Highlights from High‑Availability Architecture Discussions: Plugins, Kanban Tools, Scheduling Services, Massive Tables, and ID Generation
This article compiles recent high‑availability architecture discussions covering plugin‑based system design, simple kanban tools, independent scheduling services, challenges of managing 60‑billion‑row tables, and the trade‑offs between UUIDs and custom distributed ID generators, offering practical insights for backend engineers.
Because the high‑availability architecture group generates a large volume of technical discussions daily and continuously introduces new ideas, this post extracts several noteworthy topics for peers interested in architecture.
Discussion on Plugin Architecture
Similar to the Nginx core, its security, HTTPS, and gzip modules are injected as plugins, each implementing three fixed interfaces plus a plugin registry, achieving highly extensible programming. Java's XMPP client Spark follows the same principle.
Kanban Tools
Whiteboard Approach : Using a physical whiteboard with sticky notes provides a tangible, durable, and intuitive way to track tasks, especially for small teams; larger teams may still prefer digital tools. Tool Approach : Options mentioned include Trello, Teambition (domestic), and Microsoft Project (including Project Server).
Scheduling Services
Looking for a standalone service that allows manual or API‑driven creation of scheduled jobs, supports email/SMS/queue notifications, and provides a monitoring UI, as an alternative to building on Quartz.
In‑house Quartz‑based solution: custom wrappers for database or Redis storage, job pausing, error monitoring, and retries. Other solutions: Alibaba's tbschedule, Jenkins‑triggered web services or scripts, which can replace Quartz with easier configuration. Conclusion: tbschedule appears to meet the requirements.
Handling a Single Table with 6 Billion Records
A senior DBA from Meituan shared MySQL selection criteria, raising concerns about bottlenecks at such scale.
Concerned View : Questions about B‑tree node order, horizontal scaling impact on efficiency, SSD effects, and risks of not partitioning. Supportive View : InnoDB controls tree depth, works fine on mechanical drives; switching to TokuDB increased I/O pressure.
ID Generation Discussion
Architecture uses non‑auto‑increment, distributed, partitioned IDs generated by a global service (e.g., a Go implementation using KV store and atomic locks). Compared to UUIDs, custom IDs are shorter, index‑friendly, and have lower storage and I/O overhead, especially for InnoDB primary and secondary indexes.
Industry News
How would you program a computer with 10 TB of RAM? ( link )
Microsoft open‑sourced WinObjC to provide an Objective‑C development environment for Visual Studio/Windows ( link )
For further high‑availability architecture exchange, reply with "arch" to apply for the group. A new group has been opened to accommodate more participants.
Edited by Tim Yang. Follow the "High‑Availability Architecture" public account for more content; please credit the source when reposting.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
High Availability Architecture
Official account for High Availability Architecture.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.