Moonglade

Modern Markdown blogging with AI-powered localization and search.

All Posts

Recent Featured
14
SEP

Understanding DHT: Welcome to Onboarding at a Microsoft Without a Central Directory You've joined a parallel universe...

This article uses the analogy of "Cyber Microsoft" to explain, in a clear and accessible manner, how Distributed Hash Tables (DHT) work. It points out that the core of the Kademlia algorithm lies in nodes maintaining hierarchical contact lists according to specific rules, ensuring that queries can progressively narrow down the scope and locate the target. This mechanism resolves the "false valley" dead-end problem that may exist in traditional networks. The article clarifies that DHT is solely responsible for routing based on precise key-value pairs, rather than semantic search. It also distinguishes between the mathematical uniqueness of the nearest neighbor and the lookup uncertainty caused by missing contact entries in real-world networks. Furthermore, the article reveals how DHT achieves decentralized key-value storage: publishers and queriers converge in a specific address region using the same resource hash value, without the need for a central server. Finally, the article emphasizes that DHT is a decentralized, peer-to-peer network structure that relies on all nodes collectively maintaining local knowledge to achieve global navigation, rather than depending on a single super node or arbitrary connections.

Views 63
11
AUG

Level 3 Classified Protection: A Compliance Stage Built by Everyone Without Malice, Yet Together

The author provides an in-depth analysis of how the Level 3 Classified Protection (Dengbao) standard has devolved into formalism in real-world implementation. Initially, the author questioned the "buy equipment to achieve compliance" model, arguing that security should be a systemic engineering capability rather than a mere accumulation of hardware. Subsequently, through a shift in perspective, the author realized that Dengbao was not designed for highly mature enterprises but rather to lift a large number of organizations with weak security foundations from low scores to a passing grade, serving a mandatory social baseline function. However, influenced by Goodhart's Law, once regulatory metrics are quantified for assessment, they mutate into purchasable, acceptable compliance performances, leading to the absurd reality of "buying security." The author points out that truly effective security measures (such as Zero Trust) are difficult to standardize and regulate, while easily regulatable metrics are prone to being reduced to KPIs. Ultimately, the author concludes that although Level 3 Dengbao is rigid and outdated, it is indispensable as a "60-point system" for raising the overall security baseline. The true absurdity does not lie in the standard itself, but in society mistaking "passing the exam" for "true security," even compelling high-level organizations to purchase inefficient equipment to prove compliance. This reveals the inherent limitations and costs of managing complex, dynamic security systems using administrative metrics.

Views 277
Compliance Cybersecurity InfoSec Culture Level 3 Protection MLPS MLPS Level 3
27
MAY

Set Up DGX Spark with Qwen 3.6 27B Dense

This article details the complete process of deploying the Qwen 3.6 27B BF16 model on an NVIDIA DGX Spark. The hardware utilizes the Grace-Blackwell architecture with 128GB of unified memory. By disabling the desktop environment, sleep mechanisms, and hardening SSH security configurations, the setup ensures maximum resource utilization and system stability. The deployment architecture uses vLLM to provide an OpenAI-compatible API, with OllamaGateway handling authentication and model alias management. Key configurations include full BF16 precision, 95% memory utilization, a 256K context window, and prefix caching, balancing quality and performance. Tests show that, limited by memory bandwidth, inference speed is approximately 5.5-16 tok/s, but it supports switching thinking modes and tool calls per request. The system uses systemd for automatic startup and fault recovery, with a cold start time of about 150 seconds. The overall solution involves no quantization and no external dependencies, with sufficient memory headroom, making it suitable for high-reliability local LLM service deployment within a local area network.

Views 519
BF16 Blackwell DGX Spark LLM Deployment Qwen3.6 vLLM
11
APR

Ultimate Fix: Rudder (Yaw) Axis Conflict for X-Plane 12 on Linux

This article addresses the issue of the rudder failing to return to center when using an Xbox controller in X-Plane 12 on Linux, and proposes a Python-based solution. Due to the long-term lack of updates and compatibility defects in the traditional `xboxdrv` driver, the article introduces a bridging method that uses the `evdev` library to intercept LT/RT signals and create a persistent virtual joystick device. The core logic involves merging the left and right trigger keys into a single axis and mapping the difference to a 16-bit high-precision range, thereby achieving perfect compatibility with advanced add-on aircraft such as the ZIBO 737. The implementation plan includes writing a bridging script for automatic controller detection, configuring a Systemd service for auto-start on boot and reconnection after disconnection, and setting up udev rules to resolve permission issues. Ultimately, users can bind this virtual device in X-Plane to achieve stable control, with calibration information persistently saved, eliminating the need for reconfiguration after every restart, significantly enhancing the flight simulation experience on the Linux platform.

Views 284
Anduin Linux Python X Plane 12 Xbox Series X evdev
09
MAR

Highly Reliable Data Disaster Recovery Architecture: Asynchronous Snapshot Backup Solution for Heterogeneous File Systems

This article addresses the data disaster recovery requirements in high-performance computing environments and proposes a "versatile" backup solution that balances IO efficiency with historical version rollback. It points out that traditional `rsync` synchronization carries the risk of accidental deletion propagation, while full snapshot backups incur massive resource overhead and are constrained by ZFS performance. To address these issues, the author designs a heterogeneous filesystem decoupling architecture: the production side employs Ext4 combined with bcache to achieve peak IO performance, while the backup side deploys ZFS to leverage its snapshot capabilities for versioned management. In specific implementation, `rsync` performs incremental data synchronization every 30 minutes, triggering immediate ZFS snapshots. Coupled with Sanoid-configured tiered snapshot retention policies, this approach achieves high-frequency data protection while establishing a read-only time-travel mechanism to effectively defend against logical errors. The solution supports fine-grained single-file recovery as well as volume-level full rollback, avoiding the performance bottlenecks associated with direct ZFS mounting. The final conclusion indicates that this architecture maintains high performance in the production environment while providing reliable disaster recovery capabilities, making it a highly valuable balanced solution for core business systems.

Views 169
Data Backup Disaster Recovery Snapshot ZFS ext4 rsync
24
FEB

Anduin's CQE Theorem: Why Procurement in Large Organizations Always Ends Up Buying Electronic Garbage

Using AI-generated low-quality web pages as a starting point, the author delves into the absurdities and flaws of the domestic enterprise bidding and procurement system. He points out that in a game where "procedural justice" overrides "substantive justice," compliance and efficiency become rigid metrics, while quality is often sacrificed. Through the "CQE Theorem," he reveals the impossible trilemma faced by large organizations in balancing compliance, quality, and efficiency, leading to a situation where bad money drives out good, ultimately resulting in the procurement of "electronic garbage" with poor compatibility and terrible user experience. This mechanism, centered on process-based liability avoidance, not only distorts the value of technology but also inflicts mental torment on developers who hold themselves to high product standards. In contrast, markets such as Hong Kong prioritize outcome orientation and product experience, allowing users to vote directly with their feet. The article argues that in a domestic environment filled with "compliance-driven electronic garbage," the only way for technologists to break free from systemic constraints and return to the essence of technology is to step out of the quagmire and face real users directly.

Views 209
AI Generated Bidding Scandal Compliance Trap Enterprise IT Product vs Process Technical Integrity
22
FEB

Run Human Fall Flat on AnduinOS

This article addresses the severe lag or crash issues encountered with *Human: Fall Flat* on Linux systems, identifying the root cause as the official discontinuation of native Linux support since 2018. This has resulted in critical defects such as Split Lock penalties, out-of-bounds memory access, and an inability to join online multiplayer sessions. The article emphasizes that completely abandoning the native version and switching to run the Windows version via Proton 9 is the only viable solution. However, if the game still fails to launch or performs poorly in Proton mode, the core issue is typically the absence of critical 32-bit Vulkan and OpenGL driver libraries on the system, which prevents the DXVK component from functioning correctly. The resolution steps include: first, installing the corresponding 32-bit graphics drivers and Vulkan libraries; second, clearing the Steam compatibility layer containers and shader caches to eliminate environmental conflicts; third, clearing the launch options in Steam properties and explicitly forcing Proton 9.0; and finally, verifying the integrity of the game files and re-downloading necessary Windows runtime components. By following this series of operations, you can completely resolve game crashes and low frame rate issues, achieving full GPU acceleration and normal online connectivity with Windows players.

Views 226
32 bit Drivers Game Crashes Human Fall Flat Linux Gaming Proton 9 Steam Proton
21
FEB

Say Goodbye to Ugly Tag Suffixes: Building True Multi-Architecture Images with Docker Buildx

This article provides a comprehensive guide to building dual-architecture images for x86 and ARM64 using Docker buildx and QEMU. It begins with the installation of the base environment and the initialization of the builder, explaining how the `binfmt_misc` mechanism enables cross-platform instruction translation, allowing x86 machines to compile ARM code. The core practical section demonstrates how to use the `--platform` parameter to build multi-architecture images in one step, and explains the Manifest List principle: the builder independently compiles images for each architecture and then generates an index, ensuring that clients only download the layers corresponding to their specific architecture upon pull, thereby avoiding redundancy. The article specifically points out that multi-architecture images cannot be directly loaded into the traditional Docker local image list, recommending direct pushing or exporting in OCI format. For CI/CD scenarios, it recommends injecting QEMU support non-invasively by running the `tonistiigi/binfmt` container to keep the environment clean. Additionally, it covers verifying images using `regctl`, testing ARM images on x86 systems, a guide to avoiding common pitfalls regarding Dockerfile architecture compatibility, and an explanation of the purpose of buildx's background BuildKit containers and their resource management. Mastering these techniques enables efficient building and management of enterprise-level multi-architecture automated release pipelines.

Views 275
ARM64 Container Docker Docker Buildx Linux Multi Arch bash
10
JAN

Use Cloudflare with Caddy to Scrub Traffic

This article introduces a high-security reverse proxy configuration solution based on Caddy and Cloudflare. The core process uses a script to automatically fetch Cloudflare's latest IPv4 and IPv6 IP ranges and generate a `cloudflare_ips.conf` file containing the trusted proxy list and mandatory mutual TLS (mTLS) verification configuration. This configuration ensures that only Cloudflare nodes holding valid certificates can initiate requests, thereby supporting Full Strict mode. The build script `build_proxy.sh` is responsible for concatenating independent configuration files scattered across different business directories with the base baseline configuration to generate the final Caddyfile, achieving modularization and decoupling of business configurations for easier independent maintenance. During the deployment phase, the actual Cloudflare certificates are mounted into the container via Docker Volume, replacing the temporary certificates used during the build, enabling the service to perform actual work. Ultimately, this architecture implements a triple security mechanism: the use of Cloudflare certificates, strict mTLS client authentication, and trusting only Cloudflare as the front-end proxy. This design ensures extreme security while allowing backend services to transparently obtain the real client IP, balancing security and maintainability.

Views 158
ACME Caddy Cloudflare Docker MTLS Reverse Proxy