It looks like I came up with new leaderless consensus protocol like EPaxos, a post about it: "Pacified consensus: how to retrofit leaderlessness into a Paxos or Raft based protocol" -

Well, always has been.. Grep is essential for coding — AI or not

Now it's paying off again with SwiftUI, which feels a lot like React

Surprising how wrestling with JavaScript years ago helped me onboard with modern C++ & Seastar (cooperative concurrency within core-pinned threads) to develop Redpanda, a distributed streaming platform

Took me 4 months to get proficient in SwiftUI & iPhone development

I delayed implementing persistence, which made me repeatedly enter data. This exposed every UX flaw until I couldn't ignore them anymore. A great way to overcome initial blindness to your own errors!

Started iOS programming to build a Strava-like app for surfers and stumbled on a great life hack.

One idea for a high perf web framework would be to adopt a native pipes and filters architecture without controllers or concepts like that on top. Each pipe or filter gets to pre-inspect the req before processing to demand a contiguous memory arena for the response…

I guess software complexity is often hidden beneath the surface, but with physical objects, you can’t ignore it when it’s right in front of you. This forces management to prioritize it sooner

Funemployment means fun blog posts! Please enjoy a categorization of replication algorithms by failure handling strategies; their optimizations, tradeoffs, and complexities; and how this all affects their resource efficiency, latency, and availability. https://t.co/yGh75oMsgX

Just wrote an article with quick recap of the first @localfirstconf. Like that “X event in 15 minutes” videos, but in text. For me, it was the best conference this year and the whole Local-First are is right now the coolest webdev topic. https://t.co/cflNeCKQGU

🚨BREAKING NEWS: #Redpanda has raised $100M in Series C funding, led by @LightspeedVP, @GVteam, and @HaystackVC! Our team is dedicated to enabling #data-intensive apps with new foundational capabilities beyond the scope of legacy #streamingdata platforms. https://t.co/xnTHT8E23J

@rystsov Hey Denise, great article! You might also find this paper interesting: https://t.co/XmR3u2Sf89 They talk about how local file-system faults often lead to global, cluster-wide negative consequences!

If you like working on distributed systems and think through hard and nuanced problems like 👇, join us!

In my latest blog post, I demonstrate that this is a fundamental finding that applies to all non-Byzantine replication protocols. Furthermore, I provide a compelling example illustrating how data loss on a single node can trigger the loss of data in the whole Kafka cluster.

The argument mentioned above is a misconception. Even the loss of power on a single node, resulting in the local loss of unsynchronized data, can lead to silent global data loss in a replicated system that does not employ fsync, regardless of the replication protocol utilized 😮

Of course, it's important to acknowledge that data loss can still occur if all nodes experience power outages or simultaneous OS crashes. Nevertheless, there are measures we can take to minimize the likelihood of such events, such as leveraging availability zones.

Node crashes can lead to data loss if the application doesn't utilize fsync. However, there are replication protocols designed to withstand node crashes while maintaining data consistency. This suggests that replication can render fsync unnecessary for replicated systems.

📢 Just published a blog post discussing the ramifications of running a replicated system without fsync. Contrary to popular belief, the loss of data on a single node can actually result in the loss of replicated data on all nodes. 🔄 https://t.co/Fai0KKU38p

Around 70% of traffic in Azure is RDMA. Azure Storage is powered by RDMA between clients and frontends and within the clusters.