.
42// Takes output (io.Writer) which receives the statement counts and the summary.
43//
43//
44// Returns error when a query fails.
44// Returns error when a query fails.
45func run(ctx context.Context, database *sql.DB, output io.Writer) error {
45func run(ctx context.Context, database *sql.DB, output io.Writer) error {
46	counter := &statementCounter{DB: database}
46	counter := &statementCounter{DB: database}
47	queries := store.New(counter)
47	queries := store.New(counter)
48
48
49	// An empty batch is a no-op: nothing is sent to the database.
49	// An empty batch is a no-op: nothing is sent to the database.
50	if err := queries.RecordReadings(ctx, nil); err != nil {
50	if err := queries.RecordReadings(ctx, nil); err != nil {
51		return fmt.Errorf("recording no readings: %w", err)
51		return fmt.Errorf("recording no readings: %w", err)
52	}
52	}
53	fmt.Fprintf(output, "0 readings sent in %d statements\n", counter.statements)
53	fmt.Fprintf(output, "0 readings sent in %d statements\n", counter.statements)
54
54
55	readings := tideReadings("harbour", "pier")
55	readings := tideReadings("harbour", "pier")
56
56
57	// A batch query takes a slice of Params, one per row. On database/sql the generated
57	// A batch query takes a slice of Params, one per row. On database/sql the generated
58	// method writes them as multi-row INSERTs, chunked so that no statement binds more than
58	// method writes them as multi-row INSERTs, renumbering the placeholders $1, $2 and on,
59	// 65535 variables, the limit declared by Seagull's DuckDB dialect: four columns a row
59	// chunked so that no statement binds more than 65535 variables, the most the PostgreSQL
60	// allows 16383 rows a statement, so a week's 20160 rows take two. copyfrom generates the
60	// protocol can carry: four columns a row allows 16383 rows a statement, so a week from
61	// same code on database/sql.
61	// two gauges takes two. copyfrom generates the same code on database/sql; only the pgx
62	// driver uses COPY.
62	//
63	//
63	// Each chunk is its own statement, so run the call inside RunInTx (concept 23) when
64	// Each chunk is its own statement, so run the call inside RunInTx (concept 23) when
64	// every row must land or none.
65	// every row must land or none.
65	if err := queries.RecordReadings(ctx, readings); err != nil {
66	if err := queries.RecordReadings(ctx, readings); err != nil {
66		return fmt.Errorf("recording readings: %w", err)
67		return fmt.Errorf("recording readings: %w", err)
67	}
68	}
68	fmt.Fprintf(output, "%d readings sent in %d statements\n", len(readings), counter.statements)
69	fmt.Fprintf(output, "%d readings sent in %d statements\n", len(readings), counter.statements)
69
70
70	summary, err := queries.SummariseReadings(ctx)
71	summary, err := queries.SummariseReadings(ctx)
71	if err != nil {
72	if err != nil {
72		return fmt.Errorf("summarising readings: %w", err)
73		return fmt.Errorf("summarising readings: %w", err)
73	}
74	}
74	for _, row := range summary {
75	for _, row := range summary {
75		fmt.Fprintf(output, "%-8s %d readings, %d with a note\n", row.Station, row.Readings, row.Noted)
76		fmt.Fprintf(output, "%-8s %d readings, %d with a note\n", row.Station, row.Readings, row.Noted)
76	}
77	}
77	return nil
78	return nil
78}
79}
79
80
80// tideReadings makes one reading a minute for a week at each station. The height rises
81// tideReadings makes one reading a minute for a week at each station. The height rises
81// and falls in a simple saw-tooth, and the reading on the hour carries a note.
82// and falls in a simple saw-tooth, and the reading on the hour carries a note.
82//
83//
83// Takes stations (...string) which name the gauges.
84// Takes stations (...string) which name the gauges.
84//
85//
85// Returns []store.RecordReadingsParams which holds every reading, ready for the batch.
86// Returns []store.RecordReadingsParams which holds every reading, ready for the batch.
86func tideReadings(stations ...string) []store.RecordReadingsParams {
87func tideReadings(stations ...string) []store.RecordReadingsParams {
87	readings := make([]store.RecordReadingsParams, 0, len(stations)*minutesPerWeek)
88	readings := make([]store.RecordReadingsParams, 0, len(stations)*minutesPerWeek)
88	for _, station := range stations {
89	for _, station := range stations {
89		for minute := range int32(minutesPerWeek) {
90		for minute := range int64(minutesPerWeek) {
90			var note *string
91			var note *string
91			if minute%60 == 0 {
92			if minute%60 == 0 {
92				note = new(fmt.Sprintf("hour %d check", minute/60))
93				note = new(fmt.Sprintf("hour %d check", minute/60))
93			}
94			}
94			readings = append(readings, store.RecordReadingsParams{
95			readings = append(readings, store.RecordReadingsParams{
95				Station:  station,
96				Station:  station,
96				Minute:   minute,
97				Minute:   minute,
97				HeightCm: 200 + minute%360,
98				HeightCm: 200 + minute%360,
98				Note:     note,
99				Note:     note,
99			})
100			})
100		}
101		}
101	}
102	}
102	return readings
103	return readings
103}
104}

Go type differences

The same SQL types map to different Go types, because each database reports them differently.

Query	Field	DuckDB	PostgreSQL
RecordReadings	parameter `minute`	`int32`	`int64`
RecordReadings	parameter `height_cm`	`int32`	`int64`