feat(filemanager): ensure events are ingested in the correct order

[mmalenic]

Closes #73

There's still a bit of repetitive code in the application logic and queries, but I think for now it's okay. It will be interesting to benchmark these queries eventually, especially within the update_reordered_for_*.sql queries because the select statement locks all rows with the same bucket, key and version_id. The update_reordered_for_*.sql queries are a bit long, but I think it's okay because it reduces the number of requests that the Lambda function needs to make to the database. Potentially the update and insert queries could be merged to reduce code repetition, with a flag variable indicating if it is a created or deleted event.

Changes

• Make sure that events are ingested in the correct order by using the created_sequencer and deleted_sequencer values in the database table. The process is:
  • For each event, check if the sequencer condition is met, where there is a sequencer value that better matches any already ingested event with the same bucket, key and version id.
  • Then, replace the already ingested event with the new one, and return the old event to be re-ingested, as it could belong to another object.
  • If there is no out of order event, then just insert the event normally.
  • This process allows the database to correct any ordering issues as it receives events, where the current state of the database represents the best known order of all events received.
• Move the size and checksum from object to s3_object because this simplifies the reordering logic.
• Add various tests:
  • More ordering and duplication tests.
  • Tests for the queries themselves (rather than inside the ingest function).
  • Long permutation test to make sure event reordering works correctly. This could also be rewritten as a benchmark or performance test.

[victorskl]

I will review in tick, soon

[victorskl]

Very detailed impl, Marko. Kudos.! Can't wait to see this in action.