Delays Are Punished

To prevent the chkurl() worker from dilly-dallying, line 16 sets a context with a timeout, which it passes as the ctx variable to the worker bee. It uses the default net/http package in an inner goroutine (starting in line 44) running inside an outer goroutine to retrieve the web page, and pushes the status code returned by the web server into the local httpch channel when it's available.

This means that the inner goroutine is stuck until someone at the other end of the httpch channel starts reading. This will happen in the subsequent select construct starting in line 53, which kicks in on one of two events: either when a web request response comes from the httpch channel, or when the main program loses patience and ctx.Done() returns with an error. In the latter case, the function outputs a Timeout!! message and sets the HTTP return code to 501. If, on the other hand, everything works just fine, line 55-56 pushes the returned code and the associated URL into the results channel.

Figure 2 shows the flow of the compiled binary. The first three requests come back relatively quickly. The fourth one would keep dawdling for a massive five seconds, but the context timeout of two seconds as defined in line 16 sends an early termination signal, and the worker bee's status code comes back as a 501, as specified in line 60.

Figure 2: The results from the first three websites are returned quickly, but the fourth is too slow. This prompts the context to call a timeout.

The combination of two nested goroutines in chkurl() is necessary, by the way, because sending data into a channel and extracting the data at the other end needs to be synchronized. Letting the program simply send to the channel first and then read from it will not work, as the sender will hang forever if no receiver is listening. To ensure that the whole enchilada in Listing 3 runs smoothly, the sender sends its values to the channel in an asynchronous goroutine. The receiver can dock afterwards at their leisure; as soon as they do so, the sender unblocks and data gets sent through the channel.

Precision, No Sleep

You might have noticed that Listing 3 does not rely on unreliable Sleep commands for synchronization when collecting worker data, as done sloppily in Listings 1 and 2. Sleeping for a set time is no guarantee that things have finished, as data sent over the network can arrive unusually slowly. Anything is possible on the Internet; worst case, the program might exit before the last goroutine had a chance to send its results up the channel.

Listing 3 therefore uses two for loops in lines 28 and 32, each of which counts to four: once to call chkurl() four times and later to collect the results from the return channel four times. The order of requests and responses can get mixed up this way, but the program structure guarantees that all the results are complete and nothing gets accidentally dropped.

Goroutines are a great way to deploy code with components that can run concurrently. Note, however, that the concurrent code will only run in parallel if the platform supports it (e.g., thanks to multithreading and/or a multi-core processor). The difference between concurrency and parallelism is therefore quite relevant, as Go guru Rob Pike impressively explains in a video [2] that is definitely worth watching.

The Author

Mike Schilli works as a software engineer in the San Francisco Bay area, California. Each month in his column, which has been running since 1997, he researches practical applications of various programming languages. If you email him at mailto:mschilli@perlmeister.com he will gladly answer any questions.