Golang Sync Mutex: Normal and Starvation Mode
victoriametrics.com> Mutex, or MUTual EXclusion, in Go is basically a way to make sure that only one goroutine is messing with a shared resource at a time. This resource can be a piece of code, an integer, a map, a struct, a channel, or pretty much anything.
This is untrue right? It can only protect code, not data, right?
The relationship with data is that it protects (controls) access to it
> This is untrue right? It can only protect code, not data, right?
Do traffic lights protect the flow of traffic, or do they protect people? It seems reductive to fixate on traffic lights stopping intersecting flows of traffic, when the intent is to ultimately protect people from slamming into (and thus injuring/killing) each other. It may be a tempting counterpoint that traffic lights don't do anything to protect someone from cancer, heart attack, etc -- but that's simply out of the scope of the protection that traffic lights are intended to provide.
Similarly, it seems odd to me to suggest that a mutex protects code -- what is it protecting that code from? How can the code be corrupted, damaged, or otherwise harmed? A mutex does prevent other non-lock-holding threads from running simultaneously, but that's all in service of protecting the data. The data is protected by imposing order on the execution of the code -- by serializing access to that data.
Controlling something isn't the same as protecting that something. A traffic light controls traffic, but it protects the people sharing the road. A mutex controls the execution of code, but it protects data.
> Do traffic lights protect the flow of traffic, or do they protect people? It seems reductive to fixate on traffic lights stopping intersecting flows of traffic, when the intent is to ultimately protect people from slamming into (and thus injuring/killing) each other.
Had the author of the article weighed in, then yes: They protect the flow of traffic, people, an int, pretty much anything!
> Similarly, it seems odd to me to suggest that a mutex protects code -- what is it protecting that code from?
It is protecting the code from simultaneous execution. The burden of protecting the data still falls on the programmer.
Mutexes literally appear so in the code. They wrap sections of imperative statements, or methods. They don't wrap data, eg:
> Controlling something isn't the same as protecting that something. A traffic light controls trafficMutex<Int> myInt;Great example. Drivers die in intersections. The cause is driver inattention. The traffic lights don't lose their licence.
I only asked it as a question because others could hypothetically point out that Go mutexes can protect data (and NOT via metaphor) - maybe I missed something!
mutexes, in general, are considered to protect data. one associates a mutex with a piece of data. any number of bits of code might access that data, not just a single synchronized function.
I have misgivings about such long spinlocks in user space. A millisecond is over a million instructions.
Does the spinlock last a millisecond? It sounds from the article that it lasts 120 cycles.
"Starvation mode kicks in if a goroutine fails to acquire the lock for more than 1 millisecond."
That may mean that after 1ms, the mutex becomes more fair. That seems to be a further fallback than ending spinning.
Go tends to use only one real thread per CPU, so the problems are different than they are in, say, Rust. Go is doing some CPU dispatching in user space.
Indeed, starvation mode is just a way to ensure fairness, at the cost of surrendering control and being woken up later. Mutexes can be costly when there is a lot of competition, it is up to us to use it correctly
Who's "us"? In Go, the threads are a shared resource across all the code. Multiple libraries can be contending for a resource.
I've hit something like this in Wine's storage allocator. Wine has its own "malloc", not the Microsoft one. It has three nested locks, and the innermost one is a raw spinlock. When a buffer is resized ("realloc"), and a new buffer has to be allocated, the copying is done inside a spinlock. If you have enough threads doing realloc calls (think "push") the whole thing goes into futex congestion collapse and performance drops by two orders of magnitude.
A spinlock in user space is an optimistic assumption. It's an assumption that the wait will be very short. It's important to have something tracking whether that assumption is working. At some load level, it's time to do an actual lock and give up control to another thread. "Starvation mode" may be intended to do that. But it seems more oriented towards preventing infinite overtaking.
Can you push Go's mutexes into futex congestion?
By “us”, I was referring to normal, everyday go devs (versus working on go itself).
I think starvation mode is a pragmatic way of solving the issue: overtaking happens when the fast path assumption is wrong - but absolute fairness is not a requirement. So it’s ok. The gains of the fast path is enough to justify having that starvation mode.
I would argue that if some piece of code ends up wasting lots of time on futex calls, it suggests that this code was not designed with the right use case in mind.
Interesting comment about wine’s allocator, did not know about this
It differs by CPU model.
So some CPU models wait 120 cycles and others wait a millisecond (millions of cycles)? That seems like a pretty drastic difference. I wonder why they would write code with such a drastic difference between CPU models.
No, Haswell's PAUSE is 9 cycles, the P-cores of Alder Lake are 160 cycles. Zen 4 is 65.