AskaEth
6e28b38b08
【MT-A：护栏 + 诚实截断 + 试算异步】
- maxTidyCompareItems 400(写死)→2000 并开放为 MEMORY_TIDY_COMPARE_LIMIT；
MEMORY_TIDY_CANDIDATE_LIMIT 默认 20→200（语义不变，仍每分区）
- 依据是首跑实测：459 条的分区被 400 截断 ⇒ compared(705) < scanned(784)，
候选数(77)由上限而不是规则决定；三处上限的 WARN 都补齐了总数/上限/分区名
- 新增「上限 + 时间预算」双闸（实测 2000 条/分区 ≈234s > 3 分钟墙钟）：
整分区跳过 / 两两比对中途收手，一律如实记 (budget) + WARN
- ?async=1 ⇒ 202 {runId,status}；GET …/{runId} 轮询（writeAuth 同档）；
done 的 result 与同步 data 逐字节同形；并发=复用既有 runId、TTL 30min 自完成起算、
重启即如实 404 tidy_run_not_found；不带 async 时同步路径逐字不变
【MT-B：D3 入库前审查（design/23 §15.1）】
- 新记忆经 Store.Add（仓库里唯一 INSERT INTO memory_items）落库后，由独立记忆管理
会话（sess-memorytidy）在有界后台队列异步判定 skip/merge/modify/create
- 软合并回写（status + merged_into/merged_from/merge_count，可 revert、绝不改 content）
- 隔离来源（群聊/陌生人）**概不审查**、召回/超时失败**零改写**（fail-safe 结构性成立）
- 写入侧仅 +62.7µs/条（非阻塞入队）；默认 ENABLED=true（主人裁定：入库前先审查）
【耦合说明（重要）】两片同包且互相依赖：MT-B 的 preinsert_recall.go 用 MT-A 的
collectTidyPairs 新签名 + tidyBudget ⇒ **各自单独都 build 不过**，故一并落地。
MT-A 为让整包编译，最小幅度适配了 MT-B 的 preinsert_recall.go / preinsert_test.go
（零值预算=行为不变；符号重名改名）。
【我的验收】隔离 worktree（37 文件）build/vet/linux/darwin 全过，切片包全绿；
我亲手打红两个变异：① 去掉 (compare) 截断标记 ⇒ TestTidyCompareLimit_HitIsLoud 红；
② 中和隔离来源守卫 ⇒ TestReviewRestrictedSourceIsNeverTouched 红。
唯一红是 internal/runtime/tool 的两条超时用例（该包无人改）⇒ 主仓库 -count=8 全绿，
判定为负载敏感 flake（当时两个整树验收 + 子代理并发），已记入开发习惯与约定 §7。
.env.example 补 MEMORY_REVIEW_* 四键（MT-B 无该文件权限，由我按其代码语义补写）。
280 lines
11 KiB
Go
280 lines
11 KiB
Go
package memory
|
||
|
||
import (
|
||
"context"
|
||
"encoding/json"
|
||
"fmt"
|
||
"strings"
|
||
"time"
|
||
)
|
||
|
||
// 本文件是记忆整理的存储层：分区扫描、候选落库与查询、运行留痕。它复用
|
||
// memory 包既有的 *sql.DB 与格式化辅助（formatTime/parseTime/scanItem），
|
||
// 不新建存储机制。
|
||
//
|
||
// 第一版唯一的写入就是候选表与运行表：**没有任何一条 memory_items 会被改写
|
||
// 或删除**（design/23 §3 非目标 + §7「apply 才写记忆」）。
|
||
|
||
// tidyCandidatePending 是候选的初始状态（design/23 §10）。
|
||
const tidyCandidatePending = "pending"
|
||
|
||
// tidyPartition 是整理的隔离单元：design/23 §6 的分区键
|
||
// (persona_id, scope, project_id)。分区之间**永不比较**。
|
||
type tidyPartition struct {
|
||
personaID string
|
||
scope Scope
|
||
projectID string
|
||
}
|
||
|
||
// label 渲染分区标签，供日志与截断登记使用（不含任何记忆正文）。
|
||
func (p tidyPartition) label() string {
|
||
return fmt.Sprintf("%s/%s/%s", p.personaID, p.scope, p.projectID)
|
||
}
|
||
|
||
// tidyStatusActive 是「在册」状态。整理只扫描在册条目：已合并/已归档的行
|
||
// 不再参与比对（design §10 status 列）。
|
||
const tidyStatusActive = "active"
|
||
|
||
// listTidyPartitionItemsSQL 按分区读取**在册**（status='active'）条目，
|
||
// 最新在前；rowid 打破 created_at 并列，使同一刻写入的条目顺序稳定
|
||
// （与 ListByPersona 同一口径）。分区过滤写在 SQL 里，而不是读全表再在内存
|
||
// 里筛 —— design/23 §6 要求在分区切分层就隔断。
|
||
const listTidyPartitionItemsSQL = `SELECT id, persona_id, role_id, project_id, scope, kind,
|
||
content, tags, confidence, source_conversation, created_at, updated_at,
|
||
decay_weight FROM memory_items
|
||
WHERE persona_id = ? AND scope = ? AND project_id = ? AND status = ?
|
||
ORDER BY created_at DESC, rowid DESC LIMIT ?`
|
||
|
||
const listTidyPartitionsSQL = `SELECT DISTINCT persona_id, scope, project_id
|
||
FROM memory_items WHERE status = ?
|
||
ORDER BY persona_id, scope, project_id`
|
||
|
||
const countTidyPartitionItemsSQL = `SELECT COUNT(*) FROM memory_items
|
||
WHERE persona_id = ? AND scope = ? AND project_id = ? AND status = ?`
|
||
|
||
const insertTidyCandidateSQL = `INSERT OR IGNORE INTO memory_tidy_candidates
|
||
(id, persona_id, scope, project_id, kind, action, source_ids, target_id,
|
||
proposed_content, merged_content, reason, evidence, rule_version,
|
||
score_min, score_max, status, idempotency_key, created_at, decided_at, decided_by)
|
||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, '', '')`
|
||
|
||
const insertTidyRunSQL = `INSERT INTO memory_tidy_runs
|
||
(id, started_at, finished_at, scanned, candidates, applied, skipped, errors,
|
||
vetoed, veto_reasons, rule_version)
|
||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`
|
||
|
||
const listTidyCandidatesSQL = `SELECT id, persona_id, scope, project_id, kind, action,
|
||
source_ids, target_id, proposed_content, merged_content, reason, evidence,
|
||
rule_version, score_min, score_max, status, idempotency_key, created_at
|
||
FROM memory_tidy_candidates
|
||
WHERE status = ?
|
||
ORDER BY created_at DESC, rowid DESC LIMIT ?`
|
||
|
||
const listTidyRunsSQL = `SELECT id, started_at, finished_at, scanned, candidates,
|
||
applied, skipped, errors, vetoed, veto_reasons, rule_version
|
||
FROM memory_tidy_runs ORDER BY started_at DESC, rowid DESC LIMIT ?`
|
||
|
||
// listTidyPartitions 返回所有含在册条目的分区。DISTINCT 让每个分区的
|
||
// 扫描上限（design §9 每分区 5000 条）各自独立。
|
||
func (s *Store) listTidyPartitions(ctx context.Context) ([]tidyPartition, error) {
|
||
rows, err := s.db.QueryContext(ctx, listTidyPartitionsSQL, string(tidyStatusActive))
|
||
if err != nil {
|
||
return nil, fmt.Errorf("memory: list tidy partitions: %w", err)
|
||
}
|
||
defer func() { _ = rows.Close() }()
|
||
|
||
parts := make([]tidyPartition, 0)
|
||
for rows.Next() {
|
||
var p tidyPartition
|
||
var scope string
|
||
if err := rows.Scan(&p.personaID, &scope, &p.projectID); err != nil {
|
||
return nil, fmt.Errorf("memory: scan tidy partition: %w", err)
|
||
}
|
||
p.scope = Scope(scope)
|
||
parts = append(parts, p)
|
||
}
|
||
if err := rows.Err(); err != nil {
|
||
return nil, fmt.Errorf("memory: iterate tidy partitions: %w", err)
|
||
}
|
||
return parts, nil
|
||
}
|
||
|
||
// listTidyPartitionItems 读取一个分区内的在册条目（最新在前，最多 limit 条）。
|
||
func (s *Store) listTidyPartitionItems(ctx context.Context, p tidyPartition, limit int) ([]MemoryItem, error) {
|
||
rows, err := s.db.QueryContext(ctx, listTidyPartitionItemsSQL,
|
||
p.personaID, string(p.scope), p.projectID, string(tidyStatusActive), limit)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("memory: list tidy partition items: %w", err)
|
||
}
|
||
return scanItems(rows, "tidy partition item")
|
||
}
|
||
|
||
// countTidyPartitionItems 数一个分区里在册条目的**真总数**。只在扫描上限命中时
|
||
// 调用一次：日志里的「总数」必须是真数，而不是被 LIMIT 截出来的 len(items)
|
||
// （那是**下限**，把「至少 5000」写成「正好 5000」就是一句谎）。
|
||
func (s *Store) countTidyPartitionItems(ctx context.Context, p tidyPartition) (int, error) {
|
||
var total int
|
||
if err := s.db.QueryRowContext(ctx, countTidyPartitionItemsSQL,
|
||
p.personaID, string(p.scope), p.projectID, string(tidyStatusActive)).Scan(&total); err != nil {
|
||
return 0, fmt.Errorf("memory: count tidy partition items: %w", err)
|
||
}
|
||
return total, nil
|
||
}
|
||
|
||
// insertTidyCandidate 落库一条候选，返回它是否是**新建**的。幂等键
|
||
// （design/23 §9）唯一：同一批数据重复运行命中已存在行时返回 (false, nil)，
|
||
// 绝不重复产候选。
|
||
func (s *Store) insertTidyCandidate(ctx context.Context, c TidyCandidate) (bool, error) {
|
||
c.IdempotencyKey = tidyIdempotencyKey(c)
|
||
sourceJSON, err := json.Marshal(c.SourceIDs)
|
||
if err != nil {
|
||
return false, fmt.Errorf("memory: encode tidy source ids: %w", err)
|
||
}
|
||
evidenceJSON, err := json.Marshal(c.Evidence)
|
||
if err != nil {
|
||
return false, fmt.Errorf("memory: encode tidy evidence: %w", err)
|
||
}
|
||
if c.CreatedAt.IsZero() {
|
||
c.CreatedAt = time.Now().UTC()
|
||
}
|
||
res, err := s.db.ExecContext(ctx, insertTidyCandidateSQL,
|
||
c.ID, c.PersonaID, string(c.Scope), c.ProjectID, string(c.Kind), string(c.Action),
|
||
string(sourceJSON), c.TargetID, c.ProposedContent, c.MergedContent, c.Reason,
|
||
string(evidenceJSON), c.RuleVersion, c.ScoreMin, c.ScoreMax, c.Status,
|
||
c.IdempotencyKey, formatTime(c.CreatedAt))
|
||
if err != nil {
|
||
return false, fmt.Errorf("memory: insert tidy candidate: %w", err)
|
||
}
|
||
n, err := res.RowsAffected()
|
||
if err != nil {
|
||
return false, fmt.Errorf("memory: tidy candidate rows affected: %w", err)
|
||
}
|
||
return n > 0, nil
|
||
}
|
||
|
||
// persistTidyRun 写一条运行留痕（design/23 §10 memory_tidy_runs）。
|
||
func (s *Store) persistTidyRun(ctx context.Context, run TidyRun) error {
|
||
if run.ID == "" {
|
||
run.ID = mustNewTidyID()
|
||
}
|
||
reasons, err := json.Marshal(run.VetoReasons)
|
||
if err != nil {
|
||
return fmt.Errorf("memory: encode tidy veto reasons: %w", err)
|
||
}
|
||
if _, err := s.db.ExecContext(ctx, insertTidyRunSQL,
|
||
run.ID, formatTime(run.StartedAt), formatTime(run.FinishedAt),
|
||
run.Scanned, run.Candidates, run.Applied, run.Skipped, run.Errors,
|
||
run.Vetoed, string(reasons), run.RuleVersion); err != nil {
|
||
return fmt.Errorf("memory: insert tidy run: %w", err)
|
||
}
|
||
return nil
|
||
}
|
||
|
||
// ListTidyCandidates 返回指定状态的整理候选（最新在前，最多 limit 条，
|
||
// 非正 limit 取 defaultTidyCandidateLimit）。它是读侧入口：候选正文含隐私，
|
||
// 只应由 owner-only 的读侧使用（design/23 §11）。
|
||
func (s *Store) ListTidyCandidates(ctx context.Context, status string, limit int) ([]TidyCandidate, error) {
|
||
if strings.TrimSpace(status) == "" {
|
||
status = tidyCandidatePending
|
||
}
|
||
if limit <= 0 {
|
||
limit = defaultTidyCandidateLimit
|
||
}
|
||
rows, err := s.db.QueryContext(ctx, listTidyCandidatesSQL, status, limit)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("memory: list tidy candidates: %w", err)
|
||
}
|
||
defer func() { _ = rows.Close() }()
|
||
|
||
out := make([]TidyCandidate, 0)
|
||
for rows.Next() {
|
||
c, err := scanTidyCandidate(rows)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("memory: scan tidy candidate: %w", err)
|
||
}
|
||
out = append(out, c)
|
||
}
|
||
if err := rows.Err(); err != nil {
|
||
return nil, fmt.Errorf("memory: iterate tidy candidates: %w", err)
|
||
}
|
||
return out, nil
|
||
}
|
||
|
||
// ListTidyRuns 返回整理运行留痕（最新在前，最多 limit 条）。非正 limit 取
|
||
// 一个保守的默认条数。
|
||
func (s *Store) ListTidyRuns(ctx context.Context, limit int) ([]TidyRun, error) {
|
||
if limit <= 0 {
|
||
limit = defaultTidyCandidateLimit
|
||
}
|
||
rows, err := s.db.QueryContext(ctx, listTidyRunsSQL, limit)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("memory: list tidy runs: %w", err)
|
||
}
|
||
defer func() { _ = rows.Close() }()
|
||
|
||
out := make([]TidyRun, 0)
|
||
for rows.Next() {
|
||
var (
|
||
r TidyRun
|
||
startedAt string
|
||
finishedAt string
|
||
reasonsJSON string
|
||
)
|
||
if err := rows.Scan(&r.ID, &startedAt, &finishedAt, &r.Scanned, &r.Candidates,
|
||
&r.Applied, &r.Skipped, &r.Errors, &r.Vetoed, &reasonsJSON, &r.RuleVersion); err != nil {
|
||
return nil, fmt.Errorf("memory: scan tidy run: %w", err)
|
||
}
|
||
var err error
|
||
if r.StartedAt, err = parseTime(startedAt); err != nil {
|
||
return nil, err
|
||
}
|
||
if r.FinishedAt, err = parseTime(finishedAt); err != nil {
|
||
return nil, err
|
||
}
|
||
if err := json.Unmarshal([]byte(reasonsJSON), &r.VetoReasons); err != nil {
|
||
return nil, fmt.Errorf("memory: decode tidy veto reasons: %w", err)
|
||
}
|
||
out = append(out, r)
|
||
}
|
||
if err := rows.Err(); err != nil {
|
||
return nil, fmt.Errorf("memory: iterate tidy runs: %w", err)
|
||
}
|
||
return out, nil
|
||
}
|
||
|
||
// scanTidyCandidate 读取一行候选：source_ids 与 evidence 以 JSON 存储。
|
||
func scanTidyCandidate(sc rowScanner) (TidyCandidate, error) {
|
||
var (
|
||
c TidyCandidate
|
||
scope, kind string
|
||
action string
|
||
sourceJSON string
|
||
evidenceRaw string
|
||
createdAt string
|
||
)
|
||
if err := sc.Scan(&c.ID, &c.PersonaID, &scope, &c.ProjectID, &kind, &action,
|
||
&sourceJSON, &c.TargetID, &c.ProposedContent, &c.MergedContent, &c.Reason,
|
||
&evidenceRaw, &c.RuleVersion, &c.ScoreMin, &c.ScoreMax, &c.Status,
|
||
&c.IdempotencyKey, &createdAt); err != nil {
|
||
return TidyCandidate{}, err
|
||
}
|
||
c.Scope = Scope(scope)
|
||
c.Kind = Kind(kind)
|
||
c.Action = TidyAction(action)
|
||
if err := json.Unmarshal([]byte(sourceJSON), &c.SourceIDs); err != nil {
|
||
return TidyCandidate{}, fmt.Errorf("memory: decode tidy source ids: %w", err)
|
||
}
|
||
if c.SourceIDs == nil {
|
||
c.SourceIDs = []string{}
|
||
}
|
||
if err := json.Unmarshal([]byte(evidenceRaw), &c.Evidence); err != nil {
|
||
return TidyCandidate{}, fmt.Errorf("memory: decode tidy evidence: %w", err)
|
||
}
|
||
t, err := parseTime(createdAt)
|
||
if err != nil {
|
||
return TidyCandidate{}, err
|
||
}
|
||
c.CreatedAt = t
|
||
return c, nil
|
||
}
|