pgdriver: discarding bad listener connection

[ivanakimov]

When running something similar to this, sometimes I notice:

pgdriver: 2022/06/15 07:16:00 listener.go:321: pgdriver: &{%!s(*context.emptyCtx=0xc00003a038) %!s(*pgdriver.Listener=&{0xc00007c720 0xc001473620 [queue_channel bun:ping] {0 0} 0xc0002af2c0 false 0xc008666a80}) %!s(int=100) %!s(time.Duration=5000000000) %!s(time.Duration=60000000000) %!s(chan pgdriver.Notification=0xc00007c900) %!s(chan struct {}=0xc003a7e000)} channel is full for 1m0s (notification is dropped)

pgdriver: 2022/06/15 07:16:00 listener.go:115: bun: discarding bad listener connection: bun: ping timeout

Two questions:

1. When do full channels and bad listener connections happen?
2. Is there anything to do about these, or are they simply notifications?

[vmihailenco]

That error happens when you can't consume all messages from the Listener channel fast enough. Probably the consuming loop crashed or is blocked for too long.

[comerc]

I got a stable reproduction of the error.

Change this lines: https://github.com/uptrace/bun/blob/master/example/pg-listen/main.go#L41-L43

To above:

	// if err := ln.Listen(ctx, "mychan1", "mychan2"); err != nil {
	// 	panic(err)
	// }

	for {
		select {
		case notif := <-ln.Channel():
			fmt.Println(notif.Channel, notif.Payload)
		}
	}

[john8329]

More like "make it like the example in the link". select was the cause of my problem for this, you need one goroutine with a range loop for each channel you listen to.