Top Tweets for #rxJava

about 19 hours ago

#hms #relationship #disruptor #allocator #oomkiller #rxjava I think I got closer why the "Relationship Disruptor" prevents or cancels any meaningful relationship formation/continuation of me, despite on paper, my stats and dynamically discoverable fetures are fine. ~~~~ Hint was the RxJava issue with groubBy excursions and the linux out-of-memory process killer plaguing GitHub many years ago. The main drawback with groupBy is that a new item may trigger a new group creation or feed one of the existing group, while they are being backpressured. There is no way to know upfront without introducing latencies and deadlocks. The solution I'm going to implement with RxJava 4 is a buffer for group creation excursions, default 4x of the main prefetch amount. So if the item->group dispatch is relatively balanced, progress can be made on the more frequent groups. When the buffer gets full, the operator switches to LRU group closing, meaning older unused groups get completed and their slot is reused for the new group. This of course requires losing the group reference identity and treat the group as a 'value stream' itself. As for the OOMKiller, it was partly systematically fixed by natively giving the CI/CD more memory, as well as we updated many tests to use less memory, the pipeline to use caching and other small test optimizations. ~~~~ What does it have to do with my relationship establishing issues? When I see a candidate, my mind starts doing heavy processing: simulations, evaluations, getting all past experience, data, skills loaded into working memory, which causes a huge spike in memory requirements by the ledger. Since the process is not very data efficient and mostly not async, it adds further burden on the conformal compactor during the process to save up on memory before more allocation is needed. This all spends huge amounts of entropy and risks process cancellation due to local bulk-boundary instability. In layman terms, I make some peculiar face while processing which is often off-putting and the time it takes to get some early results makes the initital meeting opportunity window gone, decohered of hard limited. So the solution is, optimization, zero-copy solutions, async loading, dynamic-adaptive loading, past history and experience compaction for quicker wibe processing at least. This is doable and a similar thing already happened many years ago with my local quantum search capability (Orch-OR collapse + NN processing improvements). I have to post the search query and it does it asynchronously with optional progress or partial result retrieval. It is very efficient and high quality usually. Think of it like posting on X and the X algorithm routing interesting topics back to me via 'For You' without me needing to refresh every 0.5 seconds. Wir werden schaffen das! 🫡

Paradigma Digital

@paradigmate

29 days ago

¿Sientes que el mundo secuencial ya no es suficiente? Te guiamos para que des el salto a #RxJava sin dolores de cabeza. Aprende a manejar flujos, errores y concurrencia con esta guía práctica 👇

paradigmate's tweet photo. ¿Sientes que el mundo secuencial ya no es suficiente? Te guiamos para que des el salto a #RxJava sin dolores de cabeza. Aprende a manejar flujos, errores y concurrencia con esta guía práctica 👇

David Karnok

@akarnokd

2 months ago

#java #rxjava #language Turns out, we already have a clumsy but working way for a few language features I unkindly requested: - auto Varhandles ~ AtomicXXX classes - async ~ create() with an emitter.emit() call for yield returns - await ~ await(CompletionStage [, DisposableContainer]) Universal Turing Complete von Neuman computers rule! Given a 10 foot pile of sh|t, a programmer can build an 11 foot high ladder and step over it. ~ Cory Doctorow, Ensh|ttification 🫡🥚✝️

243

David Karnok

@akarnokd

3 months ago

Wow. Interesting thesis. https://t.co/PtLZ7J5YNu If you can, please redo it with #RxJava 4's Schedulers.virtual() instead of wasting 1,000+ cached threads on blocking IO at once. Also it is not explicitly mentioned by the document, but Structured Concurrency won't save you much. Coordination is one of the hardest part of async programming and you'll sweat blood getting it to work for non-trivial and/or multi-valued use cases. I know, because I did sweat. Worry not. We'll fix that with Streamable<T> / Streamer<T> just fine. 😏Stay tuned!

David Karnok

@akarnokd

3 months ago

Now that #RxJava 4 is brewing, expect job listings requiring 5 years of experience with it. So if you did Rx 4 already in 2021, you'll be golden. 😆

451

David Karnok

@akarnokd

3 months ago

#java #idea #rxjava Here is my list of Java language enhancements that would make our RxJava life much much easier: - Extension methods. - Generators á la C#, yield return X and yield return async X. - Syntactic sugar async/await on top of Virtual Threads - Syntactic sugar atomics, atomic_read(x), atomic_cmpxchg(x), etc. - Obtaining names via nameof(method|param|variable) at compile time. These all seem to be doable, and anomalies worked around just fine. Please do it. Please. 🥺🙏

David Karnok

@akarnokd

3 months ago

#rxjava #proposal I've just posted a new hybrid reactive-virtual type Streamable<T> and its provider Streamer<T>. It is a multi-valued sequence that complements the executor.submit() type one-shot calls with Virtual Threads. It is going to be modelled after C#'s IAsyncEnumerable<T> type. Virtual threads make it very convenient to block on join() calls. However, I'd rather have syntactic sugar support in Java: await someCompletionStage;. Any state machine builder for yield return and yield return async would be also welcome. https://t.co/SM2U7nHkby

476

David Karnok

@akarnokd

3 months ago

#scifi #rxjava I've been thinking about why 🖖 Star Trek Discovery was interesting and odd at the same time. Michael's hyperactive inference made it unstable-to-mundane to my eyes. However, when she was contrasted with a great captain such as Lorca, Pike and Rayner, the duo shined. Add in a McCoy-esque unity of Booker, Tilly, Stamets, Saru, et all, and now we have a stable 3-body system. Season 5 was good, but not perfect. I think the issue was that Rayner became number-1. He should have been the captain of the situation, making Michael his number-1 instead. The show needed a seasoned leader who can filter through Michael's and the others' ideas through his experienced lense and make the decision. Not Michael. Rayner not knowing or being capable of all the inference was not a bug. It was a feature. 🤔 Which brings us to RxJava. I have many ideas and resolves to make it great again. But I'm too technical, driven, loopy. The best years were when Ben Christensen was at the helm. He is an industry veteran who could filter through the ideas and resolves in a reliable, safe and measured way. Our McCoy was probably the engaged community members who also provided contrast and feedback to the library. I miss those days. So what does this mean with regards to RxJava 4? Technically doable, all doable. But we'd end up with a huge library if we do the full reactive triad / tetrahedron. We need a filter. I need a co-operator, maybe even two. To be clear, I'm not stepping down or release my admin powers. I invite qualified and established people like Ben & others to show up and get involved. We can go slow, no need to do everything within 10 days. I can, but won't. But I could 😉 🙏 for listening.

769

Top Tweets for #rxJava

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users