add support for custom hostname

[andyxning]

pynsq now only supports override user_agent and not supporting override hostname.

In cases we run a consumer in a container, the hostname shell command will return a docker id which is a random string. When we check the client info in nsqadmin we can not get any useful info about the host name where consumer running.

With enable override hostname, we can set the consumer, running in a container, hostname to the host name where the container is running which is helpful when we want to determine which host to get the consumers.

/cc @mreiferson @alpaker

[andyxning]

@mreiferson @alpaker

The error in travis ci seems nothing to do with the change. Could you please take a look at this?

[ploxiln]

Just looking at the test matrix, it seems to be the combo of python-3.5 and tornado-3.2 that fails.

[andyxning]

@ploxiln Can you please help to figure out why it is failed.

[andyxning]

@ploxiln I am not familiar with tornado. I googled the error and change the test by adding a single IOLoop instance according to tornado #663 comment.

Also according to this google group thread, it means that run_sync with a coroutine that also calls run_sync for TimeoutError: Operation timed out after None seconds error.

@mreiferson @alpaker could you please take a look at this?

[ploxiln]

We'll investigate, it'll just be a couple days ...

[andyxning]

@ploxiln Thanks.

BTW, could you also please take a look at nsq #921.

[mreiferson]

@andyxning thanks for the contribution!

I reran those broken builds and they worked, which means the issue is transient. From reading the links you've pasted, it looks like we've got some object(s) somewhere that aren't using the correct IOLoop instance.

I'm not particularly concerned with fully investigating this and fixing it "correctly", so I'd rather just leave the test code alone for now. Can you remove those changes and we can land this?

[andyxning]

@mreiferson I have removed the changes to test. But it will error again in CI. PTAL.