High performance MySQL optimization, backups, and replication (Baron Schwartz, Peter Zaitsev, Vadim Tkachenko)（Z-Library）

【One-Line Pitch】 A practitioner's field manual for making MySQL fast and keeping it alive: it teaches you to measure performance by response time, then tune schemas, queries, servers, and replication accordingly. Best for engineers who already run MySQL in production and need depth beyond the manual. 【Book Arc】 - **Opening (~0%–10%)**: Frames the whole discipline — what "high performance" means, how MySQL's architecture and storage engines (with MVCC and isolation-level trade-offs) shape everything later. Solves the "where do I even start" problem. - **Early (~10%–30%)**: Establishes benchmarking and profiling as the foundation — designing realistic benchmarks, reading the MySQL Benchmark Suite, and using query profiling tools to find the worst offenders by time, not by guesswork. - **Middle (~30%–50%)**: Moves into diagnosis and schema design — capturing diagnostic data for intermittent problems, using INFORMATION_SCHEMA statistics and system-call tracing, and choosing column types (e.g., ENUM vs VARCHAR joins) that actually change query speed. - **Late (~50%–70%)**: Advanced application-side patterns — materialized views via Flexviews, counter tables, and other ways to restructure hot data so concurrency stops being the bottleneck. - **Ending (~70%–100%)**: Operational durability — backup and recovery strategy (logical dumps, filesystem snapshots, point-in-time recovery, InnoDB crash recovery), binary-log management, replication, and the tooling ecosystem (XtraBackup, mydumper, monitoring tools). 【Key Takeaways】 - **Performance is response time, not resource usage** (Early): the book defines performance by task completion time, which reframes optimization around queries and statements rather than CPU or memory graphs. - **Benchmark deliberately, not accidentally** (Early): match the benchmark to the workload (don't use a decision-support benchmark for OLTP), snapshot production data, and accept that single-node tests trade realism for fewer variables. - **Profile before you optimize** (Early): tools like pt-query-digest rank queries by total time and flag high-variance queries as prime candidates — the variance-to-mean ratio is a practical signal. - **Intermittent problems need a trigger and a data collector** (Middle): the hard part is knowing when the problem happens; thresholds should sit above healthy baselines, not at them. - **Storage engine choice changes query behavior** (Early): a query like `SELECT COUNT(*)` is fast on MyISAM but not necessarily elsewhere — engine migration can make things slower, not faster. - **Schema details have measurable costs** (Middle): joining ENUM columns to VARCHAR columns was slower than ENUM-to-ENUM, so lookup tables with integer keys remain a sound design habit. - **Restructure hot data instead of fighting it** (Late): materialized views and separate counter tables reduce contention and let complex queries become simple, fast ones. - **Backups are a design problem, not a checkbox** (Ending): the book treats binary-log management, snapshots, point-in-time recovery, and crash recovery as one connected discipline. 【Reading Tips】 - Deep-read the benchmarking and profiling chapters early — they give you the measurement vocabulary the rest of the book assumes. - Skim the tool listings (backup utilities, monitoring systems) on a first pass; return when you actually need to pick one. - Treat the storage-engine and schema chapters as reference: revisit them when a specific query misbehaves. - The backup/recovery material is the highest-stakes section — read it with your own recovery plan open beside you. - Don't expect copy-paste recipes; the book teaches judgment about when a technique applies. 【Coverage Limits】 The excerpts are heavily front-loaded (table of contents, acknowledgments, early chapters) with only fragments from the middle and late sections, so the replication, high-availability, and failover material is only partially represented here.