Distributed coroutines with a native Python extension and Dispatch
stealthrocket.tech
stealthrocket.tech
That being said, I like the idea and the blog post is wonderfully written.
Serializing file handles doesn't work, but in our experience, programs rarely run into constructs where this becomes a problem, and when it happens, there are mitigation measures that are usually easy to implement (small restructure of the program, capturing resource metadata to reconstruct them later, etc...).
We have a few features on the roadmap to help mitigate the security implications as well, including allowing users to store their program state in a S3 bucket that they own. Our scheduler can operate with only metadata about the program, so splitting the two can be an effective model to mitigate those risks.
Your original machine might not have the required computational resources to run the job quickly enough.
> Shouldn't it be the job of the load balancer, to choose the least busy machine, before any coroutine is run?
It is not always simple or possible to know, when what part of a computation will be finished and as such cannot easily be perfectly planned. If you mean a load balancer as in traefik or similar, then that entails serializing your intermediate results or writing your code in a specific way, so that computation can be split using a load balancer.
> Isn't it expensive to serialize coroutines and transfer them between machines?
Probably, but not taking advantage of idle cores on another machine might be more expensive (in terms of time needed to finish the computation).
Also, I wonder what observability is like. If a coroutine crashes, what its stacktrace will look like?
No idea, have not used it.
This kind of thing is what Erlang excels at. Serializing functions and their entire environments is a difficult to solve problem. I think in Python it probably means moving into a whole different space of types and objects, because Python has not been developed with such a thing in mind from the start, while Erlang has.
If there are idle cores on another machine, why not just point the load balancer to schedule new jobs/coroutines on those idle machines? Why reschedule existing coroutines on another machine if they run just fine on the current node? (i.e. the node is at full capacity, and it's fine) I can see that the problem can arise if a coroutine wants to spawn new coroutines, and if we schedule them on the same node which is at full capacity, they can end up waiting for a while before they are run... It makes sense to schedule new jobs on different machines, but why reschedule existing coroutines? You have to serialize/deserialize stuff and send over the wire, and there are several gotchas as explained in the article (pickling arbitrary objects doesn't sound very reliable/safe, and judging from the article, can break in future versions of Python), plus there's a lot of magic involved which will probably make it harder to investigate things in production... I'd personally just stick with local coroutine sheduler + global load balancer which picks nodes for new jobs + coroutines, when created, only receive plain values as arguments (string, int, float and basic structures) to be reliably serializable and so that it was transparent what's going on... (i.e. do NOT store internal coroutine state, assume they are transient). Maybe I don't understand the idea.
A load balancer can help distribute new jobs across a fleet, but even the shortest of jobs can become "long running" when it hits timeouts, rate limits, and other transient errors. You quickly need a scheduler to effectively orchestrate the retries without DDoS-ing your systems, and need to keep track of the state to carry jobs to completion.
Combine a scheduler (like Dispatch) with a primitive like distributed coroutines, and you've got a powerful foundation to create distributed applications of all kinds without seeing complexity skyrocket.
I can see the benefit that with Dispatch, logic is simpler to read/to write as just ordinary functions, while in our approach, we have to scatter it around various event handlers/job processors. However, I still like that in our approach, event handlers/job processors are entirely stateless (the only state is jobs/event payloads), I've found it to be good for scalability and reliability + easier to reason about, compared to passing around internal coroutine state.
Dispatch can also handle the "one-off" jobs you describe, where you don't need to track the coroutine state. In a way, it's a subset/special case of the distributed coroutine (just like functions are a special case of coroutines with no yield point).
Our take is that distributed coroutines can be brought to general-purpose programming languages like Python so that engineering teams can adopt these features incrementally in their existing applications instead of adding a whole new language or framework.
In my opinion, the value is in being able to reuse the software tools and processes you're familiar with, major shifts are rarely the right call.
Say your coroutine is performing a transactional operation with multiple steps, but for some reason the second step starts getting rate limits from an upstream API, you need to enter a retry loop to reschedule the operation a bit later. What if your program stops then? Without capturing the state somewhere, it remains volatile in memory and the operation would be lost when the program stops.
This scenario occurs in a wide range of systems, from simple async jobs for email delivery to complex workflows with fan-out and fan-in.
> Isn't it expensive to serialize coroutines and transfer them between machines?
We only capture the local scope of the call stack, so the amount of data that we need to record is pretty small and stays in the order of magnitude of the data being processed (e.g., the request payload).
> I wonder what observability is like. If a coroutine crashes, what its stacktrace will look like?
We maintain stack traces across await points of coroutine calls, even if the concurrent operations are scheduled on different program instances. When an error occurs, the stack traces show the entire code path that led to the error.
There's definitely a lot of opportunities for creating powerful introspection tools, but since Dispatch is just code, you can also use your typical metrics, logs, and traces to instrument your distributed coroutines.
What if the local stack references a deep object hierarchy? The classical "I wanted a banana, but got the gorilla holding the banana and the entire jungle"
>What if your program stops then? Without capturing the state somewhere, it remains volatile in memory and the operation would be lost when the program stops.
You persist coroutines on disk?
This is why these solutions have often seen the creation of a new programming language as a necessary step to avoid those issues altogether. We think it's important to bring those capabilities to mainstream languages, we can't always go back from scratch with a new language.
Regarding where the state is stored, we keep everything in S3 and have plans to allow users to provide their own object store if they need to keep the application state in a storage medium that they own.
Also happy to answer any questions :)
With Dispatch, you just write code, as if you would write a local program, and our scheduler can manage the distribution and state of the execution. Testing the code especially gets really simple.
Having a hosted scheduler also means that you don't need to deploy any extra infrastructure like RabbitMQ or Redis.
A good starting point to answer your other questions is the "getting started" section in our documentation https://docs.dispatch.run/dispatch/getting-started
If you have more questions, join our Discord! We're always happy to chat and get feedback on what's difficult to grasp.
I took a gander at the source I'll wait till/if there's ever a 100% local version of this. But good luck all the same!
Take a look at the "Local Testing" section in the README of dispatch-py https://github.com/stealthrocket/dispatch-py?tab=readme-ov-f...
TLDR: python -m dispatch.test http://127.0.0.1:8000
This is just a mock implementation but it lets you test your code and experiment with the solution.
People have often tried to use Temporal to solve similar problems. We think Dispatch is a very different solution because we don’t require you to architect your system in terms of a rigid framework of activities and workflows... You just write regular code with simple functions, enabling incremental adoption into existing applications and making the code easy to test.
A friend of mine once described Dispatch as "Temporal without the galaxy brain", it's a pretty good way of phrasing it in my opinion :)