Make all integration tests safe to run in parallel

[corylanou]

Packages test in parallel by default, and with the recent addition of integration tests in other packages, this is now creating intermittent port collision, and hence, test failures.

This change also required the following changes:

• Track Graphite listeners so we could close them properly and assign 0 port for testing
• Track OpenTSDB listeners so we could close them properly and assign 0 port for testing
• Change they way we do default config values for Graphite/OpenTSDB endpoints as to allow for testing. All other default styles should also likely move to this new style.

Before the change:

$ time go test ./... -p=1
ok      github.com/influxdb/influxdb    3.597s
ok      github.com/influxdb/influxdb/admin      0.011s
ok      github.com/influxdb/influxdb/client     0.009s
ok      github.com/influxdb/influxdb/cmd/influx 4.647s
ok      github.com/influxdb/influxdb/cmd/influxd        111.165s
ok      github.com/influxdb/influxdb/collectd   0.007s
ok      github.com/influxdb/influxdb/graphite   0.006s
ok      github.com/influxdb/influxdb/httpd      1.539s
ok      github.com/influxdb/influxdb/influxql   0.013s
ok      github.com/influxdb/influxdb/messaging  3.766s
?       github.com/influxdb/influxdb/opentsdb   [no test files]
ok      github.com/influxdb/influxdb/raft       8.218s
?       github.com/influxdb/influxdb/statik     [no test files]
?       github.com/influxdb/influxdb/test       [no test files]
?       github.com/influxdb/influxdb/tests/urlgen       [no test files]
?       github.com/influxdb/influxdb/udp        [no test files]
?       github.com/influxdb/influxdb/uuid       [no test files]

real    2m21.503s
user    0m20.528s
sys     0m7.652s

After the change:

$ time go test ./... -parallel 100
ok      github.com/influxdb/influxdb    3.605s
ok      github.com/influxdb/influxdb/admin      0.012s
ok      github.com/influxdb/influxdb/client     0.011s
ok      github.com/influxdb/influxdb/cmd/influx 3.458s
ok      github.com/influxdb/influxdb/cmd/influxd        57.129s
ok      github.com/influxdb/influxdb/collectd   0.012s
ok      github.com/influxdb/influxdb/graphite   0.006s
ok      github.com/influxdb/influxdb/httpd      1.497s
ok      github.com/influxdb/influxdb/influxql   0.012s
ok      github.com/influxdb/influxdb/messaging  3.739s
?       github.com/influxdb/influxdb/opentsdb   [no test files]
ok      github.com/influxdb/influxdb/raft       7.043s
?       github.com/influxdb/influxdb/statik     [no test files]
?       github.com/influxdb/influxdb/test       [no test files]
?       github.com/influxdb/influxdb/tests/urlgen       [no test files]
?       github.com/influxdb/influxdb/udp        [no test files]
?       github.com/influxdb/influxdb/uuid       [no test files]

real    0m59.438s
user    0m20.251s
sys     0m5.510s

[otoolep]

Style: any reason there are multiple blank lines here?

[otoolep]

I like the change -- especially the randomization of ports.

[otoolep]

There is a tiny window here where the port has been freed, and could then be re-used by the next caller of newPort(), and still be used by the first caller?

[corylanou]

When watching the ports log out it always went up continuously. Could it create a race condition? I'm not sure. I suspect it has more to do with how the underlying platform implements it. I will ask on Gopher slack though to see if anyone knows if there is some type of delay at the system level before it puts that port back on the market.

[otoolep]

Unfortunately we confirmed that on Ubuntu 14.04, port collision occur in this scenario.

[corylanou]

I'm not super excited about this. I'm open to suggestions.

[otoolep]

How about?

return &url.URL{
    Scheme: "http",
    host: s.ClusterAddr().String(),
}

[otoolep]

Some small feedback, +1. Don't forget the changelog.

[otoolep]

All done in a different PR.