Reduce duplication of DM's syncer and CDC's MySQL sink

[lance6716]

noticed there're lot of functions can be reused:

(a finished item)

---

https://github.com/pingcap/ticdc/blob/4bc1853a10287f12aff5706123c0cfc39feae7d2/cdc/sink/mysql.go#L1203 vs https://github.com/pingcap/ticdc/blob/4bc1853a10287f12aff5706123c0cfc39feae7d2/dm/syncer/dml.go#L713

---

sink/causality.go vs syncer/causality.go

[fatelei]

can I have a try to make a pull request?

[lance6716]

can I have a try to make a pull request?

thanks! we’re looking forward to your contribution.

also please keep pr in a small size, that’s better for reviewing.

[fatelei]

can I have a try to make a pull request?

thanks! we’re looking forward to your contribution.

also please keep pr in a small size, that’s better for reviewing.

ok

[amyangfei]

Causality detection is also a common library used both in DM and TiCDC. ref: https://github.com/pingcap/ticdc/issues/3286

[lance6716]

some comparison of DML part of mysqlSink and Syncer

dml comes from:

• CDC: will periodically get resolved transaction from txnCache
• DM: will read the (DML) job from a channel, the sender of the channel is the binlog event loop

processing stages:

• CDC: casuality -> generate DML -> batch a SQL transactoin
• DM: compact (optional) -> casuality -> generate DML -> batch a SQL transactoin

compact

• DM: for frequently updated row, only replicate the latest data

generate DML:

• CDC: only generate DELETE and REPLACE per row
• DM: will generate many types including INSERT ON DUPLICATE UPDATE, and wil retry to merge successive rows of one table and one type into one DML, in other words, one DML with many row changes

batch a SQL transactoin

• CDC: when the batch is not full, minimal lag is ticker interval
• DM: when the batch is not full but channel is empty, directly execute the batch

cyclic replication

• CDC: when generate DML, a cyclic marker row is also generated

cc @amyangfei @GMHDBJD