test fails when more than five pixels are requested

[gka]

While now the simple test with just 3 requests succeeds, a very similar test with six requests fails

❯ cake test                                                                                                                        
4:  one
8:  two
6:  three
--- flushed ---
Test Failed  { one: 4, two: 8, three: 6 }

Apparently the last requests is timed out, and the flush is only happening after the 60-seconds interval defined in test/config.json has passed. Also note that the counts are completely wrong, the expected result should be either

{ "one": 1, "two": 2, "three": 3 }

or, since the last request failed, at least:

{ "one": 1, "two": 2, "three": 2 }

I could reproduce this multiple counting of single pixel.gif requests in a real server setting, so this is a serious bug.

[knowtheory]

Hey @gka, I can replicate that the test you've linked there fails. But i wanted to make sure we had a different external check, so i wrote up a little set of ruby scripts just as a sanity check, and things seem to be working as intended there: https://gist.github.com/knowtheory/e3c370ad28a6503a9cf6

[thejefflarson]

I'm looking at this, and yes, it seems the last request in the test isn't going through, but this:

 I could reproduce this multiple counting of single pixel.gif requests in a real server setting, 
 so this is a serious bug.

Seems like a different bug? Are you saying that the requests are being double counted? Cause that is definitely strange.

[gka]

Yes they are counted multiple times. There are exactly five requests (maybe six), but the resulting JSON data that is flushed to the test endpoint is { one: 4, two: 8, three: 6 }, which indicates 18 requests. I was also testing this on a remote server, and I fired exactly 100 requests but pixel-ping flushed way more than that. It is indeed a strange bug...

[knowtheory]

I think we need to revisit the test, because all the external checks i've written in ruby have indicated that pixel-ping isn't double counting even across flush boundaries.

I took the scripts from the gist i posted above, upped the sleep to 5 seconds, and did 100 count per threads and got the expected results https://gist.github.com/knowtheory/2accf4151d6cdd07fe7e