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.
When a request crosses N microservices, how do you identify which one failed? Distributed tracing solves this problem. This article explains traces, spans, context propagation, the OpenTelemetry standard, sampling strategies, and how to diagnose slow calls and cross-service failures.
Clicking through cloud consoles is slow, error-prone, and difficult to reproduce. Infrastructure as Code (IaC) describes resources in code, with Terraform as the leading tool. This guide covers its declarative model, core workflow, state, modules, production backends, and team collaboration.
Keyset queries avoid deep OFFSET but cannot automatically freeze updates, deletes, and permission changes during paging. This article wraps filters, sorting, snapshot identity, and boundary tuples into a signed cursor and provides a sustainable snapshot implementation.
Re-running the past year's partitions on a production table easily overwrites real-time writes and leaves half-new, half-old results. This article has backfill write into an independent generation, completing a reversible cutover with a half-open range, dual-compute diffs, and a pointer CAS.
Setting each application instance's local connection pool to 50 does not mean the database will see only 50 connections. This article uses global permits, workload queues, and cancellation confirmation to cap total connections and prevent one tenant or a slow query from taking down every request.
Developers accustomed to relational modeling often struggle with MongoDB. Its core ideas are query-driven schema design, embedding vs. referencing, and denormalization. This article explains document modeling, its trade-offs, and when to choose a document database over a relational one.
Slow database? Tune before scaling up. This guide covers PostgreSQL configuration (shared_buffers, work_mem, and connections), query analysis with EXPLAIN and index optimization, connection pooling, VACUUM/ANALYZE maintenance, hardware, and a systematic locate-first tuning workflow.
Direct calls between services create tight coupling; event-driven design loosens it. This article explains EDA’s core model (producers publish events and consumers subscribe), events vs. messages, Event Sourcing, CQRS, Saga, and where it fits or does not. Do not force EDA solely for “decoupling.”
Knowing how to add a B-tree index is not the same as understanding indexes. This guide covers composite indexes and the leftmost prefix, covering indexes that avoid heap fetches, partial and expression indexes, specialized index types, and the write and maintenance costs involved in query-driven index design.
Acquiring a lock and continuing to own it are not the same fact. After a process pause, network partition, or lost renewal response, an old owner may still write. This article uses monotonic fencing tokens to reject stale mutations at the resource actually being protected.
Serverless lets you focus on code instead of servers, but its stateless design, cold starts, and billing model impose new constraints. This article explains its FaaS + BaaS core, where it fits (event-driven, bursty, and low-traffic workloads), its costs (cold starts, vendor lock-in, and difficult debugging), and when to use or avoid it.
Compose is not just for local development. For small and midsize projects, it can be ten times simpler than K8s in production. This guide covers environment layering, health checks, dependency ordering, restart policies, log drivers, resource limits, security, and when to choose Swarm or standalone Compose.