knowledge-governance/internal/platform/postgres/test_publication_removal.go

package postgres

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"strings"
	"time"

	"github.com/jackc/pgx/v5"
)

// ErrPublicationRemovalDone：该发布单元的检索已经移除过（路由映射 409；
// 条目不存在复用既有的 ErrPublicationNotFound，状态不允许复用 ErrPublicationStateConflict）。
var ErrPublicationRemovalDone = errors.New("test publication retrieval already removed")

// StartTestPublicationRemoval 让运维把一个「被禁止导入、但仍存活在测试库」的知识从测试库里移除：
// 把该发布单元置为 disabling 并入队既有的 test.remove 工作流（该工作流会调 WeKnora 删除知识并读回校验，
// 只有它能把状态推进到 disabled/failed）。
//
// 为什么要有这条入口：禁止导入本身只撤销待入库条目并置 policy_state='blocked'，测试库里已经建好的知识
// 不会自动消失；此前只有"变更集删除路径"会入队 test.remove，界面既列不出条目也没有按钮，运维无法闭环。
//
// 幂等：按 publication 固定 idempotency_key（ON CONFLICT DO NOTHING）。重复调用（含并发与用户重试）
// 只会有一个任务；已在 disabling 时直接返回既有任务 id；已 disabled 时报 ErrPublicationRemovalDone。
func (s *Store) StartTestPublicationRemoval(ctx context.Context, publicationID string, now time.Time) (string, error) {
	publicationID = strings.TrimSpace(publicationID)
	if publicationID == "" || len(publicationID) > 128 || strings.ContainsAny(publicationID, "\x00\r\n") {
		return "", ErrPublicationNotFound
	}
	beginner, ok := s.db.(interface {
		Begin(context.Context) (pgx.Tx, error)
	})
	if !ok {
		return "", errors.New("test publication removal transactions are unavailable")
	}
	tx, err := beginner.Begin(ctx)
	if err != nil {
		return "", fmt.Errorf("begin test publication removal: %w", err)
	}
	defer tx.Rollback(ctx)

	var retrievalState, knowledgeID, processingState string
	var superseded bool
	err = tx.QueryRow(ctx, `SELECT retrieval_state, knowledge_id, processing_state, superseded_at IS NOT NULL
FROM test_publications WHERE id=$1 FOR UPDATE`, publicationID).Scan(&retrievalState, &knowledgeID, &processingState, &superseded)
	if errors.Is(err, pgx.ErrNoRows) {
		return "", ErrPublicationNotFound
	}
	if err != nil {
		return "", fmt.Errorf("load test publication for removal: %w", err)
	}
	key := "operator-remove:" + publicationID
	if retrievalState == "disabled" {
		return "", ErrPublicationRemovalDone
	}
	if retrievalState == "disabling" {
		// 已经在移除中：返回既有任务，不重复入队。
		var existingJobID string
		if err := tx.QueryRow(ctx, `SELECT id FROM jobs WHERE idempotency_key=$1`, key).Scan(&existingJobID); err == nil {
			if err := tx.Commit(ctx); err != nil {
				return "", fmt.Errorf("commit test publication removal: %w", err)
			}
			return existingJobID, nil
		}
	}
	if superseded || knowledgeID == "" || processingState != "completed" {
		// 只处理"测试库里确实活着"的条目：没上传过的走撤销，不进这里。
		return "", ErrPublicationStateConflict
	}
	tag, err := tx.Exec(ctx, `UPDATE test_publications SET retrieval_state='disabling', last_error='', updated_at=$2 WHERE id=$1`, publicationID, now)
	if err != nil {
		return "", fmt.Errorf("mark test publication removal: %w", err)
	}
	if tag.RowsAffected() != 1 {
		return "", ErrPublicationStateConflict
	}
	jobID, err := newOpaqueID("job-")
	if err != nil {
		return "", err
	}
	payload, err := json.Marshal(map[string]string{"publication_id": publicationID})
	if err != nil {
		return "", errors.New("encode test publication removal job")
	}
	// business_* 三列必须写 NULL，不能写空串：jobs 上有部分唯一索引
	//   UNIQUE (business_change_set_id, business_version_id, business_target_kb_id, business_action)
	//   WHERE business_change_set_id IS NOT NULL
	// 变更集路径写的是真实 id，空串则是"非 NULL"，于是 ('','','','remove') 对**所有**运维移除都是同一个键，
	// 第一条之后每次插入都撞唯一索引；而 ON CONFLICT DO NOTHING 会把它静默吞掉，留下"状态已 disabling、
	// 却没有任务"的永久卡死条目（车道还看不见它）。NULL 表达"没有变更集身份"，索引因此不适用，幂等只由
	// 下面的 idempotency_key 决定。
	tag, err = tx.Exec(ctx, `INSERT INTO jobs(id,type,payload,state,priority,available_at,max_attempts,idempotency_key,created_at,updated_at,business_change_set_id,business_version_id,business_target_kb_id,business_action)
VALUES($1,'test.remove',$2,'pending',20,$3,10,$4,$3,$3,NULL,NULL,NULL,'remove')
ON CONFLICT (idempotency_key) DO NOTHING`, jobID, payload, now, key)
	if err != nil {
		return "", fmt.Errorf("enqueue test publication removal: %w", err)
	}
	if tag.RowsAffected() != 1 {
		// 键已存在（并发重复调用）：必须是同一个任务，不能留下没有任务的 disabling。
		var existingJobID string
		if err := tx.QueryRow(ctx, `SELECT id FROM jobs WHERE idempotency_key=$1`, key).Scan(&existingJobID); err != nil {
			return "", fmt.Errorf("test publication removal job was not enqueued: %w", err)
		}
		if err := tx.Commit(ctx); err != nil {
			return "", fmt.Errorf("commit test publication removal: %w", err)
		}
		return existingJobID, nil
	}
	if err := tx.Commit(ctx); err != nil {
		return "", fmt.Errorf("commit test publication removal: %w", err)
	}
	return jobID, nil
}

func mustJSONString(value string) string {
	var builder strings.Builder
	builder.WriteByte('"')
	for _, r := range value {
		switch r {
		case '"':
			builder.WriteString(`\"`)
		case '\\':
			builder.WriteString(`\\`)
		default:
			if r < 0x20 {
				builder.WriteString(fmt.Sprintf(`\u%04x`, r))
				continue
			}
			builder.WriteRune(r)
		}
	}
	builder.WriteByte('"')
	return builder.String()
}