Bilingual

Hanami 3.0: In Full Bloom
4 days ago
- Hanami 3.0 introduces major features: first-class mailers with integrated email delivery and templates, built-in internationalization (i18n) for translations and localization, and Minitest support alongside existing RSpec.
- Performance improvements include default component memoization, reducing allocations by up to 80% and increasing throughput by nearly 3x, plus optimizations in Hanami Action and View.
- Enhancements to developer experience: colorized logs, improved asset watching, body parsing integrated into Hanami Action, undecorated view exposures by default, and a cleaner foundation with renamed gems.
- Additional refinements: new generators (provider, mailer), options for app generation (--test, --template-engine), updated dependencies (Ruby 3.3+, Puma 7.1), and various fixes for routing and templates.
- Easy setup with commands to install and run Hanami 3.0, plus upgrade notes for existing 2.3 users, supported by contributions from the community and sponsors.
Database Traffic Control
5 days ago
- Postgres lacks built-in traffic management, making it vulnerable to performance degradation or crashes during query spikes.
- PlanetScale introduces Database Traffic Control™, enabling real-time resource budgeting and limits for queries via rules like query patterns, application names, users, and custom tags.
- Traffic Control allows setting limits on CPU, concurrency, and query timing, with modes for warning or enforcement to throttle or block excessive queries.
- Use cases include incident response for rogue queries, priority-based traffic shaping, isolating AI agent traffic, and prioritizing paid tiers in multi-tenant apps.
- Available via PlanetScale dashboard, API, and CLI, it supports automated budget creation and sqlcommenter tags for enhanced traffic categorization.
NUMA: Cores, memory, and the distance between them
11 days ago
- NUMA (Non-Uniform Memory Access) means memory access costs vary depending on CPU and memory bank locations, unlike UMA (Uniform Memory Access).
- NUMA emerged to scale beyond single-socket limits, with each socket having its own memory controller and interconnects like AMD's HyperTransport/Infinity Fabric and Intel's QPI/UPI.
- Modern servers can have multiple NUMA nodes per socket due to designs like AMD's EPYC with chiplets and Intel's Sub-NUMA Clustering, moving beyond the 'one socket, one node' model.
- Remote memory access on modern servers is 1.5x to 3x slower than local access in microbenchmarks, worsening under load due to interconnect contention, affecting tail latency.
- Memory interleaving spreads memory across nodes to flatten costs but sacrifices peak performance for predictability, making 'everything equally bad'.
- NUMA involves two affinities: CPU affinity (pinning processes to CPUs) and memory affinity (controlling memory allocation per node), which must align for optimal performance.
- Linux's 'first touch' policy allocates memory on the node of the CPU that first accesses it, which can cause remote accesses if allocation and usage threads are on different nodes.
- Xen's dom0 lacks NUMA awareness, leading to mismatches between guest vCPUs and memory placement, silently causing cross-interconnect access and performance issues.
- Virtualized systems like Xen have two stacks (hypervisor and guest) making separate placement decisions, unlike KVM where the host Linux kernel manages NUMA for VMs.
- Edera's work makes Xen NUMA-aware end-to-end, allowing guests to see accurate topology and optimize placement, avoiding the 'blind and interleave' dilemma.
Superpowers 6
5 days ago
- Superpowers 6 offers significant performance improvements: builds are up to 50% faster and up to 60% cheaper in token costs.
- Key optimizations include merging spec compliance and code quality review agents, pre-generating review packets to reduce git commands, and improving task-specific agent guidance.
- The improvements were validated through an autoresearch loop with 25 experiments, identifying effective strategies like terse reviewer contracts and conditional implementer tiering.
- The evals suite at https://github.com/prime-radiant-inc/superpowers-evals was crucial for testing changes across various coding agents and harnesses.
- Superpowers 6 is available for installation from https://github.com/obra/superpowers, with updates to plugin marketplaces following soon.
- Prime Radiant is hiring for a Superpowers Community Engineer role at https://primeradiant.com/jobs/superpowers-community-engineer/.
CachyOS June 2026 Release
6 days ago
- CachyOS June 2026 release introduces Hyprland Noctalia desktop, DNS-over-QUIC support, and performance improvements.
- Package updates include Python extended PGO for better performance, GCC branch misprediction tuning for Intel/AMD CPUs, and pacman network isolation.
- Installer now features Hyprland Noctalia with a preview video, removed paru in favor of Shelly, and improved hardware/keyboard detection.
- CachyOS-Welcome adds DNS-over-QUIC via blocky, a troubleshooting page, terminal support, and new localizations including Azerbaijani and Greek.
- Fixes cover installer handling of keyboard layouts, pacman config, cleanup steps, and crashes in CachyOS-Welcome when cachyos-pi is missing.
- Manual updates not required; users can update with sudo pacman -Syu and download ISOs from provided mirrors for desktop and handheld editions.
How we made WINDOW JOIN parallel and vectorized
10 days ago
- QuestDB introduced a dedicated WINDOW JOIN operator to efficiently aggregate one table over a time window around each row of another, simplifying complex SQL queries that previously required multiple joins and unions.
- The operator achieves high performance through data-level parallelism, where the left-hand side (LHS) table is split into page frames processed by worker threads, and a low-cardinality fast path that enables SIMD vectorized aggregation.
- Key optimizations include building per-key indexes and contiguous value buffers for RHS data, allowing binary searches for window bounds and efficient SIMD kernels for aggregates like sum, avg, min, and max.
- Benchmarks against Timescale, DuckDB, and ClickHouse on a 50M-row trades and 150M-row prices dataset showed QuestDB's parallel + SIMD implementation running 5x faster than its single-threaded version and 25x faster than ClickHouse's best rewrite.
- The WINDOW JOIN operator supports various configurations, including one-sided windows, include/exclude prevailing rows, and multiple aggregate functions, with plans to expand temporal operators like HORIZON JOIN in future releases.
JPEG-XL Libjxl 0.12 Brings More Performance Optimizations
2 days ago
- libjxl 0.12 release improves JPEG-XL image format performance and features.
- Includes new jxltran tool for extracting codestreams from JXLP boxes.
- Progressive lossless images are 30~40% smaller and encode 2-5x faster with multi-threading.
- Lossless images decode faster and are 30~80% smaller overall.
- SIMD optimizations boost encoding, with EstimateCost SIMDified for 5% improvement.
- Modular encoding/decoding up to 4x faster via SIMD, and histogram encoding/decoding overhauled.

Hasty Briefsbeta

#performance optimization

Hanami 3.0: In Full Bloom

Database Traffic Control

NUMA: Cores, memory, and the distance between them

Superpowers 6

CachyOS June 2026 Release

How we made WINDOW JOIN parallel and vectorized

JPEG-XL Libjxl 0.12 Brings More Performance Optimizations