What is the next step in the project?

[oseasandrepro]

Now that the "Stable Foundations" was released, it is time to decide, will you continue with refactoring and add new features only for prototype/academic proposes?, or will you define a strategic niche and conduct the development to a level of maturity, so the Lib can be used in production scenario?

[oseasandrepro]

Things are getting serious! 🤔
What you gonna do?

[oseasandrepro]

I just decide to push the library to production level of maturity.

[oseasandrepro]

You should read the "Procedure Call Protocol Specification", because now since we gonna to produce production code, we need a more solid reference.

https://datatracker.ietf.org/doc/html/rfc1057#section-9

[oseasandrepro]

Tackle Problems:

1. Data Definition language - Python for Interface definition and Data definition language
2. Serialization across multiple languages
3. A compact protocol
4. Security (I decide to use TLS)

[oseasandrepro]

Ok, I was think about, This order can be changed to

1. Data Definition language - Python for Interface definition and Data definition language
2. A compact protocol - Framing-Octet-counting, look this On the Design of Application Protocols, is a excelente reference about design of Application Protocols
3. Security (I decide to use TLS)
4. (D)Serialization across multiple languages

[oseasandrepro]

Ok, now the use of TLS was implementend lets think about integration with multiple languages.

Does exist an important discussion about types here:

1. I give the responsability of (de)serialization do msgpack and my definition language does not have specific data types for differente int sizes. I am assuming that an int has 4bytes(32bits) - how do I make it? I must explore more types and type hints in Python. suggestion:

• Python Type Hints: Writing Clearer and More Reliable Code
• Use NewType (from typing import NewType)
• Advanced Type Annotations in Python: Part 2

2. msgpack does not suport (de)serialize of custom objects natively, so if I want de(serialize) a custom Class , now, I must implement it with "Extended types" of msgpack- An alternative for this is use tuple of tuples to represent complexy objects.

I decide to :

1. Implement integration with JavaScritpt
2. Benchmark tests
3. Discussions about performance

Why? the intention is to stablish the limits of the current implementation aproach, and think how can I make it better, so I can use this knowledge in the implementation of the others langues - in future releases.

After doing this I gonna stablish the future of the project 🤔 👐

The Benchmark tests metrics suggestion:

• Latency distribution: Track P95 and P99 response times, not just averages, to catch hidden delays.
• Throughput (RPS): Measure successful requests per second before saturation sets in.
• Resource utilization: Watch server CPU, memory consumption, and thread pool queues to spot resource bottlenecks

Possible combinations(Languages, server or client) of Benchmark test:

• Python-Server <--> Python-Client
• Python-Server <--> JS-Client
• JS-Server <--> JS-Client
• JS-Server <--> Python-Client

Tests with cross languages gonna be interstin becouse the raw socket Lib in JS is event oriented, so the stubs gonna be event driven, on the other hand
in Python I am using an procedural aproach 🥶

Another aspect is how to make scalabele integration tests cross languages? 🥶

[oseasandrepro]

I was thinking that add another language before benchmark test is not a good a ideia 🤔 , I will implement an aproach in more than one language before know the limits of currents aproach.

Ok, if this was an interprise project maibe the implementation in one target langue would be the right choice - thinking in pratical apect(eg: we need the custom RPC calls runing in the next week).
But, since since this in not my position I gonna move on another way: First make the benchmark test(only with Python) improve the arcthecture, learn the industry patterns of the implementation of RPC in the process.

I decide to :

1. Benchmark tests
2. Discussions about performance
3. Implement integration with JavaScritpt

[oseasandrepro]

After the Python benchmarks, I ran SRPC in Kubernetes through Tsuru. That experiment raised some questions about the next step 🤔

Exposing a custom TCP service required extra configuration, and using k6 was less straightforward than it would be for HTTP. I also saw much lower throughput when the server had limited CPU and memory, even with 256 threads. This does not tell me yet whether the main cost came from the thread model, resource limits, connection handling, or the test setup.

So I am updating the scope:

• Keep TCP and the current application protocol. Cloud deployment is not a requirement or a priority for SRPC. The Tsuru test was an experiment to learn about its operational costs.
• Focus on I/O-bound applications. I will evaluate the server against that workload, avoiding CPU-bound calls.
• Investigate the current Python implementation before changing its concurrency model. The next step is to repeat the benchmarks with explicit CPU and memory limits, measure latency, throughput and resource usage, and identify where requests start to queue.
• Decide on architectural changes from those results. An event loop with a thread pool is one option to investigate, but I do not yet have evidence that it will improve this workload.

I also run some experiments to use Generic socket server classes of Python, maybe I will use it, one vantage is code redution and better server life cycle.

This keeps the direction of my previous comment: understand the limits of the Python implementation, improve what the measurements justify, and then add another language. I will revisit JavaScript versus Java when the protocol and the target use case are clear enough to guide that choice. 👐