From e5b3ef8eb645347ae1fd76118f02f1d2d756c13f Mon Sep 17 00:00:00 2001 From: cat-shark <1716967236@qq.com> Date: Fri, 9 Oct 2026 21:00:33 +0800 Subject: [PATCH] =?UTF-8?q?fix:=20=E7=BC=A9=E7=9F=AD=E6=A0=87=E8=AE=B0?= =?UTF-8?q?=E5=94=AF=E4=B8=80=E9=94=AE=E9=81=BF=E5=85=8D=E8=B6=85=E5=87=BA?= =?UTF-8?q?=20InnoDB=20=E7=B4=A2=E5=BC=95=E9=95=BF=E5=BA=A6=E4=B8=8A?= =?UTF-8?q?=E9=99=90?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../db/migration/V20261009_9__add_markers.sql | 9 +- .../MigrationIndexKeyLengthTest.java | 304 ++++++++++++++++++ 2 files changed, 312 insertions(+), 1 deletion(-) create mode 100644 src/test/java/com/guo/learningprogresstracker/migration/MigrationIndexKeyLengthTest.java diff --git a/src/main/resources/db/migration/V20261009_9__add_markers.sql b/src/main/resources/db/migration/V20261009_9__add_markers.sql index afcf138..ce8c0e5 100644 --- a/src/main/resources/db/migration/V20261009_9__add_markers.sql +++ b/src/main/resources/db/migration/V20261009_9__add_markers.sql @@ -25,7 +25,14 @@ CREATE TABLE markers deleted INT DEFAULT 0 NOT NULL COMMENT '逻辑删除符', -- 同一对象上「同类 + 同名 + 同值」只保留一条(重复打标是幂等的)。 -- 值参与唯一键,因为系统标记是「名 = 值」成对表达的。 - UNIQUE KEY uk_markers_object_marker (created_by, object_type, object_num, kind, marker_name, marker_value), + -- + -- 注意:这里**不含 created_by**,两个原因: + -- 1. object_num 已经全局唯一(碎片自增 id 或任务编号),带上 created_by 是冗余; + -- 2. InnoDB 的索引键长度上限是 3072 字节(DYNAMIC 行格式);带上 created_by 后 + -- 在 utf8mb4 下会达到 3620 字节,建表直接报 error 1071。 + -- H2 不校验这个限制,所以只有真实 MySQL 才会暴露——改动索引时务必用 + -- MigrationIndexKeyLengthTest 复算。 + UNIQUE KEY uk_markers_object_marker (object_type, object_num, kind, marker_name, marker_value), -- 按标记找对象(点标签即筛选) KEY idx_markers_lookup (created_by, marker_name, object_type, object_num), -- 列出某对象的全部标记 diff --git a/src/test/java/com/guo/learningprogresstracker/migration/MigrationIndexKeyLengthTest.java b/src/test/java/com/guo/learningprogresstracker/migration/MigrationIndexKeyLengthTest.java new file mode 100644 index 0000000..98d8de1 --- /dev/null +++ b/src/test/java/com/guo/learningprogresstracker/migration/MigrationIndexKeyLengthTest.java @@ -0,0 +1,304 @@ +package com.guo.learningprogresstracker.migration; + +import org.junit.jupiter.api.Test; + +import java.nio.file.Files; +import java.nio.file.Path; +import java.util.ArrayList; +import java.util.Comparator; +import java.util.LinkedHashMap; +import java.util.List; +import java.util.Locale; +import java.util.Map; +import java.util.regex.Matcher; +import java.util.regex.Pattern; +import java.util.stream.Stream; + +import static org.junit.jupiter.api.Assertions.assertTrue; +import static org.junit.jupiter.api.Assertions.fail; + +/** + * 迁移脚本的索引键长度守卫。 + * + *

背景:InnoDB 对索引键长度有硬上限——DYNAMIC/COMPRESSED 行格式是 3072 字节, + * 超了直接报 error 1071 让建表/加索引失败。而本项目在 utf8mb4 下 VARCHAR(n) 要按 4n 字节算, + * 所以一个看起来人畜无害的「多列唯一键」很容易越界。

+ * + *

危险之处在于测试环境用 H2,它不校验这个限制:迁移能在测试里跑通、单个 INSERT 也正常, + * 但一部署到真实 MySQL 就起不来。这条测试直接读迁移文件复算每个索引的键长度, + * 把这类问题挡在提交之前。

+ * + *

新增/修改索引后若本测试失败:优先减少参与索引的列,或确认某些列在业务上是否真的需要进键。

+ */ +class MigrationIndexKeyLengthTest { + + /** InnoDB DYNAMIC 行格式下的索引键前缀长度上限(字节) */ + private static final int MAX_KEY_BYTES = 3072; + + /** utf8mb4 每字符最多 4 字节 */ + private static final int UTF8MB4_BYTES_PER_CHAR = 4; + + private static final Path MIGRATION_DIR = Path.of("src/main/resources/db/migration"); + + /** 形如 varchar(255) / tinyint(1) 的列声明 */ + private static final Pattern WITH_LENGTH = Pattern.compile("^([a-z]+)\\((\\d+)(?:\\s*,\\s*\\d+)?\\)$"); + + /** 定长类型的字节数;VARCHAR/CHAR 按字符数 × 4 另算 */ + private static final Map FIXED_BYTES = Map.ofEntries( + Map.entry("tinyint", 1), + Map.entry("smallint", 2), + Map.entry("mediumint", 3), + Map.entry("int", 4), + Map.entry("integer", 4), + Map.entry("bigint", 8), + Map.entry("float", 4), + Map.entry("double", 8), + Map.entry("decimal", 8), + Map.entry("datetime", 8), + Map.entry("timestamp", 8), + Map.entry("date", 3), + Map.entry("time", 3), + Map.entry("year", 1)); + + private record IndexDef(String file, String table, String name, List columns) { + } + + @Test + void everyIndexFitsWithinInnoDbKeyLimit() throws Exception { + Map> tableColumns = new LinkedHashMap<>(); + List indexes = new ArrayList<>(); + + for (Path file : migrationFiles()) { + String sql = stripComments(Files.readString(file)); + Parser parser = new Parser(file.getFileName().toString()); + parser.parse(sql, tableColumns, indexes); + } + + assertTrue(indexes.size() > 10, "应至少解析出若干索引,实际 " + indexes.size() + "(解析器是否失效?)"); + + List violations = new ArrayList<>(); + for (IndexDef index : indexes) { + Map columns = tableColumns.get(index.table()); + if (columns == null) { + violations.add(index.file() + " 的索引 " + index.name() + " 引用了未知表 " + index.table()); + continue; + } + int total = 0; + List unknown = new ArrayList<>(); + for (String column : index.columns()) { + String declaredType = columns.get(column); + if (declaredType == null) { + unknown.add(column); + continue; + } + Integer bytes = bytesOfDeclared(declaredType); + if (bytes == null) { + // 只对「真的进了索引」的列报错:TEXT 必须先确认是否需要前缀索引 + violations.add(index.file() + " 的索引 " + index.name() + " 引用了列 " + column + + "(类型 " + declaredType + "),无法按定长计算键长"); + continue; + } + total += bytes; + } + if (!unknown.isEmpty()) { + // 解析不到就报出来,避免新增列/写法变化时这条守卫静默失效 + violations.add(index.file() + " 的索引 " + index.name() + " 引用了无法解析的列 " + unknown + + "(表 " + index.table() + ")"); + continue; + } + if (total > MAX_KEY_BYTES) { + violations.add(String.format("%s 的索引 %s 键长 %d 字节,超过 InnoDB 上限 %d 字节;列=%s", + index.file(), index.name(), total, MAX_KEY_BYTES, index.columns())); + } + } + + if (!violations.isEmpty()) { + fail("迁移脚本里有索引键长度超限(真实 MySQL 会报 error 1071,H2 测不出来):\n - " + + String.join("\n - ", violations)); + } + } + + /** 把列声明(如 varchar(255)、datetime)换算成索引键占用字节;无法定长计算的返回 null */ + private static Integer bytesOfDeclared(String declaredType) { + String normalized = declaredType.toLowerCase(Locale.ROOT).strip(); + Matcher withLength = WITH_LENGTH.matcher(normalized); + if (withLength.matches()) { + String base = withLength.group(1); + int length = Integer.parseInt(withLength.group(2)); + if (base.equals("varchar") || base.equals("char")) { + return length * UTF8MB4_BYTES_PER_CHAR; + } + return FIXED_BYTES.get(base); + } + return FIXED_BYTES.get(normalized); + } + + private static List migrationFiles() throws Exception { + try (Stream files = Files.list(MIGRATION_DIR)) { + return files.filter(path -> path.getFileName().toString().endsWith(".sql")) + .sorted(Comparator.comparing(path -> path.getFileName().toString())) + .toList(); + } + } + + /** 去掉整行注释,避免注释里的示例被当成 DDL */ + private static String stripComments(String sql) { + StringBuilder result = new StringBuilder(); + for (String line : sql.split("\r?\n")) { + if (line.trim().startsWith("--")) { + continue; + } + result.append(line).append('\n'); + } + return result.toString(); + } + + /** 极简 DDL 解析器:只覆盖本仓库实际用到的写法,遇到不认识的列会让测试失败而不是静默放过 */ + private static final class Parser { + + private static final Pattern CREATE_TABLE = + Pattern.compile("(?i)^\\s*create\\s+table\\s+(?:if\\s+not\\s+exists\\s+)?(\\w+)\\s*\\(?\\s*$"); + private static final Pattern COLUMN = + Pattern.compile("(?i)^\\s*(\\w+)\\s+([a-z]+)\\s*(?:\\(\\s*(\\d+)(?:\\s*,\\s*\\d+)?\\s*\\))?"); + private static final Pattern ALTER_TABLE = Pattern.compile("(?i)^\\s*alter\\s+table\\s+(\\w+)\\s+(.*)$"); + private static final Pattern ADD_OR_MODIFY_COLUMN = + Pattern.compile("(?i)^(?:add|modify)\\s+column\\s+(\\w+)\\s+([a-z]+)\\s*(?:\\(\\s*(\\d+)(?:\\s*,\\s*\\d+)?\\s*\\))?"); + private static final Pattern RENAME_COLUMN = + Pattern.compile("(?i)^rename\\s+column\\s+(\\w+)\\s+to\\s+(\\w+)"); + private static final Pattern INLINE_KEY = + Pattern.compile("(?i)^\\s*(?:unique\\s+)?(?:key|index)\\s+(\\w+)\\s*\\(([^)]*)\\)"); + private static final Pattern CREATE_INDEX = + Pattern.compile("(?i)^\\s*create\\s+(?:unique\\s+)?index\\s+(\\w+)\\s+on\\s+(\\w+)\\s*\\(([^)]*)\\)"); + private static final Pattern ADD_UNIQUE_CONSTRAINT = + Pattern.compile("(?i)^add\\s+constraint\\s+(\\w+)\\s+unique\\s*\\(([^)]*)\\)"); + + private final String fileName; + private String currentTable; + private boolean insideTableBody; + + Parser(String fileName) { + this.fileName = fileName; + } + + void parse(String sql, Map> tableColumns, List indexes) { + for (String rawLine : sql.split("\r?\n")) { + String line = rawLine.strip(); + if (line.isEmpty()) { + continue; + } + + Matcher createTable = CREATE_TABLE.matcher(line); + if (createTable.matches()) { + currentTable = createTable.group(1).toLowerCase(Locale.ROOT); + insideTableBody = true; + tableColumns.computeIfAbsent(currentTable, key -> new LinkedHashMap<>()); + continue; + } + + if (insideTableBody) { + if (line.startsWith(")")) { + insideTableBody = false; + continue; + } + if (line.startsWith("(")) { + continue; + } + Matcher inlineKey = INLINE_KEY.matcher(line); + if (inlineKey.find()) { + indexes.add(new IndexDef(fileName, currentTable, inlineKey.group(1), + splitColumns(inlineKey.group(2)))); + continue; + } + if (isConstraintKeyword(line)) { + continue; + } + Matcher column = COLUMN.matcher(line); + if (column.find()) { + String name = column.group(1).toLowerCase(Locale.ROOT); + if (isTypeKeyword(name)) { + continue; + } + // 只记声明,等索引真正引用到它时再算字节(避免对没进索引的 TEXT 列误报) + tableColumns.get(currentTable).put(name, declaredType(column.group(2), column.group(3))); + } + continue; + } + + Matcher createIndex = CREATE_INDEX.matcher(line); + if (createIndex.matches()) { + indexes.add(new IndexDef(fileName, createIndex.group(2).toLowerCase(Locale.ROOT), + createIndex.group(1), splitColumns(createIndex.group(3)))); + continue; + } + + Matcher alterTable = ALTER_TABLE.matcher(line); + if (!alterTable.matches()) { + continue; + } + String table = alterTable.group(1).toLowerCase(Locale.ROOT); + String rest = alterTable.group(2).strip(); + Map columns = tableColumns.computeIfAbsent(table, key -> new LinkedHashMap<>()); + + Matcher addUnique = ADD_UNIQUE_CONSTRAINT.matcher(rest); + if (addUnique.find()) { + indexes.add(new IndexDef(fileName, table, addUnique.group(1), splitColumns(addUnique.group(2)))); + continue; + } + + Matcher rename = RENAME_COLUMN.matcher(rest); + if (rename.find()) { + String from = rename.group(1).toLowerCase(Locale.ROOT); + String to = rename.group(2).toLowerCase(Locale.ROOT); + String declared = columns.remove(from); + if (declared != null) { + columns.put(to, declared); + } + continue; + } + + Matcher column = ADD_OR_MODIFY_COLUMN.matcher(rest); + if (column.find()) { + String name = column.group(1).toLowerCase(Locale.ROOT); + columns.put(name, declaredType(column.group(2), column.group(3))); + } + } + } + + private static boolean isConstraintKeyword(String line) { + String lower = line.toLowerCase(Locale.ROOT); + return lower.startsWith("primary ") + || lower.startsWith("foreign ") + || lower.startsWith("constraint ") + || lower.startsWith("check "); + } + + private static boolean isTypeKeyword(String name) { + return switch (name) { + case "primary", "foreign", "constraint", "check", "unique", "key", "index" -> true; + default -> false; + }; + } + + private static List splitColumns(String raw) { + List columns = new ArrayList<>(); + for (String part : raw.split(",")) { + String name = part.strip().toLowerCase(Locale.ROOT); + // 去掉可能的后缀(如 name(50) 前缀索引、ASC/DESC) + int paren = name.indexOf('('); + if (paren > 0) { + name = name.substring(0, paren); + } + name = name.replaceAll("\\s+(asc|desc)$", "").strip(); + if (!name.isEmpty()) { + columns.add(name); + } + } + return columns; + } + + private static String declaredType(String type, String length) { + String base = type.toLowerCase(Locale.ROOT); + return length == null ? base : base + "(" + length + ")"; + } + } +}