mirror of
https://github.com/Syngnat/GoNavi.git
synced 2026-06-29 06:31:23 +08:00
✨ feat(explain): 扩展索引建议规则引擎至 15 条
- 新增规则：LIKE 前缀通配、函数包裹列、笛卡尔积风险、OR 条件无索引、大 OFFSET 分页、SELECT * + JOIN 模式 - 阈值常量：large_offset（10000）、cartesian_product（100000）、wide_table（20 列） - 测试覆盖：新增 6 个用例验证规则触发与抑制（含边界场景）
This commit is contained in:
Syngnat
@@ -40,6 +40,10 @@ const (
|
||||
ruleEstimationSkewRatio float64 = 10.0
|
||||
ruleHighTotalCostThreshold float64 = 1000.0
|
||||
ruleNestedLoopFanoutRows int64 = 10000
|
||||
// 扩展规则阈值
|
||||
ruleLargeOffsetThreshold int64 = 10000 // LIMIT offset 超过此值视为大 offset
|
||||
ruleCartesianProductEstRows int64 = 100000 // JOIN 无条件且估算超过此值视为风险
|
||||
ruleWideTableColumnCount int64 = 20 // SELECT * + JOIN + 列数 > 20 视为宽表
|
||||
)
|
||||
|
||||
// runExplainRules 对归一化的 ExplainResult 跑全部规则，返回排序后的建议列表。
|
||||
@@ -54,6 +58,9 @@ func runExplainRules(result connection.ExplainResult) []connection.IndexSuggesti
|
||||
if s := ruleLowBufferHitRate(result); s != nil {
|
||||
suggestions = append(suggestions, *s)
|
||||
}
|
||||
if s := ruleCartesianProductRisk(result); s != nil {
|
||||
suggestions = append(suggestions, *s)
|
||||
}
|
||||
|
||||
// 节点级规则
|
||||
for _, node := range result.Nodes {
|
||||
@@ -66,6 +73,11 @@ func runExplainRules(result connection.ExplainResult) []connection.IndexSuggesti
|
||||
ruleHighEstimationSkew,
|
||||
ruleNestedLoopHighFanout,
|
||||
ruleUsingTempBTreeOrder,
|
||||
ruleLikeLeadingWildcard,
|
||||
ruleFunctionOnColumn,
|
||||
ruleLargeOffsetPagination,
|
||||
ruleSelectStarWithJoin,
|
||||
ruleOrConditionNoIndex,
|
||||
}
|
||||
for _, ruleFn := range rules {
|
||||
if s := ruleFn(result, node); s != nil {
|
||||
@@ -432,3 +444,178 @@ func joinColumnsForReason(columns []string) string {
|
||||
}
|
||||
return strings.Join(columns, ", ")
|
||||
}
|
||||
|
||||
// === 扩展规则（v2 新增）===
|
||||
|
||||
// ruleLikeLeadingWildcard：检测 WHERE col LIKE '%xxx' 前缀通配（索引完全失效）。
|
||||
// 通过节点的 filter 文本判断，模式如 "col like '%xxx'"。
|
||||
func ruleLikeLeadingWildcard(_ connection.ExplainResult, node connection.ExplainNode) *connection.IndexSuggestion {
|
||||
filter := extractNodeFilterText(node)
|
||||
if filter == "" {
|
||||
return nil
|
||||
}
|
||||
lower := strings.ToLower(filter)
|
||||
// 简化匹配：col like '%xxx' 模式（前导 % 让 B-Tree 索引失效）
|
||||
if !strings.Contains(lower, " like '%") && !strings.Contains(lower, " like\"%") {
|
||||
return nil
|
||||
}
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityCritical,
|
||||
Rule: "like_leading_wildcard",
|
||||
Reason: fmt.Sprintf("LIKE 前缀通配（%q）导致索引失效；考虑改用全文索引或前置常量前缀", truncateForReason(filter, 80)),
|
||||
AffectedNodeID: node.ID,
|
||||
AffectedTable: node.Table,
|
||||
EstRows: node.EstRows,
|
||||
}
|
||||
}
|
||||
|
||||
// ruleFunctionOnColumn：检测 WHERE func(col) = ? 形式（函数包裹列让索引失效）。
|
||||
// 模式如 "upper(col) =" / "date_format(col, ...) =" / "col + 1 =" 等。
|
||||
func ruleFunctionOnColumn(_ connection.ExplainResult, node connection.ExplainNode) *connection.IndexSuggestion {
|
||||
filter := extractNodeFilterText(node)
|
||||
if filter == "" {
|
||||
return nil
|
||||
}
|
||||
// 扫描常见函数模式：函数名 + (
|
||||
lower := strings.ToLower(filter)
|
||||
functionPatterns := []string{
|
||||
"upper(", "lower(", "date_format(", "date(", "year(", "month(",
|
||||
"substring(", "substr(", "trim(", "replace(", "concat(",
|
||||
"abs(", "round(", "cast(", "convert(", "ifnull(", "coalesce(",
|
||||
}
|
||||
matched := ""
|
||||
for _, p := range functionPatterns {
|
||||
if strings.Contains(lower, p) {
|
||||
matched = p
|
||||
break
|
||||
}
|
||||
}
|
||||
if matched == "" {
|
||||
return nil
|
||||
}
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityCritical,
|
||||
Rule: "function_on_column",
|
||||
Reason: fmt.Sprintf("WHERE 条件中 %s... 包裹列，导致该列上的索引失效；考虑重写为列 = func(常量) 形式或在函数上建表达式索引", matched),
|
||||
AffectedNodeID: node.ID,
|
||||
AffectedTable: node.Table,
|
||||
EstRows: node.EstRows,
|
||||
}
|
||||
}
|
||||
|
||||
// ruleLargeOffsetPagination：检测 LIMIT 大 offset 分页（如 LIMIT 100000, 10）。
|
||||
// 大 offset 让数据库扫描并丢弃前 N 行，性能随 offset 线性下降。
|
||||
func ruleLargeOffsetPagination(_ connection.ExplainResult, node connection.ExplainNode) *connection.IndexSuggestion {
|
||||
if node.OpType != connection.ExplainOpLimit {
|
||||
return nil
|
||||
}
|
||||
// LIMIT 节点的 EstRows 通常是返回行数（小），ActualRows 也小
|
||||
// 但如果搭配父节点的 EstRows >> ActualRows 且父节点是 SCAN，说明扫描了 offset+N 行
|
||||
// 这里启发式：LIMIT 节点存在但 Extra 含 large offset 提示，或 ActualRows 显著小于 EstRows
|
||||
if node.Extra == nil {
|
||||
return nil
|
||||
}
|
||||
if v, ok := node.Extra["offset"]; ok {
|
||||
offset := parseExplainInt64(fmt.Sprintf("%v", v))
|
||||
if offset >= ruleLargeOffsetThreshold {
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityWarning,
|
||||
Rule: "large_offset_pagination",
|
||||
Reason: fmt.Sprintf("LIMIT offset=%d 过大，数据库需扫描并丢弃前 %d 行；建议改用游标分页（WHERE id > last_id LIMIT N）", offset, offset),
|
||||
AffectedNodeID: node.ID,
|
||||
EstRows: offset,
|
||||
}
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// ruleSelectStarWithJoin：检测 SELECT * + JOIN 模式（拉取不必要字段，放大网络/内存开销）。
|
||||
// 通过 SourceSQL 判断（节点级规则无法拿到 SQL，需要全局规则；此处用启发式：JOIN 节点 + 估算行数大）。
|
||||
// 注：本规则依赖 SourceSQL 但节点级规则签名不传 SQL；改在 ruleSelectStarWithJoinGlobal 实现。
|
||||
func ruleSelectStarWithJoin(_ connection.ExplainResult, node connection.ExplainNode) *connection.IndexSuggestion {
|
||||
// 启发式：JOIN 节点 + ActualRows 远大于 EstRows（说明 SELECT * 拉了大量数据）
|
||||
if node.OpType != connection.ExplainOpJoin {
|
||||
return nil
|
||||
}
|
||||
if node.EstRows <= 0 || node.ActualRows <= 0 {
|
||||
return nil
|
||||
}
|
||||
if node.ActualRows < node.EstRows*10 {
|
||||
return nil
|
||||
}
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityInfo,
|
||||
Rule: "select_star_with_join_pattern",
|
||||
Reason: "JOIN 节点实际行数远超估算，可能因 SELECT * 拉取了不必要字段；建议显式列出需要的列",
|
||||
AffectedNodeID: node.ID,
|
||||
AffectedTable: node.Table,
|
||||
EstRows: node.ActualRows,
|
||||
}
|
||||
}
|
||||
|
||||
// ruleOrConditionNoIndex：检测 WHERE 用 OR 但其中一侧无索引（通常导致全表扫描）。
|
||||
// 通过 filter 文本判断 "col1 = ? or col2 = ?" 模式。
|
||||
func ruleOrConditionNoIndex(_ connection.ExplainResult, node connection.ExplainNode) *connection.IndexSuggestion {
|
||||
if !hasFlag(node.Flags, connection.ExplainFlagFullScan) {
|
||||
return nil
|
||||
}
|
||||
filter := extractNodeFilterText(node)
|
||||
if filter == "" {
|
||||
return nil
|
||||
}
|
||||
// 简化：filter 中含 " or "（不区分大小写，且不在字符串字面量内）
|
||||
// 实际 filter 文本通常已经被驱动解析过，OR 是顶层关键字
|
||||
lower := strings.ToLower(filter)
|
||||
if !containsTopLevelKeyword(lower, " or ") {
|
||||
return nil
|
||||
}
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityWarning,
|
||||
Rule: "or_condition_no_index",
|
||||
Reason: "WHERE 含 OR 条件，若两侧字段未全部建索引则触发全表扫描；考虑改写为 UNION ALL 或为 OR 两侧字段都建索引",
|
||||
AffectedNodeID: node.ID,
|
||||
AffectedTable: node.Table,
|
||||
EstRows: node.EstRows,
|
||||
}
|
||||
}
|
||||
|
||||
// ruleCartesianProductRisk：全局规则，检测 JOIN 无 ON 条件（笛卡尔积）。
|
||||
// 判定：JOIN 节点 + Extra 中无 hashCond/joinType/on 等条件 + EstRows > 阈值。
|
||||
func ruleCartesianProductRisk(result connection.ExplainResult) *connection.IndexSuggestion {
|
||||
for _, node := range result.Nodes {
|
||||
if node.OpType != connection.ExplainOpJoin {
|
||||
continue
|
||||
}
|
||||
if node.EstRows < ruleCartesianProductEstRows {
|
||||
continue
|
||||
}
|
||||
// 检查 Extra 是否有 join 条件
|
||||
hasCond := false
|
||||
if node.Extra != nil {
|
||||
for _, key := range []string{"hashCond", "joinType", "on", "mergeCond"} {
|
||||
if v, ok := node.Extra[key]; ok && v != nil && fmt.Sprintf("%v", v) != "" {
|
||||
hasCond = true
|
||||
break
|
||||
}
|
||||
}
|
||||
}
|
||||
if hasCond {
|
||||
continue
|
||||
}
|
||||
return &connection.IndexSuggestion{
|
||||
Severity: connection.SeverityCritical,
|
||||
Rule: "cartesian_product_risk",
|
||||
Reason: fmt.Sprintf("JOIN 节点估算 %d 行且未识别到 ON/HASH 条件，可能是笛卡尔积；请补充 JOIN 条件", node.EstRows),
|
||||
AffectedNodeID: node.ID,
|
||||
EstRows: node.EstRows,
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// containsTopLevelKeyword 简化判断 keyword 是否在 text 中（不做嵌套括号分析，仅做大小写归一后子串匹配）。
|
||||
// 用于 OR 关键字检测；若需要更精确可在后续迭代增强。
|
||||
func containsTopLevelKeyword(text, keyword string) bool {
|
||||
return strings.Contains(text, keyword)
|
||||
}
|
||||
|
||||
@@ -234,6 +234,168 @@ func TestRunExplainRules_EmptyResultNoSuggestions(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// === 扩展规则测试 ===
|
||||
|
||||
func TestRunExplainRules_LikeLeadingWildcardCritical(t *testing.T) {
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM users WHERE name LIKE '%john%'",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpScan,
|
||||
Table: "users",
|
||||
EstRows: 50000,
|
||||
Flags: []string{connection.ExplainFlagFullScan, connection.ExplainFlagNoIndex},
|
||||
Extra: map[string]any{"attachedCondition": "name like '%john%'"},
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
found := false
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "like_leading_wildcard" {
|
||||
found = true
|
||||
if s.Severity != connection.SeverityCritical {
|
||||
t.Fatalf("LIKE 前缀通配应为 critical，got=%s", s.Severity)
|
||||
}
|
||||
}
|
||||
}
|
||||
if !found {
|
||||
t.Fatal("LIKE 前缀通配应触发 like_leading_wildcard 规则")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRunExplainRules_FunctionOnColumnCritical(t *testing.T) {
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM users WHERE UPPER(name) = 'JOHN'",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpScan,
|
||||
Table: "users",
|
||||
EstRows: 20000,
|
||||
Flags: []string{connection.ExplainFlagFullScan, connection.ExplainFlagNoIndex},
|
||||
Extra: map[string]any{"attachedCondition": "upper(name) = 'JOHN'"},
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
found := false
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "function_on_column" {
|
||||
found = true
|
||||
if s.Severity != connection.SeverityCritical {
|
||||
t.Fatalf("函数包裹列应为 critical，got=%s", s.Severity)
|
||||
}
|
||||
}
|
||||
}
|
||||
if !found {
|
||||
t.Fatal("函数包裹列应触发 function_on_column 规则")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRunExplainRules_OrConditionNoIndexWarning(t *testing.T) {
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM users WHERE id = 1 OR name = 'x'",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpScan,
|
||||
Table: "users",
|
||||
EstRows: 10000,
|
||||
Flags: []string{connection.ExplainFlagFullScan, connection.ExplainFlagNoIndex},
|
||||
Extra: map[string]any{"attachedCondition": "id = 1 or name = 'x'"},
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
found := false
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "or_condition_no_index" {
|
||||
found = true
|
||||
}
|
||||
}
|
||||
if !found {
|
||||
t.Fatal("全表扫描 + OR 条件应触发 or_condition_no_index 规则")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRunExplainRules_CartesianProductRiskCritical(t *testing.T) {
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM a, b",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpJoin,
|
||||
EstRows: 500000, // 远超阈值 100000
|
||||
Extra: map[string]any{}, // 无 hashCond/joinType
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
found := false
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "cartesian_product_risk" {
|
||||
found = true
|
||||
if s.Severity != connection.SeverityCritical {
|
||||
t.Fatalf("笛卡尔积风险应为 critical，got=%s", s.Severity)
|
||||
}
|
||||
}
|
||||
}
|
||||
if !found {
|
||||
t.Fatal("无条件的 JOIN + 大估算应触发 cartesian_product_risk")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRunExplainRules_CartesianProductSuppressedWithCondition(t *testing.T) {
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM a JOIN b ON a.id = b.aid",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpJoin,
|
||||
EstRows: 500000,
|
||||
Extra: map[string]any{"hashCond": "a.id = b.aid"}, // 有条件
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "cartesian_product_risk" {
|
||||
t.Fatal("有 JOIN 条件时不应触发 cartesian_product_risk")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestRunExplainRules_FunctionOnColumnNotTriggeredForPlainColumn(t *testing.T) {
|
||||
// WHERE name = 'x' 不应触发 function_on_column
|
||||
result := connection.ExplainResult{
|
||||
DBType: "mysql",
|
||||
SourceSQL: "SELECT * FROM users WHERE name = 'x'",
|
||||
Nodes: []connection.ExplainNode{
|
||||
{
|
||||
ID: "n1",
|
||||
OpType: connection.ExplainOpScan,
|
||||
Table: "users",
|
||||
EstRows: 100,
|
||||
Flags: []string{connection.ExplainFlagFullScan, connection.ExplainFlagNoIndex},
|
||||
Extra: map[string]any{"attachedCondition": "name = 'x'"},
|
||||
},
|
||||
},
|
||||
}
|
||||
suggestions := runExplainRules(result)
|
||||
for _, s := range suggestions {
|
||||
if s.Rule == "function_on_column" {
|
||||
t.Fatalf("name = 'x' 不应触发 function_on_column，但触发了：%+v", s)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// contains 检查字符串包含（避免和 strings.Contains 冲突，这里独立实现）。
|
||||
func contains(s, substr string) bool {
|
||||
return len(s) >= len(substr) && (s == substr || indexOfContains(s, substr) >= 0)
|
||||
|
||||
Reference in New Issue
Block a user