What's the expected way to do logging AND tracing?

[danielbprice]

Hi! I'm in the process of porting from v4 to v5. At our company, we use Datadog for tracing. But Datadog does not have a tracing wrapper for v4 pgx; it does have a way to add tracing to sql.DB. A pgx-native tracer seems to be in the works for v5. I prefer to use zerolog for logging.

As a result, in v4 I use a l somewhat complex initialization process:

        connConfig, err := pgx.ParseConfig(connString)
        ...
        // hook up zerolog
        connConfig.Logger = zerologadapter.NewLogger(...)
        ...
        // magic incantations
        connStr := stdlib.RegisterConnConfig(connConfig)
        driver := stdlib.GetDefaultDriver()
        Driver := driver.(*stdlib.Driver)
        connector, err := Driver.OpenConnector(connStr)
        if err != nil {
                return nil, err
        }

        // sqltrace "gopkg.in/DataDog/dd-trace-go.v1/contrib/database/sql"
        sqltrace.Register("pgx", &stdlib.Driver{}) // 
        db := sqltrace.OpenDB(connector)

This accomplishes zerolog for logging and Datadog for tracing and has been in production for a year.

As I look at the Tracer interface in v5, I'm having trouble understanding what the expected method is for accomplishing both logging and tracing? Do I need to implement an MultiTracer adaptor, which relays all of the Tracer calls to multiple underlying Tracers, one of which is logging, and one of which is my tracing client (Datadog)?

[jackc]

Do I need to implement an MultiTracer adaptor, which relays all of the Tracer calls to multiple underlying Tracers, one of which is logging, and one of which is my tracing client (Datadog)?

I think this is probably necessary.

[esiqveland]

I want to have a instrumented logger and the otel tracer.

Did you write a MultiTracer to support this use case?

[uh-zz]

I have the same problem.

I was able to set up both loggers and tracers by providing the following implementation in my application.

#1061 (comment)

Here is a sample of my implementation using pgx.

// https://github.com/jackc/pgx/issues/1061#issuecomment-1205942323
type MultiQueryTracer struct {
	Tracers []pgx.QueryTracer
}

func (m *MultiQueryTracer) TraceQueryStart(ctx context.Context, conn *pgx.Conn, data pgx.TraceQueryStartData) context.Context {
	for _, t := range m.Tracers {
		ctx = t.TraceQueryStart(ctx, conn, data)
	}

	return ctx
}

func (m *MultiQueryTracer) TraceQueryEnd(ctx context.Context, conn *pgx.Conn, data pgx.TraceQueryEndData) {
	for _, t := range m.Tracers {
		t.TraceQueryEnd(ctx, conn, data)
	}
}

func New(ctx context.Context, slogger *slog.Logger) *pgxpool.Pool {
	config, err := pgxpool.ParseConfig(dsn)
	if err != nil {
		panic(err)
	}

	// FYI: NewLoggerAdapter: https://github.com/mcosta74/pgx-slog
	adapterLogger := logger.NewLoggerAdapter(slogger)

	m := MultiQueryTracer{
		Tracers: []pgx.QueryTracer{
			// tracer: https://github.com/exaring/otelpgx
			otelpgx.NewTracer(),

			// logger
			&tracelog.TraceLog{
				Logger:   adapterLogger,
				LogLevel: tracelog.LogLevelTrace,
			},
		},
	}

	config.ConnConfig.Tracer = &m

	pool, err := pgxpool.NewWithConfig(ctx, config)
	if err != nil {
		panic(err)
	}

	if err := pool.Ping(ctx); err != nil {
		panic(err)
	}

	return pool
}

[zaydek]

If anyone's struggling with this, I threw this together based on a comment in this issue and another (which are in the code file). This uses the slog logger interface.

https://gist.github.com/zaydek/91f27cdd35c6240701f81415c3ba7c07