the margin of full and incremental replication

[actiontech-bot]

Is your feature request related to a problem? Please describe. No

Describe the solution you'd like Need a solution to handle margin of full and incremental replication

Describe alternatives you've considered None

---

Phase 1

commit: "Keep polling until get a consistent TX with GTID":

1. Get GTID (G1)
2. start transaction with consistent snapshot
3. Get GTID (G2)
4. continue if G1 = G2，otherwise retry

Phase 2

remove start transaction with consistent snapshot

It's conflict with homogeneous replication, so ignored.

Phase 3

full-replication and incremental-replication start at the same time

[ghost]

5865a97a before：全量在snapshot TX中提取所有表的数据 after：全量在snapshot TX中提取无PK表的数据，随后commit，在TX外提取其余表的数据

2a2cb5da：将所有update分为delete+insert执行

[ghost]

问题：

update->delete+insert, 导致FK on delete cascade 被误执行。

[ghost]

• dtle现有机制: 支持同步外键, 支持外键级联更新.
• 某DTS机制: “在全量同步和增量同步过程中，DTS会以Session级别暂时禁用约束检查以及外键级联操作。 若任务运行时源库存在级联更新、删除操作，可能会导致数据不一致。”

问题: 是否将dtle变更为和某DTS一样的机制?

FK级联同步与将要进行的变更有冲突, 原因如下.

下列机制

• A 全量不在consistent snapshot中执行
• B update分为delete+insert
• C 全量断点续传(待实现)
• D 已有job增加表, 对新增表进行全量(待实现)

其中

• A依赖B (否则引起数据不一致 gh-ost#519)
• C依赖A
• D依赖A
• B在有FK delete cascade时, 引起问题(FK子表的级联update变为delete)