debezium/dbz#1879 Add SMT EnforceRecordSize

[twthorn]

debezium/dbz#1879

Add an SMT for enforcing record size with the current strategy of proportional column truncation.

We still need to performance test this. Some optimizations with caching may be needed.

We may support other strategies in the future, one such is payload offloading debezium/dbz#1949

[Naros]

I wonder whether there should be more surgical algorithms, such as largest-first or greedy. I also wonder if the SMT should explicitly prefer a list of field names.

Let's say we have an event with 10 fields, 9 of which are 10 bytes each, and one is 5000 bytes, for a total of 5090 bytes. If we assume that the event's size exceeds a 2000-byte threshold, all 9 smaller fields are dropped to 6 bytes, which I think risks making the entire event potentially irrelevant.

Let's wait and see how Jiri views the PR behavior; maybe he sees something I am glossing over.

[jpechane]

@twthorn Hi, thanks for the PR! Two comments

• Is this something that is Vitess specific? If not then the SMT could/should reside in the core repository
• Could you please check ApproximateStructSizeCalculator? Is there is opportunity to re-use the class or extend it so it is usable with the SMT too?