Capturing thoughts and reflections while coding.
This article analyzes nginx 1.31's newly open-sourced least_time load balancing, detailing its EMA-based algorithm, header vs last_byte semantics for streaming, and real-world tests showing 9x latency reduction on heterogeneous backends and 3x difference in streaming scenarios, plus adaptive behavior trade-offs.
The uni-app mini-program compiler (alpha) incorrectly replaces `!b` with undeclared `_a_b` when a local variable is named `b`, causing a runtime ReferenceError only when specific data triggers the condition; the bug evades build, TypeScript, and ESLint checks, and is fixed by renaming the variable.
This guide explains why Docker Compose v1 is obsolete, details the pitfalls of manually installing the v2 CLI plugin, and recommends using Ubuntu's apt package docker-compose-v2 for automatic updates, network reliability, and seamless migration with zero container downtime.
The author details removing filesystem tool plugins from DeepSeek Harness (DSH) to secure multi-tenant container deployments, covering plugin architecture, overlay-based disabling, impact on 6 tools, real-model attack testing showing bash as a residual channel mitigated by sandbox fail-closed behavior, and resource metrics showing negligible savings.
The article walks through four carefully designed JMH benchmarks—method inlining, SIMD vectorization, lock elimination, and a Netty AsciiString regression—to show how JVM JIT optimizations appear in assembly, explain the performance impact (up to 1500×), and demonstrate how to diagnose and fix real‑world issues using assembly analysis.
The article explains why the default 7200‑second TCP KeepAlive idle time causes RSTs in cloud environments, demonstrates Netty configuration for finer‑grained keepalive settings, shares real‑world cloud‑networking failures, and recommends adjusting Linux kernel parameters as a reliable solution.
A remote Docker container kept restarting with no logs; by inspecting the ExitCode and analyzing build cache and file permissions, the author identified a polluted build layer that produced an empty binary and a UID mismatch on a mounted key file, then applied concrete fixes.
The article details a complete HeFeng weather integration for a mini‑program, covering front‑end icon‑font rendering, Rust‑based backend JWT authentication with Ed25519, three upstream API calls, a two‑tier cache for real‑time and historical data, gzip auto‑decompression, request‑timeout layering, and environment‑variable driven deployment configuration.
A benchmark on an 8‑core Linux 6.6 system shows that switching Netty from epoll to io_uring lets a half‑sized thread pool achieve 3 % higher throughput, more than double per‑thread efficiency, and a 67 % reduction in CPU migrations, challenging the traditional rule of using twice‑the‑core thread counts.
The article compares two Java‑based integration methods for sending pure‑text documents to RagFlow—first uploading an empty placeholder file then adding chunks, and later writing the text to a temporary file and uploading it directly—detailing implementation, pitfalls, and why the latter is preferred.
