For anyone new to building web applications with Go, it's important to realise that all incoming HTTP requests are served in their own goroutine. This means that any code in or called by your application handlers will be running concurrently, and there is a risk of race conditions occurring.
In case you're not familiar with race conditions, I'll quickly explain the risk with an example. Consider a situation where you have two goroutines try to add money to a shared bank balance at the same time, like so:
|Instruction||Goroutine 1||Goroutine 2||Bank Balance|
|1||Read balance ⇐ £50||£50|
|2||Read balance ⇐ £50||£50|
|3||Add £100 to balance||—|
|4||Add £50 to balance||—|
|5||Write balance ⇒ £150||£150|
|6||Write balance ⇒ £100||£100|
Despite making two separate deposits, only the second one is reflected in the final balance because the two goroutines were racing each other to make the change.
The Go blog describes the downsides:
Race conditions are among the most insidious and elusive programming errors. They typically cause erratic and mysterious failures, often long after the code has been deployed to production. While Go's concurrency mechanisms make it easy to write clean concurrent code, they don't prevent race conditions. Care, diligence, and testing are required.
This specific type of race condition is known as a data race. And they can occur when two or more goroutines try to use a piece of shared data (in this example the bank balance) at the same time, but the result of their operations is dependent on the exact order that the scheduler executes their instructions.
Go provides a number of tools to help us avoid data races. These include Channels for communicating data between goroutines, a Race Detector for monitoring unsynchronized access to memory at runtime, and a variety of 'locking' features in the Atomic and Sync packages. One of these features are Mutual Exclusion locks, or mutexes, which we'll be looking at in the rest of this post.
Creating a Basic Mutex
Let's create some toy code to mimic the bank balance example:
We know that if there are multiple goroutines using this code and calling
myBalance.Display() frequently enough, then at some point a data race is likely to occur.
One way we could prevent the data race is to ensure that if one goroutine is using the
myBalance variable, then all other goroutines are prevented (or mutually excluded) from using it at the same time.
We can do this by creating a Mutex and setting a lock around particular lines of code with it. While one goroutine holds the lock, all other goroutines are prevented from executing any lines of code protected by the same mutex, and are forced to wait until the lock is yielded before they can proceed.
In practice, it's more simple than it sounds:
Here we've created a new mutex and assigned it to
mu. We then use
mu.Lock() to create a lock immediately before both racy parts of the code, and
mu.Unlock() to yield the lock immediately after.
There's a couple of things to note:
- The same mutex variable can be used in multiple places throughout your code. So long as it's the same mutex (in our case
mu) then none of the chunks of code protected by it can be executed at the same time.
- Holding a mutex lock doesn't 'protect' a memory location from being read or updated. A non-mutex-locked line of code could still access it at any time and create a race condition. Therefore you need to be careful to make sure all points in your code which are potentially racy are protected by the same mutex.
Let's tidy up the example a bit:
So what's changed here?
Because our mutex is only being used in the context of a
balance object, it makes sense to embed it in the
balance struct (an idea borrowed from Andrew Gerrard's excellent
10 things you (probably) don't know about Go slideshow). If you look at a larger codebase with lots of mutexes, like Go's HTTP Server, you can see how this approach helps to keep locking rules nice and clear.
We've also made use of the defer statement, which ensures that the mutex gets unlocked immediately before the function executing it returns. This is common practice for functions that contain multiple return statements, or where the return statement itself is racy like in our example.
Read Write Mutexes
It is important point to emphasize that data races aren't a concern if the only thing you are doing concurrently is reading the shared data.
In our bank balance example, having a full mutex lock on the
Display() function isn't strictly necessary. It would be OK for us to have multiple reads of
myBalance happening at the same time, so long as nothing is being written.
We can achieve this using RWMutex, a reader/writer mutual exclusion lock which allows any number of readers to hold the lock or one writer. This tends to be more efficient than using a full mutex in situations where you have a high ratio of reads to writes.
Reader locks can be opened and closed with
RUnlock() like so: