v5.3.1 relay log may turn off after master reboot

[lance6716]

What did you do?

prerequisite: all conditions below must be satisfied

• v5.3.1
• user use start-relay -s source01 worker01 to start relay log for source01 on worker01
• source configuration file of source01 has enable-relay: true (in v5.3.1, this field does not take effect, start-relay command is the only way to start relay)
• master reboot or leader switch

What did you expect to see?

relay is still turned on for source01 on worker01

What did you see instead?

relay is turned off

Versions of the cluster

v5.3.1

current status of DM cluster (execute query-status <task-name> in dmctl)

(paste current status of DM cluster here)

[lance6716]

Root cause

during cherry-pick and resolve conflict for #4430 , we wrongly included the logic of v5.4 into v5.3, this logic is about cleaning etcd relay KV which causes the issue.

Walkaround

If user need relay log feature, he/she must not have enable-relay: true in source configuration file

[niubell]

enable-relay is depreacted from v2.0.2, so we don't need to fix this issue, right ?

[lance6716]

depreacted means it doesn't take effects, but

tiflow/dm/dm/master/scheduler/scheduler.go

Line 1515 in 641df5d

if sourceCfg.EnableRelay {

it can cause relay KV deleted.

User may upgrade from v2.0.1 and not turn off enable-relay

[lance6716]

closed by #4877