alexcesaro / statsdbench Goto Github PK

#1. Fix destination address

The const addr is being used as the destination address for the test clients. It seems the intent was for this address to send to the server instance that is started earlier in the benchmarks.

c, err := cactus.NewBufferedClient(addr, prefix, flushPeriod, 1432)

This should instead probably be...

c, err := cactus.NewBufferedClient(s.conn.LocalAddr().String(), prefix, flushPeriod, 1432)

#2. Reset timers so setup is not counted as part of benchmarks.

func BenchmarkCactus(b *testing.B) {
    s := newServer()
    c, err := cactus.NewBufferedClient(s.conn.LocalAddr().String(), prefix, flushPeriod, 1432)
    if err != nil {
        b.Fatal(err)
    }
    b.ResetTimer()  ## <== this is the key here. this reset the counters and timers,
                    ##     so only the following lines are counted.

    for i := 0; i < b.N; i++ {
        c.Inc(counterKey, 1, 1)
        c.Gauge(gaugeKey, gaugeValue, 1)
        c.Timing(timingKey, int64(timingValue), 1)
    }
    c.Close()
    s.Close()
}

The two above changes should help the clients uniformly (at the very least #2 will).

It seems to me the current code gives a misleading benchmark, since the Timing() method of the alexcesaro client takes in interface{} as its value - which is passed all the way down to appendNumber(), but since it's not a numeric it's not appended at all.

So calls to
func (c *Client) Timing(bucket string, value interface{})
should take something which appendNumber can recognizes. But not a time.Duration.