Metrics endpoint closed before Loki modules

[rfratto]

Even when running an ingester with a sleep period of 30 seconds, no shutdown metrics (loki_ingester_sent_chunks, etc.) are scraped by Prometheus. This is seemingly because the endpoint serving metrics may be shut down before any of the modules close:

loki/pkg/loki/loki.go

Lines 140 to 146 in 459e4e3

	// Stop gracefully stops a Loki.
	func (t *Loki) Stop() error {
	t.stopping(t.cfg.Target)
	t.server.Shutdown()
	t.stop(t.cfg.Target)
	return nil
	}

The server should likely stay alive for as long as the modules are alive so metrics can continue to be collected during shutdown.