cdc cli changefeed create --start-ts does not work properly and may lead to data loss

[overvenus]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. What did you do? If possible, provide a recipe for reproducing the error.

TiCDC cli creates a changefeed that does not use the start-ts specified in command line. It leads to data loss.

Note that the start-ts in command line is invalid and it points to 1975-02-25 15:14:19.597000.

mysql> select tidb_parse_tso(426318359552065537); # start-ts in cli output

+------------------------------------+
| tidb_parse_tso(426318359552065537) |
+------------------------------------+
| 2021-07-14 23:11:20.469000         |
+------------------------------------+

mysql> select tidb_parse_tso(42610054816776198); # start-ts in command line

+-----------------------------------+
| tidb_parse_tso(42610054816776198) |
+-----------------------------------+
| 1975-02-25 15:14:19.597000        |
+-----------------------------------+

2. What did you expect to see?

a. create a changefeed with a start-ts specified in command line.
b. report warnings about invalid start-ts.

3. What did you see instead?

A wrong changefeed is created and it leads to data loss.

4. Versions of the cluster

v5.0.1

[amyangfei]

Strange, the start-ts should be checked against GC safepoint, what does invalid mean?

[overvenus]

Strange, the start-ts should be checked against GC safepoint, what does invalid means?

Sorry, I mean the ts is way behind TiDB gc safepoint.

[overvenus]

@liuzix finds that there is an unexpected character ‘c2 a0’ before "--start-ts”. It may cause this issue.

[asddongmen]

It seems that any char before a flag will cause this error.!

[Rustin170506]

We can use cobra.NoArgs to report this error explicitly.