Introduction In the previous article, we reviewed the three-year evolution of stream-metrics-route and mentioned that the “dual hashmod scheduling” is the core scheduling mechanism of the entire gateway. However, during continuous production operation, one fatal flaw of hashmod became increasingly obvious—every scaling operation triggers full data redistribution. This article documents the complete decision process of migrating from hash % N (hashmod) to Jump Consistent Hash: which candidate algorithms were evaluated, why Jump Hash was ultimately chosen, and the specific impact before and after migration.

Introduction You’re debugging a frontend engineering issue — the page is behaving abnormally. You ask an AI to open the page with a browser tool and check the console for errors. The AI opens the page, scans around, and tells you: The console is clean, no errors whatsoever. You’re skeptical. You open Chrome DevTools yourself — three bright red errors are staring you in the face, the page has already crashed into a white screen. The AI visited the exact same page using a Headless browser, so why did it catch nothing?

Why This Was Built Anyone managing servers has probably had this experience: compliance audit comes, SSH into machines one by one to check—SSH config correct, SELinux enabled, firewall running, any expired accounts, password policies compliant. A few machines are fine; dozens or hundreds becomes purely manual grunt work. And the more painful part: none of this has continuous monitoring. You check compliance today, someone changes a config tomorrow, and you’d never know.

One Month In It’s been nearly a month since the previous article Switched My Blog Theme: From Hugo NexT to Self-Written Zhi. During this month, the theme has been running stably without major issues. Replacing the NexT theme was the right decision — although there was some initial adjustment, the experience is now significantly better. After a month of use, the theme’s stability has exceeded expectations. My initial concerns — whether pure Hugo Pipes without build tools could support complex requirements — were proven unfounded. Daily maintenance has become very simple; modifying a feature no longer requires digging through deeply nested SCSS files — one CSS file gets the job done.

The previous article introduced MiBeeNvr’s basic features and design philosophy. It’s only been a week since v0.1.0, and v0.2.0 follows right behind. This update is substantial — 15 new features, some I needed myself, others from community feedback. This article covers three things: what’s new in v0.2.0, how to deploy from scratch, and some practical tips for real-world use. v0.2.0 New Features Overview This update has a lot of content. Here’s a breakdown by category:

Introduction It’s been exactly three years since the previous article Applying VictoriaMetrics Stream Aggregation for Metrics was published in March 2023. In these three years, the VictoriaMetrics ecosystem has undergone tremendous changes—let’s revisit the issues raised in that blog post, see what the official project has resolved, and where our stream-metrics-route project stands today. I. Problems We Encountered Three Years Ago Let’s quickly recap the core issue list from the 2023 blog post:

Why I built a surveillance camera with ESP32-S3 before, and it worked well. Later, while rummaging through a drawer, I found an AI-Thinker ESP32-CAM development board — that classic board costing about ten bucks with a built-in OV2640 camera and TF card slot. No reason to let it go to waste, so I built another one: ai-thinker-esp32-cam. This time I wrote the firmware from scratch using ESP-IDF again, with similar capabilities to the previous project but with lots of adaptations for the AI-Thinker board. Here’s what it ended up doing:

The previous article covered the initial configuration setup. This one documents the adjustments after two weeks of running: expanding from single vendor to a four-tier model pool, adding fallback chains, hitting the GLM-4.5-air trap of analyzing without writing code. This post covers: fallback strategy design, complete free model pool inventory and analysis, concurrency control configuration, and the decision process for GLM-4.5-air replacement. After the previous article’s initial configuration, I ran it for two weeks — all the issues that needed fixing surfaced.

I have several cameras at home — a few Xiaomi cameras, some DIY ESP32 cameras, and multiple Raspberry Pi CSI cameras. I’d been using cloud storage solutions, but I was never comfortable with them: vendor lock-in, network dependency, and the costs add up. So I decided to build my own NVR system, called MiBeeNvr. Why Build MiBeeNvr To be honest, I was never satisfied with existing cloud storage solutions. Take Xiaomi cameras, for example. By default, you can only view them through the Mi Home app. Recordings are either stored on an SD card (limited capacity, frequent plugging/unplugging) or in the cloud. Cloud storage costs tens of dollars per month, and there’s the privacy concern — you never know when the manufacturer might use your video data for AI training or sell it to third parties. Not to mention vendor lock-in — switching platforms is nearly impossible.

Why We have a few parrots at home, and during the workday nobody’s around. I wanted to check in on them anytime. The requirement sounds simple: real-time video streaming, recording to storage, and ideally automatic backup to NAS. Off-the-shelf cameras are either expensive or require installing apps, registering accounts, and binding phone numbers — privacy concerns. I just want to watch my birds, not stream video to someone else’s server.