duration 420m streams messages on STDOUT

[agenteo]

using a duration higher then 420 starts displaying on STDOUT a stream of:

goroutine 114338 [select]:
^Created by github.com/tsenart/vegeta/lib.(*Attacker).Attack
        /Users/tomas/Code/go/src/github.com/tsenart/vegeta/lib/attack.go:154 +0x382

is it just a warning? I did not go trough running the app. Having 240 doesn't cause that stream.

[tsenart]

With such high durations you're better off setting the -workers flag to something like 1000. By default, Vegeta will allocate one go-routine per request as it was never designed for extremely long load tests such as the one you're doing.

By curiosity, what are you expecting to learn from a load test with this sort of parameters?

[agenteo]

thanks I'll try with workers and get back to you.

Our app uses a library calling a 3rd party API, it keep the response data in our app memory and expires after a certain amount of hours. I wanted to monitor the refetching process under load. Ideally the library would store data on a key value cache that I could manually flush.

[tsenart]

@agenteo: Any interesting results to share?

[agenteo]

Preamble, I am running this off a Macbook pro 2.6GHz i7 OSX 10.9.5 waiting for my devops to provide an EC2 instance.

I started the following attack:

vegeta attack -targets=vegeta_test_plan_dev.txt -workers=1000 -duration=360m -rate=20 > results_$(date +%Y%m%d_%H%M%S)_dev_loadtest.bin

the file has ~2000 URLs. My expectation was for the test to be running for 6 hours. I left this running Friday afternoon.

Today, Monday I did not see any goroutine output but the script was still running. When I looked at my app I did not see traffic, so I CTRL-C vegeta and looked at the report. It said it was running for 63 hours:

Requests        [total]                         78784
Duration        [total, attack, wait]           63h4m54.147124256s, 63h4m54.100175493s, 46.948763ms
Latencies       [mean, 50, 95, 99, max]         6.117199907s, 93.874484ms, 592.261451ms, 61h59m16.388871571s, 61h59m16.388871571s
Bytes In        [total, mean]                   4942771740, 62738.27
Bytes Out       [total, mean]                   0, 0.00
Success         [ratio]                         96.84%
Status Codes    [code:count]                    200:76291  505:780  404:40  0:1673

Opening the report plot I can see after 3800 seconds the red error line takes over.

I had this once before, this time I turned off HD power savings, power nap from my Mac. I asked for an EC2 instance where I'd like to run those tests I have a hunch this might be OSX related but I'll have to wait for the EC2 instance to confirm that.

I understand stretching tests for this long wasn't the design of vegeta but I appreciate you following this one up. Do you get any of this behavior when you load test for this long?

[tsenart]

What I can understand from the data you provided is that some requests never get a response. Vegeta doesn't timeout requests by default so it will wait forever in your case. It seems you would benefit from setting the -timeout option on your load test.

[agenteo]

I see, I'll try that overnight and update this.

[agenteo]

Actually I think we can close this one, --workers fixes the output issue I'll open a new one about long running tests going to sleep. Thanks