C++ Actor Framework VS moodycamel

CAF

I'd like to mention the native actor model implementation CAF, the C++ Actor Framework, and share some experiences. (Disclaimer: I've been developing on CAF in the past and have a good relationship with the creator.) CAF (1) provides native actors without an VM layer, (2) type-safe interfaces so that the compiler yells at you when a receiver cannot handle a message, and (3) transparent copy-on-write messaging so that you can still push stuff through pipelines and induce only copies only when a ref count is greater than one.

In our telemetry engine VAST, we've been using CAF successfully for several years for building a distributed system that always has a saturated write path. CAF provides a credit-based streaming abstraction as well, so that you can have backpressure across a chain of actors, making burst-induced OOM issues a blast from the past. You also get all the other benefits of actors, like linking and monitoring, to achieve well-defined failure semantics: either be up and running or collectively fail, but still allowing for local recovery—except for segfaults, this is where "native" has a disadvantage over VM-based actor models.

With CAF's network transparent runtime, a message ender doesn't need to know where receiver lives; the runtime either passes the message as COW pointer to the receiver or serializes it transparently. Other actor model runtimes support that as well, but I'm mentioning it because our experience showed that this is great value: we can can slice and dice our actors based on the deployment target, e.g., execute the application in one single process (e.g., for a beefy box) or wrap actors into single OS processes (e.g., when deploying on container auto-scalers).

The deep integration with the C++ type system allowed us to define very stable RPC-like interfaces. We're currently designing a pub/sub layer as alternate access path, because users are interested in tapping into streaming feeds selectively. This is not easy, because request-response and pub/sub are two ends of a spectrum, but it turns out we can support nicely with CAF.

Resources:

- CAF: https://github.com/actor-framework/actor-framework

- VAST: https://tenzir.github.io/vast/docs/understand-vast/actor-mod... (sorry for the incompleteness, we're in migration mode from the old docs, but this page is summarizing the benefits of CAF for us best)

- Good general actor model background: http://dist-prog-book.com/chapter/3/message-passing.html#why...

VAST is a flight recorder and security content execution engine. On the one hand, there exists a continuous stream of high-volume data sources (such as network telemetry as NetFlow, Zeek, Suricata, and endpoint telemetry). On the other hand, VAST processes needle-in-haystack queries to provide answers to questions like "has this threat been relevant to us 8 months ago?", and supports threat hunters with an interactive query capability to explore the data. From an engineering perspective, we focus especially on the separation of read and write path, concurrent message passing in an actor model runtime (CAF), and leveraging open standards, like Apache Arrow, to establish a high-bandwidth data plane for sharing data with downstream tooling. A flexible plugin API enables additional security-specific use cases on top, such as realtime matching of threat intelligence or mining of asset data for passive inventorization.

Technologies: Apache Arrow, Flatbuffers, C++ Actor Framework, Linux, Docker, Kubernetes

Other than that... if your stuff is good, that's a much better signal than a professional username. I've seen a lot of decently unprofessional usernames out there that get taken pretty seriously because of the good work behind them. My recent favorite is "moodycamel" who authored a great concurrent queue library in C++.

In c++ the moodycamel ConcurrentQueue is a good choice.

Here's a decent concurrent queue: moodycamel::ConcurrentQueue.

Lock free programming is hard, and probably harder than you think. I would not even try something like that myself. I would look for existing solutions, something like https://github.com/cameron314/concurrentqueue for example.

I needed a concurrent queue that would block when attempting to pop an empty queue, which allows the consuming thread to suspend while it's waiting for work. I found that using mutexes allowed me to develop a simple template adapter had several advantages with few drawbacks when compared to non-blocking queues: it can use a variety of containers, the code can be reviewed and verified as to its correctness (very hard to do with fancy concurrent programming that avoids mutexes), and it is only slightly slower than fancier solutions (when I benchmarked it originally, it was 4x slower than Moody Camel's concurrent queue, which to me is fine performance).

Not literatue but an example. This is a lock-free (not wait-free!) multi-producer multi-consumer queue, not a FIFO, but access patterns should be similar - if not the same: https://github.com/cameron314/concurrentqueue

If I do implement it, the most likely route I'd take is make a compiler in Clojure/clojurescript that uses Instaparse (I have a more-or-less-clojure grammar written that I was tinkering with) and generate C++ code that uses Immer for its data structures and Zug for transducers and what my not-quite-clojure would support would be heavily dependent on what the C++ code and libraries I use can do. I'd use Taskflow to implement a core.async style system (not sure how to implement channels, maybe this but I'm unsure if its a good fit, but I also haven't looked). I would ultimately want to be able to interact with C++ code, so having some way to call C++ classes (even templated ones) would be a must. I'm unsure if I would just copy (and extend as needed) Clojure's host interop functionality or not. I had toyed with the idea that you can define the native types (including templates) as part of the type annotations and then the user-level code basically just looks like a normal function. But I didn't take it very far yet, haven't had the time. The reason I'd take this approach is that I'm writing a good bit of C++ again and I'd love to do that in this not-quite-clojure language, if I did make it. A bunch of languages, like Haxe and Nim compile to C or C++, so I think its a perfectly reasonable approach, and if interop works well enough, then just like Clojure was able to leverage the Java ecosystem, not-quite-clojure could be bootstrapped by leveraging the C++ ecosystem. But its mostly just a vague dream right now.

I would recommend https://github.com/cameron314/concurrentqueue as it's very battle tested and fast.

This was explicitly considered for spdlog (using the moodycamel::ConcurrentQueue) but rejected for the above reason. I'm not involved in the development of spdlog but personally I agree, for me it's important that log output is not all mixed up.