Apache RocketMQ 从入门到实战 (丁威)（Z-Library）

【One-Line Pitch】 A hands-on field guide to running Apache RocketMQ in production, pairing core concept explanations with real incident post-mortems from large-scale clusters. Best for backend engineers, SREs, and architects who already know messaging basics and now need to debug, tune, and operate RocketMQ under load. 【Book Arc】 - **Opening (~0%–10%)**: Frames the author's path into the RocketMQ community and lays out the mental model — core concepts, consumer groups, broadcast vs. clustering, offset storage, and why `autoCreateTopicEnable` is dangerous in production. - **Early (~10%–30%)**: Gets you operational: local install, running NameServer/Broker in IDEA for source-level debugging, HA master-slave mechanics, and the read/write-splitting rules that decide whether consumers pull from master or slave. - **Middle (~30%–60%)**: Moves into troubleshooting war stories — a console showing zero TPS while backlog drops, where a new consumer group starts consuming, JVM process exits, and the classic "topic expanded but some queues go unread" trap. - **Late (~60%–90%)**: Deepens into the storage and network internals behind "system busy / broker busy" errors — PageCache pressure, `transientStorePoolEnable`, MappedFile write paths, and the Netty request-processor/thread-pool binding model. - **Ending (~90%–100%)**: Consolidates operational judgment: how to read broker logs and metrics, when to suspect configuration vs. application, and how to reason from symptom to root cause. (Excerpts do not cover the final chapters in detail.) 【Key Takeaways】 - **Consumption semantics are a design decision, not a default** (Opening): broadcast vs. clustering, concurrent vs. ordered consumption, and per-group offset storage each carry different failure and retry behavior — ordered consumption retries forever, concurrent retries 16 times. - **`autoCreateTopicEnable=true` is a production anti-pattern** (Opening): auto-created topics only land on the broker that first received the message, so routing can collapse to a single node and messages never reach other brokers. - **HA is master-slave, not failover** (Early): the slave never takes over message sending; it only serves reads, and read/write splitting is governed by `slaveReadEnable`, backlog-vs-memory thresholds, and subscription-group broker hints. - **Zero TPS with shrinking backlog is explainable, not a bug** (Middle): when backlog exceeds ~40% of physical memory, consumption shifts to the slave, so master-side TPS stats read zero while messages are still being consumed. - **Where a new consumer group starts depends on `consumeFromWhere`** (Middle): `CONSUME_FROM_LAST_OFFSET` and friends resolve differently for retry topics vs. normal topics, and misreading this looks like a bug. - **Queue expansion can silently strand queues** (Middle): after scaling a topic's queues, some queues may go unconsumed if subscription groups aren't created on all brokers — the fix is often configuration, not application code. - **"system busy / broker busy" traces back to storage and threading** (Late): PageCache busy checks, `transientStorePoolEnable` direct-buffer locking, and bounded send thread-pool queues are the levers to inspect and tune. - **Debugging method beats memorized fixes** (Middle/Late): the recurring lesson is to find the pattern by comparing broken vs. healthy consumers, form a hypothesis, then verify — not to restart blindly. 【Reading Tips】 - Deep-read the Opening concept chapter and the Early HA/read-write-splitting sections; these underpin every later troubleshooting story. - Treat the Middle chapters as case studies: skim the code listings, but read the "problem analysis" and "answer" sections closely — that's where the transferable reasoning lives. - For the storage/network chapters, focus on the configuration parameters (`transientStorePoolEnable`, `sendThreadPoolQueueCapacity`, `slaveReadEnable`) rather than line-by-line source. - Keep a running list of "symptom → likely cause" pairs as you read; this book's value is diagnostic pattern-matching, not API reference. - If you're new to RocketMQ, pair this with the author's earlier internals book; this one assumes familiarity with pull/consume flows. 【Coverage Limits】 This guide is synthesized from stratified excerpts covering roughly the first half of the book plus selected late-chapter material; later chapters and any appendices are not represented, so the Ending stage is inferred rather than fully described.