DevOps、Docker、Kubernetes、可观测性
This collection features articles and community discussions in the 云原生 space, focusing on real-world engineering scenarios, tool selection, practical paths, and risk boundaries.
The page is continuously updated and serves as a learning roadmap, technology selection reference, and technical review entry point.
Training and batch inference should not compete for GPUs by creating large numbers of Pending Pods. This guide uses Kueue LocalQueue, ClusterQueue, and ResourceFlavor to define tenant admission, quota borrowing, preemption, and rollback.
A table's ability to time travel does not mean old snapshots remain recoverable forever. This article binds reader references, snapshot expiration, orphan-file cleanup, and legal holds to the same generation, then uses cold recovery to prove that metadata and data files were not reclaimed prematurely.
Some logic is more efficient in the database, including automatic auditing, cascading updates, and data validation. But overusing triggers makes behavior implicit and hard to debug. This article explains how PostgreSQL triggers and functions work, when to use them, when not to, and why triggers demand restraint.
Sending emails, generating reports, and processing uploads time out inside a request—move them to a background queue. This article covers the queue core (enqueue/worker/retry/monitoring), tech selection (BullMQ/Sidekiq/Celery), delayed and scheduled jobs, idempotency, and how to decide “what should be async.”
Production LLM gateways cannot rely on round-robin routing alone. This article presents an auditable routing policy jointly driven by capability constraints, data boundaries, latency budgets, cost caps, health status, and fallback order.
Read-heavy and write-light? Add read replicas to spread reads across databases while the primary handles writes. This article covers primary-replica architecture, application- and proxy-level routing, replication-lag consistency traps, and when read replicas are or are not appropriate.
Locks preserve correctness during concurrent writes, but poor locking causes delays and deadlocks. This article explains PostgreSQL row, table, and advisory locks, automatic deadlock detection, concurrency control with SELECT FOR UPDATE, and ways to reduce contention through short transactions, consistent lock ordering, and optimistic locking.
Manually trigger downstream synchronization after every database change? CDC automatically captures changes and streams them to search, caches, data warehouses, and services in real time. Learn how WAL-based CDC works, how Debezium implements it, where it fits, and why it outperforms dual writes and scheduled polling.
PostgreSQL's killer feature is its extension ecosystem: pgvector for vector search, PostGIS for geospatial data, TimescaleDB for time series, and zhparser for Chinese tokenization. This guide covers their value, 10 popular extensions, installation and management, and when one PG instance can replace multiple components.
Sharding is more than splitting data: a bad shard key forces full scans, while cross-shard joins, aggregation, and pagination become painful. This article covers shard-key design, application merging, denormalization, CQRS, globally unique Snowflake IDs, and why you should usually avoid sharding at first.
Need the top 3 per category, running totals, or per-group rankings? Window functions handle them in one SQL query without GROUP BY plus subqueries. Learn the core concepts of OVER, PARTITION BY, ORDER BY, and frames; common ranking, cumulative, year-over-year, and top-N patterns; and the fundamental difference from GROUP BY.
Model providers may partially time out, rate-limit requests, or silently degrade quality. This article covers incident declaration criteria, UTC evidence timelines, isolation matrices, Prometheus queries, fallback thresholds, and recovery validation.