Mysql到TiDB迁移，双写数据库兜底方案

作者：京东零售 石磊

TiDB 作为开源 NewSQL 数据库的典型代表之一，同样支持 SQL，支持事务 ACID 特性。在通讯协议上，TiDB 选择与 MySQL 完全兼容，并尽可能兼容 MySQL 的语法。因此，基于 MySQL 数据库开发的系统，大多数可以平滑迁移至 TiDB，而几乎不用修改代码。对用户来说，迁移成本极低，过渡自然。

然而，仍有一些 MySQL 的特性和行为，TiDB 目前暂时不支持或表现与 MySQL 有差异。除此之外，TiDB 提供了一些扩展语法和功能，为用户提供更多的便利。

TiDB 仍处在快速发展的道路上，对 MySQL 功能和行为的支持方面，正按 路线图 的规划在前行。

兼容策略

先从总体上概括 TiDB 和 MySQL 兼容策略，如下表：

截至 4.0 版本，TiDB 与 MySQL 的区别总结如下表：

三种方案比较

双写方案：同时往mysql和tidb写入数据，两个数据库数据完全保持同步

•优点：此方案最安全，作为兜底方案不需担心数据库回滚问题，因为数据完全一致，可以无缝回滚到mysql

•缺点：新方案，调研方案实现，成本较高

读写分离：数据写入mysql，从tidb读，具体方案是切换到线上以后，保持读写分离一周时间左右，这一周时间用来确定tidb数据库没有问题，再把写操作也切换到tidb

•优点: 切换过程，mysql和tidb数据保持同步，满足数据回滚到mysql方案

•缺点：mysql和tidb数据库同步存在延时，对部分写入数据要求实时查询的会导致查询失败，同时一旦整体切换到tidb，无法回切到mysql

直接切换：直接一步切换到tidb

•优点：切换过程最简单，成本最低

•缺点：此方案没有兜底方案，切换到tidb，无法再回切到mysql或者同步数据回mysql风险较大，无法保证数据是否可用

Django双写mysql与tidb策略

settings.py中新增配置

# Dev Database settings
DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.mysql',
        'NAME': 'name',
        'USER': 'root',
        'PASSWORD': '123456',
        'HOST': 'db',
    },
    'replica': {
        'ENGINE': 'django.db.backends.mysql',
        'NAME': 'name',
        'USER': 'root',
        'PASSWORD': '123456',
        'HOST': 'db',
    },
    'bak': {
        'ENGINE': 'django.db.backends.mysql',
        'NAME': 'name',
        'USER': 'root',
        'PASSWORD': '123456',
        'HOST': 'db',
    },
}
# 多重写入数据库配置
MULTI_WRITE_DB = "bak"

双写中间件 basemodel.py

import copy
import logging
import traceback
from django.db import models, transaction, router
from django.db.models.deletion import Collector
from django.db.models import sql
from django.db.models.sql.constants import CURSOR
from jcdp.settings import MULTI_WRITE_DB, DATABASES

multi_write_db = MULTI_WRITE_DB


# 重写QuerySet
class BaseQuerySet(models.QuerySet):

    def create(self, **kwargs):
        return super().create(**kwargs)

    def update(self, **kwargs):
        try:
            rows = super().update(**kwargs)
            if multi_write_db in DATABASES:
                self._for_write = True
                query = self.query.chain(sql.UpdateQuery)
                query.add_update_values(kwargs)
                with transaction.mark_for_rollback_on_error(using=multi_write_db):
                    query.get_compiler(multi_write_db).execute_sql(CURSOR)
        except Exception:
            logging.error(traceback.format_exc())
            raise
        return rows

    def delete(self):
        try:
            deleted, _rows_count = super().delete()
            if multi_write_db in DATABASES:
                del_query = self._chain()
                del_query._for_write = True
                del_query.query.select_for_update = False
                del_query.query.select_related = False
                collector = Collector(using=multi_write_db)
                collector.collect(del_query)
                collector.delete()
        except Exception:
            logging.error(traceback.format_exc())
            raise
        return deleted, _rows_count

    def raw(self, raw_query, params=None, translations=None, using=None):
        try:
            qs = super().raw(raw_query, params=params, translations=translations, using=using)
            if multi_write_db in DATABASES:
                super().raw(raw_query, params=params, translations=translations, using=multi_write_db)
        except Exception:
            logging.error(traceback.format_exc())
            raise
        return qs

    def bulk_create(self, objs, batch_size=None, ignore_conflicts=False):
        try:
            for obj in objs:
                obj.save()
        except Exception:
            logging.error(traceback.format_exc())
            raise
        # objs = super().bulk_create(objs, batch_size=batch_size, ignore_conflicts=ignore_conflicts)
        # if multi_write_db in DATABASES:
        #     self._db = multi_write_db
        #     super().bulk_create(objs, batch_size=batch_size, ignore_conflicts=ignore_conflicts)
        return objs

    def bulk_update(self, objs, fields, batch_size=None):
        try:
            super().bulk_update(objs, fields, batch_size=batch_size)
            if multi_write_db in DATABASES:
                self._db = multi_write_db
                super().bulk_update(objs, fields, batch_size=batch_size)
        except Exception:
            logging.error(traceback.format_exc())
            raise


class BaseManager(models.Manager):
    _queryset_class = BaseQuerySet


class BaseModel(models.Model):
    objects = BaseManager()

    class Meta:
        abstract = True

    def delete(
            self, using=None, *args, **kwargs
    ):
        try:
            instance = copy.deepcopy(self)
            super().delete(using=using, *args, **kwargs)
            if multi_write_db in DATABASES:
                super(BaseModel, instance).delete(using=multi_write_db, *args, **kwargs)
        except Exception:
            logging.error(traceback.format_exc())
            raise

    def save_base(self, raw=False, force_insert=False,
                  force_update=False, using=None, update_fields=None):
        try:
            using = using or router.db_for_write(self.__class__, instance=self)
            assert not (force_insert and (force_update or update_fields))
            assert update_fields is None or update_fields
            cls = self.__class__
            # Skip proxies, but keep the origin as the proxy model.
            if cls._meta.proxy:
                cls = cls._meta.concrete_model
            meta = cls._meta
            # A transaction isn't needed if one query is issued.
            if meta.parents:
                context_manager = transaction.atomic(using=using, savepoint=False)
            else:
                context_manager = transaction.mark_for_rollback_on_error(using=using)
            with context_manager:
                parent_inserted = False
                if not raw:
                    parent_inserted = self._save_parents(cls, using, update_fields)
                self._save_table(
                    raw, cls, force_insert or parent_inserted,
                    force_update, using, update_fields,
                )
            if multi_write_db in DATABASES:
                super().save_base(raw=raw,
                                  force_insert=raw,
                                  force_update=force_update,
                                  using=multi_write_db,
                                  update_fields=update_fields)
            # Store the database on which the object was saved
            self._state.db = using
            # Once saved, this is no longer a to-be-added instance.
            self._state.adding = False
        except Exception:
            logging.error(traceback.format_exc())
            raise

上述配置完成以后，在每个应用的models.py中引用新的BaseModel类作为模型基类即可实现双写目的

class DirectoryStructure(BaseModel):
    """
    目录结构
    """
    view = models.CharField(max_length=128, db_index=True)  # 视图名称 eg：部门视图 项目视图
    sub_view = models.CharField(max_length=128, unique=True, db_index=True)  # 子视图名称
    sub_view_num = models.IntegerField()  # 子视图顺序号

注：目前该方法尚不支持多对多模型的双写情景，如有业务需求，还需重写ManyToManyField类，方法参考猴子补丁方式

迁移数据库过程踩坑记录

TIDB配置项差异：确认数据库配置：ONLY_FULL_GROUP_BY 禁用 （mysql默认禁用）

TIDB不支持事务savepoint，代码中需要显式关闭savepoint=False

TIDB由于是分布式数据库，对于自增主键字段的自增策略与mysq有差异，若业务代码会与主键id关联，需要注意