Files
golib/pg/integration_test.go
T
unkin-agent 5c23db8885
ci/woodpecker/pr/build Pipeline was successful
ci/woodpecker/pr/test Pipeline was successful
ci/woodpecker/pr/pre-commit Pipeline was successful
Add read/write splitting to the pg module
pg.Cluster wraps a primary pool and an optional read-replica pool.
Routing is explicit — Read(), Write(), Primary() — with no SQL
inspection: statement text misroutes CTE writes and SELECT ... FOR
UPDATE in both directions.

An unhealthy replica falls back to the primary behind a ping-based
circuit that retries on a doubling backoff window, and migrations always
run through Write().

pg.ClusterDSNsFromEnv resolves both endpoints from the environment,
naming the read-only host explicitly rather than deriving it from the
primary's.
2026-08-31 22:44:10 +10:00

240 lines
6.9 KiB
Go

package pg_test
import (
"context"
"errors"
"net/url"
"strings"
"testing"
"testing/fstest"
"time"
"git.unkin.net/unkin/golib/pg"
"git.unkin.net/unkin/golib/pg/pgtest"
)
// migrations is a two-file set exercising the things the runner promises:
// multi-statement files (simple protocol) and IF NOT EXISTS re-runnability.
var migrations = fstest.MapFS{
"0001_widgets.sql": {Data: []byte(`
CREATE TABLE IF NOT EXISTS widgets (id BIGSERIAL PRIMARY KEY, name TEXT NOT NULL);
CREATE INDEX IF NOT EXISTS widgets_name_idx ON widgets (name);
`)},
"0002_gadgets.sql": {Data: []byte(`CREATE TABLE IF NOT EXISTS gadgets (id BIGSERIAL PRIMARY KEY);`)},
"notes.md": {Data: []byte("ignored")},
}
func testCtx(t *testing.T) context.Context {
t.Helper()
ctx, cancel := context.WithTimeout(context.Background(), 3*time.Minute)
t.Cleanup(cancel)
return ctx
}
// The full path against a real server: connect, migrate, and confirm the schema
// and the bookkeeping table both landed.
func TestNewMigrated_AgainstRealPostgres(t *testing.T) {
ctx := testCtx(t)
dsn := pgtest.MustStartPostgres(ctx, t)
pool, err := pg.NewMigrated(ctx, dsn, migrations, pg.MigrateOptions{LockName: "golib-pg-integration"})
if err != nil {
t.Fatalf("NewMigrated: %v", err)
}
t.Cleanup(pool.Close)
for _, table := range []string{"widgets", "gadgets", "schema_migrations"} {
var exists bool
err := pool.QueryRow(ctx,
`SELECT EXISTS (SELECT 1 FROM information_schema.tables
WHERE table_schema = 'public' AND table_name = $1)`, table).Scan(&exists)
if err != nil {
t.Fatalf("check %s: %v", table, err)
}
if !exists {
t.Errorf("table %s was not created", table)
}
}
// The second statement of the multi-statement file must have run too.
var idx bool
if err := pool.QueryRow(ctx,
`SELECT EXISTS (SELECT 1 FROM pg_indexes WHERE indexname = 'widgets_name_idx')`).Scan(&idx); err != nil {
t.Fatalf("check index: %v", err)
}
if !idx {
t.Error("the second statement of the multi-statement migration did not run")
}
var versions []string
rows, err := pool.Query(ctx, "SELECT version FROM schema_migrations ORDER BY version")
if err != nil {
t.Fatalf("read schema_migrations: %v", err)
}
defer rows.Close()
for rows.Next() {
var v string
if err := rows.Scan(&v); err != nil {
t.Fatalf("scan: %v", err)
}
versions = append(versions, v)
}
if err := rows.Err(); err != nil {
t.Fatalf("rows: %v", err)
}
want := []string{"0001_widgets.sql", "0002_gadgets.sql"}
if len(versions) != len(want) {
t.Fatalf("recorded versions %v, want %v", versions, want)
}
for i := range want {
if versions[i] != want[i] {
t.Fatalf("recorded versions %v, want %v", versions, want)
}
}
}
// Every replica calls Migrate at startup. Running it concurrently against one
// server must apply the set exactly once and leave no lock held.
func TestMigrate_ConcurrentRepliesConverge(t *testing.T) {
ctx := testCtx(t)
dsn := pgtest.MustStartPostgres(ctx, t)
pool, err := pg.New(ctx, dsn, nil)
if err != nil {
t.Fatalf("New: %v", err)
}
t.Cleanup(pool.Close)
const replicas = 4
errs := make(chan error, replicas)
for range replicas {
go func() {
errs <- pg.Migrate(ctx, pool, migrations, pg.MigrateOptions{LockName: "golib-pg-integration"})
}()
}
for range replicas {
if err := <-errs; err != nil {
t.Fatalf("Migrate: %v", err)
}
}
var n int
if err := pool.QueryRow(ctx, "SELECT count(*) FROM schema_migrations").Scan(&n); err != nil {
t.Fatalf("count versions: %v", err)
}
if n != 2 {
t.Fatalf("schema_migrations has %d rows, want 2", n)
}
// Nothing may still hold the migration lock once every replica has finished.
var held bool
if err := pool.QueryRow(ctx,
`SELECT EXISTS (SELECT 1 FROM pg_locks WHERE locktype = 'advisory' AND objid IS NOT NULL AND granted)`,
).Scan(&held); err != nil {
t.Fatalf("check locks: %v", err)
}
if held {
t.Error("an advisory lock is still held after every replica finished")
}
}
// DSNFromEnv's output must be something pgx can actually connect with.
func TestDSNFromEnv_ConnectsToRealPostgres(t *testing.T) {
ctx := testCtx(t)
dsn := pgtest.MustStartPostgres(ctx, t)
t.Setenv("DATABASE_URL", "")
t.Setenv("APP_DATABASE_URL", "")
cfg, err := parseDSN(dsn)
if err != nil {
t.Fatalf("parse container DSN: %v", err)
}
t.Setenv("APP_DBHOST", cfg.host)
t.Setenv("APP_DBPORT", cfg.port)
t.Setenv("APP_DBUSER", cfg.user)
t.Setenv("APP_DBPASS", cfg.pass)
t.Setenv("APP_DBNAME", cfg.name)
t.Setenv("APP_DBSSL", "disable")
built, err := pg.DSNFromEnv("APP_")
if err != nil {
t.Fatalf("DSNFromEnv: %v", err)
}
pool, err := pg.New(ctx, built, nil)
if err != nil {
t.Fatalf("New(%q): %v", built, err)
}
pool.Close()
}
// A Cluster against a real server. One container plays both roles — what is
// under test is the routing, not Postgres' own replication — so the read pool
// is the same server reached through a second, distinct DSN.
func TestCluster_AgainstRealPostgres(t *testing.T) {
ctx := testCtx(t)
dsn := pgtest.MustStartPostgres(ctx, t)
c, err := pg.NewCluster(ctx, pg.ClusterConfig{
PrimaryDSN: dsn,
ReplicaDSN: dsn + "&application_name=reader",
// Short enough that the recovery probe lands inside the test.
ReplicaRetryMin: 10 * time.Millisecond,
ReplicaRetryMax: 10 * time.Millisecond,
})
if err != nil {
t.Fatalf("NewCluster: %v", err)
}
t.Cleanup(c.Close)
if c.Read() == c.Write() {
t.Fatal("a configured, healthy replica must be a distinct pool")
}
if err := c.Migrate(ctx, migrations, pg.MigrateOptions{LockName: "golib-pg-cluster-integration"}); err != nil {
t.Fatalf("Migrate: %v", err)
}
if _, err := c.Write().Exec(ctx, "INSERT INTO widgets (name) VALUES ($1)", "sprocket"); err != nil {
t.Fatalf("insert on the primary: %v", err)
}
var n int
if err := c.Read().QueryRow(ctx, "SELECT count(*) FROM widgets").Scan(&n); err != nil {
t.Fatalf("read: %v", err)
}
if err := c.Primary().QueryRow(ctx, "SELECT count(*) FROM widgets").Scan(&n); err != nil {
t.Fatalf("read-your-writes: %v", err)
}
if n != 1 {
t.Fatalf("widgets has %d rows, want 1", n)
}
// A reported replica failure moves reads to the primary; a replica that
// still answers earns them back on the next probe.
c.ReportReplicaError(errors.New("simulated replica failure"))
if c.Read() != c.Write() {
t.Fatal("a reported replica failure must move reads to the primary")
}
time.Sleep(20 * time.Millisecond)
if c.Read() == c.Write() {
t.Fatal("a replica that answers its probe must get reads back")
}
}
// dsnParts is the container DSN split back into the fields DSNFromEnv reads.
type dsnParts struct{ host, port, user, pass, name string }
func parseDSN(dsn string) (dsnParts, error) {
u, err := url.Parse(dsn)
if err != nil {
return dsnParts{}, err
}
pass, _ := u.User.Password()
return dsnParts{
host: u.Hostname(),
port: u.Port(),
user: u.User.Username(),
pass: pass,
name: strings.TrimPrefix(u.Path, "/"),
}, nil
}