dbx/README.md

dbx

PostgreSQL cluster library for Go built on pgx/v5. Master/replica routing, automatic retries with exponential backoff, round-robin load balancing, background health checking, panic-safe transactions, and context-based Querier injection.

go get git.codelab.vc/pkg/dbx

Quick start

cluster, err := dbx.NewCluster(ctx, dbx.Config{
    Master: dbx.NodeConfig{
        Name: "master",
        DSN:  "postgres://user:pass@master:5432/mydb",
        Pool: dbx.PoolConfig{MaxConns: 20, MinConns: 5},
    },
    Replicas: []dbx.NodeConfig{
        {Name: "replica-1", DSN: "postgres://user:pass@replica1:5432/mydb"},
        {Name: "replica-2", DSN: "postgres://user:pass@replica2:5432/mydb"},
    },
})
if err != nil {
    log.Fatal(err)
}
defer cluster.Close()

// Write → master
cluster.Exec(ctx, "INSERT INTO users (name) VALUES ($1)", "alice")

// Read → replica with automatic fallback to master
rows, err := cluster.ReadQuery(ctx, "SELECT * FROM users WHERE active = $1", true)

// Transaction → master, panic-safe
cluster.RunTx(ctx, func(ctx context.Context, tx pgx.Tx) error {
    tx.Exec(ctx, "UPDATE accounts SET balance = balance - $1 WHERE id = $2", 100, fromID)
    tx.Exec(ctx, "UPDATE accounts SET balance = balance + $1 WHERE id = $2", 100, toID)
    return nil
})

Components

Component	What it does
Cluster	Entry point. Connects to master + replicas, routes queries, manages lifecycle.
Node	Wraps pgxpool.Pool with health state and a human-readable name.
Balancer	Interface for replica selection. Built-in: RoundRobinBalancer.
retrier	Exponential backoff with jitter, node fallback, custom error classifiers.
healthChecker	Background goroutine that pings all nodes on an interval.
Querier injection	InjectQuerier / ExtractQuerier — context-based Querier for service layers.
MetricsHook	Optional callbacks: query start/end, retry, node up/down, replica fallback.

Routing

The library uses explicit method-based routing (no SQL parsing):

                          ┌──────────────┐
                          │   Cluster    │
                          └──────┬───────┘
                                 │
                 ┌───────────────┴───────────────┐
                 │                               │
            Write ops                       Read ops
     Exec, Query, QueryRow            ReadQuery, ReadQueryRow
     Begin, BeginTx, RunTx
     CopyFrom, SendBatch
                 │                               │
                 ▼                               ▼
          ┌──────────┐              ┌────────────────────────┐
          │  Master  │              │  Balancer → Replicas   │
          └──────────┘              │  fallback → Master     │
                                    └────────────────────────┘

Direct node access: cluster.Master() and cluster.Replica() return DB.

Multi-replica setup

cluster, _ := dbx.NewCluster(ctx, dbx.Config{
    Master: dbx.NodeConfig{
        Name: "master",
        DSN:  "postgres://master:5432/mydb",
        Pool: dbx.PoolConfig{MaxConns: 20, MinConns: 5},
    },
    Replicas: []dbx.NodeConfig{
        {Name: "replica-1", DSN: "postgres://replica1:5432/mydb"},
        {Name: "replica-2", DSN: "postgres://replica2:5432/mydb"},
        {Name: "replica-3", DSN: "postgres://replica3:5432/mydb"},
    },
    Retry: dbx.RetryConfig{
        MaxAttempts: 5,
        BaseDelay:   100 * time.Millisecond,
        MaxDelay:    2 * time.Second,
    },
    HealthCheck: dbx.HealthCheckConfig{
        Interval: 3 * time.Second,
        Timeout:  1 * time.Second,
    },
})
defer cluster.Close()

Transactions

RunTx is panic-safe — if the callback panics, the transaction is rolled back and the panic is re-raised:

err := cluster.RunTx(ctx, func(ctx context.Context, tx pgx.Tx) error {
    _, err := tx.Exec(ctx, "UPDATE accounts SET balance = balance - $1 WHERE id = $2", amount, fromID)
    if err != nil {
        return err
    }
    _, err = tx.Exec(ctx, "UPDATE accounts SET balance = balance + $1 WHERE id = $2", amount, toID)
    return err
})

For custom isolation levels use RunTxOptions:

cluster.RunTxOptions(ctx, pgx.TxOptions{
    IsoLevel: pgx.Serializable,
}, fn)

Context-based Querier injection

Pass the Querier through context so service layers work both inside and outside transactions:

// Repository
func CreateUser(ctx context.Context, db dbx.Querier, name string) error {
    q := dbx.ExtractQuerier(ctx, db)
    _, err := q.Exec(ctx, "INSERT INTO users (name) VALUES ($1)", name)
    return err
}

// Outside transaction — uses cluster directly
CreateUser(ctx, cluster, "alice")

// Inside transaction — uses tx
cluster.RunTx(ctx, func(ctx context.Context, tx pgx.Tx) error {
    ctx = dbx.InjectQuerier(ctx, tx)
    return CreateUser(ctx, cluster, "alice") // will use tx from context
})

Error classification

dbx.IsRetryable(err)          // connection errors, serialization failures, deadlocks, too_many_connections
dbx.IsConnectionError(err)    // PG class 08 + common connection error strings
dbx.IsConstraintViolation(err) // PG class 23 (unique, FK, check violations)
dbx.PgErrorCode(err)          // extract raw PG error code

Sentinel errors: ErrNoHealthyNode, ErrClusterClosed, ErrRetryExhausted.

dbxtest helpers

The dbxtest package provides test helpers:

func TestMyRepo(t *testing.T) {
    cluster := dbxtest.NewTestCluster(t, dbx.WithLogger(&dbxtest.TestLogger{T: t}))
    // cluster is auto-closed when test finishes
    // skips test if DB is not reachable
}

Set DBX_TEST_DSN env var to override the default DSN (postgres://postgres:postgres@localhost:5432/dbx_test?sslmode=disable).

Requirements

Go 1.24+, pgx/v5.