One thing I'm curious about here is the operational impact.
In production systems we often see Python services scaling horizontally because of the GIL limitations. If true parallelism becomes common, it might actually reduce the number of containers/services needed for some workloads.
But that also changes failure patterns — concurrency bugs, race conditions, and deadlocks might become more common in systems that were previously "protected" by the GIL.
It will be interesting to see whether observability and incident tooling evolves alongside this shift.
This is surely why Facebook was interested in funding this work. It is common to have N workers or containers of Python because you are generally restricted to one CPU core per Python process (you can get a bit higher if you use libs that unlock the GIL for significant work). So the only scaling option is horizontal because vertical scaling is very limited. The main downside of this was memory usage. You would have to load all of your code and libraries N types and in-process caches would become less effective. So by being able to vertically scale a Python process much further you can run less and save a lot of memory.
Generally speaking the optimal horizontal scaling is as little as you have to. You may want a bit of horizontal scaling for redundancy and geo distribution, but past that vertically scaling to fewer larger process tend to be more efficient, easier to load balance and a handful of other benefits.
> The main downside of this was memory usage. You would have to load all of your code and libraries N types and in-process caches would become less effective.
You can load modules and then fork child processes. Children will share memory with each other (if they need to modify any shared memory, they get copy-on-write pages allocated by the kernel) and you'll save quite a lot on memory.
Yes, this can help a lot, but it definitely isn't perfect. Especially since CPython uses reference counting it is likely that many pages get modified relatively quickly as they are accessed. Many other GC strategies are also pretty hostile to CoW memory (for example mark bits, moving, ...) Additionally this doesn't help for lazy loaded data and caches in code and libraries.
For big things the current way works fine. Having a separate container/deployment for celery, the web server, etc is nice so you can deploy and scale separately. Mostly it works fine, but there are of course some drawbacks. Like prometheus scraping of things then not able to run a web server in parallel etc is clunky to work around.
And for smaller projects it's such an annoyance. Having a simple project running, and having to muck around to get cron jobs, background/async tasks etc. to work in a nice way is one of the reasons I never reach for python in these instances. I hope removing the GIL makes it better, but also afraid it will expose a whole can of worms where lots of apps, tools and frameworks aren't written with this possibility in mind.
> observability tooling for Python evolving
As much as I dislike Java the language, this is somewhere where the difference between CPython and JVM languages (and probably BEAM too) is hugely stark. Want to know if garbage collection or memory allocation is a problem in your long running Python program? I hope you're ready to be disappointed and need to roll a lot of stuff yourself. On the JVM the tooling for all kinds of observability is immensely better. I'm not hopeful that the gap is really going to close.
> If true parallelism becomes common, it might actually reduce the number of containers/services needed for some workloads
Not by much. The cases where you can replace processes with threads and save memory are rather limited.
Citation needed? Tall tasks are standard practice to improve utilization and reduce hotspots by reducing load variance across tasks.
I would have thought most of those would have been moved to async Python by now.
async python still uses a single thread for the main loop, it just hides non blocking IO.
A lot of that has already been solved for by scaling workers to cores along with techniques like greenlets/eventlets that support concurrency without true multithreading to take better advantage of CPU capacity.
That's great for concurrency, but doesn't improve parallelism.
Unless you mean you have multiple worker processes (or GIL-free threads).
But you are still more or less limited to one CPU core per Python process. Yes, you can use that core more effectively, but you still can't scale up very effectively.
But python can fork itself and run multiple processes into one single container. Why would there be a need to run several containers to run several processes?
There's even the multiprocessing module in the stdlib to achieve this.
Threads are cheap, you can do N work simultaneously with N threads in one process, without serialization, IPC or process creation overhead.
With multiprocessing, processes are expensive and work hogs each process. You must serialize data twice for IPC, that's expensive and time consuming.
You shouldn't have to break out multiple processes, for example, to do some simple pure-Python math in parallel. It doesn't make sense to use multiple processes for something like that because the actual work you want to do will be overwhelmed by the IPC overhead.
There are also limitations, only some data can be sent to and from multiple processes. Not all of your objects can be serialized for IPC.
It makes sense to me that a program currently written using multiple processes would now be re-written to use multiple truly parallel threads. But it seems very odd to suggest (as your grandparent comment does) that a program currently run in multiple containers would likely be migrated to run on multiple threads.
In other words, I imagine anyone who cares about the overhead from serialization, IPC, or process creation would already be avoiding (as much as possible) using containers to scale in the first place.
I think you have a good point on IPC but process creation in Linux is almost as fast as thread creation
Unless the app would constantly be creating and killing processes then the process creation overhead would not be that much but IPC is killer
And also your types aren’t pickable or whatever and now you gotta change a lot of stuff to get it to work lol.
Forking and multi threading do not coexist. Even if one of your transitive dependencies decides to launch a thread that’s 99% idle, it becomes unsafe to fork.
Im curious as to the down votes on this. It's absolutely true, and when I was maintaining a job runner daemon that ran hundreds of thousands of who knows what Python tasks/jobs a day on some shared infra with arbitrary code for a certain megacorp from 2016-2020 or so, this was one of insidious and ugly failure modes to go debug and handle. The docs really make it sound like you can mix threading and multiprocessing but you can never really completely ensure that threading and then bare fork will ever be safe, period. It's really irritating that the docs would have you believe that this is OK or safe, but is in keeping with the Python philosophy of trying to hide the edge of the blade you're using until it's too late and you've cut the shit out of yourself.
Why is it unsafe?
In general only the thread calling fork() gets forked, so unless you call exec() soon after, there are a lot of complications with signals, shared memory.
What are the complications? A single thread with its own process sandbox with everything from the parent is exactly what I'd expect coming from C land. Are the complications you refer to specific to the python VM or more general?
I'm replying to a person that scales python by running several containers instead of 1 container with several python processes.
Fork-then-thread works, does it not?
If you have enough discipline to make sure you only create threads after all the forking is done, then sure. But having such discipline is harder than just forbidding fork or forbidding threads in your program. It turns a careful analysis of timing and causality into just banning a few functions.
Can't you check what threads are active at the time you fork?
But not the reverse, if its a bare fork and not strictly using basically mutex and shared resource free code (which is hard), and there's little or no warning lights to indicate that this is a terrible idea that fails in really unpredictable and hard to debug ways.