tests: mysql-cdc: test add sub source after corruption

[nrainer-materialize]

Create a source, purge the bin log, add a sub source?

This adds testing for #27507.

[nrainer-materialize]

@petrosagg, is this what we discussed? Should this be working?

SELECT * FROM t3 produces

ERROR:  Source error: u4: mysql server binlog frontier at Antichain { elements: [Partitioned((Interval { lower: 00000000-0000-0000-0000-000000000000, upper: 0a6ab578-2faf-11ef-ac57-0242c0a83002 }, Absent)), Partitioned((Interval { lower: 0a6ab578-2faf-11ef-ac57-0242c0a83003, upper: 0a6ab578-2faf-11ef-ac57-0242c0a83003 }, Active(28))), Partitioned((Interval { lower: 0a6ab578-2faf-11ef-ac57-0242c0a83004, upper: ffffffff-ffff-ffff-ffff-ffffffffffff }, Absent))] } is beyond required frontier Antichain { elements: [Partitioned((Interval { lower: 00000000-0000-0000-0000-000000000000, upper: 0a6ab578-2faf-11ef-ac57-0242c0a83002 }, Absent)), Partitioned((Interval { lower: 0a6ab578-2faf-11ef-ac57-0242c0a83003, upper: 0a6ab578-2faf-11ef-ac57-0242c0a83003 }, Active(27))), Partitioned((Interval { lower: 0a6ab578-2faf-11ef-ac57-0242c0a83004, upper: ffffffff-ffff-ffff-ffff-ffffffffffff }, Absent))] }

[rjobanp]

@nrainer-materialize is this working or failing with that error you posted? The CI checks seem to be green on this.

What I think is happening in the failure case is that the source doesn't actually finish the snapshot of the first tables before materialize is killed and the binlog wiped. Then when it's restarted it can't resume from a point that the first subsources needed. I think we should add a SELECT * FROM t1 before we kill MZ and wipe the binlog

[nrainer-materialize]

@nrainer-materialize is this working or failing with that error you posted? The CI checks seem to be green on this.

I guess it was failing locally.

What I think is happening in the failure case is that the source doesn't actually finish the snapshot of the first tables before materialize is killed and the binlog wiped. Then when it's restarted it can't resume from a point that the first subsources needed. I think we should add a SELECT * FROM t1 before we kill MZ and wipe the binlog

Will do!