Tutorials
Practical guides from beginner to expert: frontend, backend, DevOps, cloud and design.
5 published articles
Database Engineering
Alibaba's Panda Index gives InnoDB unique keys a version history they never had
Panda Index pushes the clustered index's versioning down into unique-key records, adding transaction columns and an independent undo chain. The result is one record lock instead of several gap locks, and index-only reads that never touch the clustered index. Independent benchmarks are still missing.
2026-08-06
Node.js Observability & APM
Node.js traces break at every await. Alibaba's ARMS agent stitches them back
Node.js services now act as BFFs, gateways, and AI orchestration layers, so one request crosses databases, caches, queues, and LLM calls. Alibaba Cloud's ARMS Node.js agent merges OpenTelemetry tracing, runtime health checks, and AI observability into a single npm package.
2026-08-05
Cloud Infrastructure
The $0.09/GB barrier to cloud GPU hopping just disappeared
By mounting Hugging Face storage as a first-class SkyPilot backend with zero egress for reads, teams can now run training and inference on whichever cloud has free GPUs without duplicating data or paying per-GB transfer fees. The integration combines lazy mounting through hf-mount with Xet-powered deduplication to keep data movement efficient.
2026-07-15
Vector Databases
Inside Alibaba's engineering push to make pgvector work at billion-vector scale
Alibaba Cloud details how it transformed pgvector from an open-source extension into a production-grade vector engine supporting billions of vectors with millisecond response times, through quantization, performance tuning, ecosystem tools, distributed scaling, and deep kernel integration.
2023-10-11
Database
Alibaba just cut 70% off database index bloat. The trick is in the prefixes.
Secondary indexes often dominate database storage, especially in SaaS and e-commerce. PolarDB-X introduces secondary prefix compression (SPC), a lightweight per-page dictionary scheme that reduces index space by 30% to 70% without heavy CPU overhead. Benchmarks show up to 47% throughput improvement in I/O-bound read workloads.
2020-10-22