DynamoDB: Use ON CONFLICT DO NOTHING clause on INSERT CDC operations

[amotl]

Reason

... to mitigate errors when events are relayed redundantly from retries after partially failed batches on the Lambda processor.

References

• DynamoDB: DuplicateKeyException after resuming CDC cratedb-toolkit#301

/cc @kneth, @surister, @hlcianfagna

[amotl]

Discussion Summary

Q&A

1. If you think this change will apply the correct solution to the corresponding problem, or if it would cause other flaws, if that option should be made configurable, or if it is sane to use as a default particularly on the CDC INSERT operations.
2. While already at it, do you think that clause should also be applied to other spots in the relevant subsystems of CTK? a) MongoDB CDC? b) Optionally also on full-load operations, both DynamoDB and MongoDB?

Do you think it is applicable then to use the upsert on all three occasions, which are
a) INSERT statements of full load operations,
b) INSERT statements of cdc operations,
c) UPDATE statements of cdc operations.

Do nothing on DuplicateKeyExceptions

@wierdvanderhaar already responded:

I think the upsert could work for all three. Only the Update needs to be transformed into an INSERT INTO .. ON CONFLICT DO UPDATE SET.

@hammerhead added:

Regarding c), there is still one semantical difference between an UPDATE and an INSERT ON CONFLICT: If there is no row matching the UPDATE, it will have no effect. If that statement gets re-written as an INSERT ON CONFLICT, the record gets inserted. Not sure if that is a problem or not, but at least it is a difference in behaviour. Personally, I would recommend to stick to using UPDATE.

Overwrite Option

@amotl asked:

I can imagine a user toggling a flag to also make it apply on a) could enable a convenience case where you can just slap a full load operation on a table that is already populated, with thinking much about needing to drop it first. That would be a cost vs. convenience tradeoff.

@hammerhead responded:

Some kind of “override if exists” flag for a) and b) could be interesting, we had a similar question the other day.

[hammerhead]

Makes sense I believe, thanks!

[kneth]

@amotl Is it possible to log the conflicts? I could imagine that the user would like to know how often it happens and maybe see if there is a pattern.