Ask HN: Are systems ready for the first negative leap second?

Kafka's log compaction corrupts data. Here's how we fixed it

Apache Kafka có lỗ hổng trong cơ chế log compaction khiến dữ liệu bị hỏng do xung đột giữa compaction và replication, gây ra bốn vấn đề: dữ liệu đã xóa tái xuất hiện, giao dịch bị hủy hiện dưới dạng đã commit, dữ liệu đã commit bị ẩn, và consumers read_committed bị đóng băng partition. Redpanda Streaming khắc phục bằng giao thức compaction phối hợp, sử dụng các cặp offset (MCCO/MTRO, MXFO/MXRO) để đảm bảo tombstones và transaction markers không bị xóa trước khi tất cả replicas xử lý xong. Lỗi này có thể tái hiện trên Kafka phiên bản 3.9 đến 4.2 bằng Docker Compose.

Lập trình viên cần đọc bài này để hiểu cách giải quyết vấn đề lỗi race condition trong log compaction của Kafka, giúp tránh mất dữ liệu và bảo đảm tính nhất quán khi xử lý các trường hợp đồng bộ hóa dữ liệu trên nhiều broker.

How to use traces to avoid breaking changes

Theo dõi (tracing) giúp phát hiện sớm vấn đề khi thay đổi hệ thống bằng cách ghi lại luồng dữ liệu trong môi trường phân tán, sử dụng các công cụ như OpenTracing, OpenTelemetry, Zipkin hay Jaeger. Nó bao gồm các thành phần như tracer, span, propagation context, correlation ID và các công cụ phân tích trực quan.

Lập trình viên nên đọc bài này để hiểu cách sử dụng tracing để phát hiện và tránh giải phóng không mong muốn trong các ứng dụng phân tán, giúp bảo đảm chất lượng và ổn định khi triển khai thay đổi.

“This Can’t Possibly Work”: What I Learned at a Temporal.io Workshop

A workshop recap from We Are Developers World Congress in Berlin covering Temporal.io's durable execution model. The session walked developers through the core problem: distributed applications crash mid-execution, leaving state inconsistent. Temporal solves this via history replay — recording workflow events so execution can resume on any machine after a crash, with variables intact and no duplicate work. The architecture splits logic into deterministic Workflows (orchestration) and Activities (side effects like API calls), with Workers replaying event history to reconstruct state. The Temporal Web UI provides full observability into every execution. The key insight: durable execution eliminates entire categories of infrastructure like cron jobs, message queues, and retry scaffolding, letting developers write clean happy-path code.

Trade-Offs in Multi-Region Architectures: Latency vs. Cost

Adding cloud regions is rarely as simple as a cost-vs-latency calculation. A framework built from multiple real-world region launches shows that roughly 40% of new region costs come from cross-region network fabric, replication, and operational overhead — often underestimated. Before expanding, decompose your latency budget: in one case study, only 45% of 260ms latency was geographic; routing optimizations and dependency elimination cut it 35% with no new infrastructure. Three deployment patterns are compared (active-active, read-local/write-global, active-passive) with trade-offs across read/write latency, consistency, cost, and DR. Key principles: eliminate cross-region dependencies in critical paths, rightsize regional footprints, invest in automation before scaling, and treat expansion as a phased optimization rather than a one-time build. A real-world phased approach achieved sub-60ms latency while revealing unexpected replication cost overruns from metadata fanout and retry feedback loops.

Formal Methods for Every Engineer in an AI-Powered Future

Gabriela Moreira, creator of the Quint specification language, discusses how formal methods are becoming practical for everyday software engineers in 2026. AI now dramatically lowers the barrier to entry by generating formal specs (Quint or TLA+) and the tedious glue code needed for model-based testing. The conversation covers how formal specifications differ from traditional testing, how to keep models and codebases in sync via model-based testing and trace validation, and why defining correct system behavior remains irreplaceable human work. A real AWS DNS outage case illustrates the cost of letting code drift from its specification.

GitHub - harmont-dev/hyper: Distributed Firecracker VM Orchestrator written in Elixir

Hyper is an open-source distributed orchestrator for Firecracker microVMs written in Elixir, targeting the same niche as Daytona and Runloop. Key features include affinity-based VM scheduling, copy-on-write disk layering for efficient VM forking, OpenTelemetry instrumentation, BEAM-native fault tolerance, and a gRPC interface for polyglot integration. It requires only a Postgres database as an external dependency and can be embedded as a Mix dependency. The author built it after being dissatisfied with existing products — citing Daytona's unreliable VM features, Freestyle's reliability issues, and Runloop's lack of forking support and high cost. Hyper is early-stage and under active development, with interfaces subject to change.

Redpanda vs Kafka vs Confluent: An Honest Comparison

A detailed comparison of Apache Kafka, Confluent, and Redpanda across performance, operational complexity, ecosystem maturity, and enterprise readiness. Redpanda's C++/Seastar thread-per-core architecture eliminates JVM GC pauses and delivers lower p99/p99.9 tail latencies, especially on NVMe hardware, and ships as a single binary versus Kafka's componentized stack. However, Kafka and Confluent win on ecosystem depth (120+ managed connectors, native Flink integration, decade of community tooling) and enterprise SLAs. The post also covers the emerging 'diskless Kafka' trend via KIP-1150, AutoMQ, and WarpStream (now Confluent-owned), noting that object-storage-native architectures may reshape the cost equation more than broker-level efficiency gains. Redpanda's BSL 1.1 license is flagged as a consideration for procurement teams. The recommendation is to run proof-of-concept benchmarks against actual production traffic rather than relying on vendor-published numbers.