Performance issue

[grafana]

Performance issue

[Dieterbe]

this should fix it f1d6f48

[Dieterbe]

btw

~/g/s/g/g/carbon-relay-ng ❯❯❯ go test -bench=Send -run=XXX
PASS
BenchmarkSendAndReceive        1    1001640486 ns/op
ok      github.com/graphite-ng/carbon-relay-ng  2.622s

1second to send 1000 metrics. unacceptable of course. i will on some profiling and optimisations (right now it's spending way too much time in GC)

[Dieterbe]

actually, those numbers are offset because it needs to initiate the connection the testendpoint first.
also i did some GC/malloc fixes

$ go test -bench=Send -run=XXX
PASS
BenchmarkSendAndReceiveThousand        1    1004344092 ns/op # 995 metrics/s
BenchmarkSendAndReceiveTenThousand         1    1004823407 ns/op # 9951 metrics/s
BenchmarkSendAndReceiveHundredThousand         1    5006029614 ns/op # 19975 metrics/s
BenchmarkSendAndReceiveMillion         1    47004007110 ns/op # 21274 metrics/s
ok      github.com/graphite-ng/carbon-relay-ng  54.088s

so clearly for the lower amounts the connection overhead is the deciding factor, then it goes up to about 20k metrics/s or 1.2M per minute, this should be enough for me, but i might tune it a bit more.

your thoughts/findings?

[Dieterbe]

also note the test now sleeps 10 microseconds between every metric send because if the connection is slow the sender will drop packets and we don't want this in the tests. it's a little too "sensitive" in the current code, i should probably look into this and improve this but anyway because of this it's artificially capped to 100k per second.

[Dieterbe]

running idle, carbon-relay-ng takes about 0.5% cpu for me now :)

[pwielgolaski]

it looks much better now, thx