Seems we fail multiple times a second on Cloud to insert a distinct ID into Postgres.
This makes sense, given a user that triggers an event is likely to trigger multiple and we see a ton of new users (distinct IDs) every day, meaning it's quite likely two events for the same non-existent ID will end up on different threads at the same time.
yakkomajuri
commented
3 years ago
Seems we fail multiple times a second on Cloud to insert a distinct ID into Postgres.
This makes sense, given a user that triggers an event is likely to trigger multiple and we see a ton of new users (distinct IDs) every day, meaning it's quite likely two events for the same non-existent ID will end up on different threads at the same time.
We do handle this properly but:
a) we swallow errors b) we're doing a lot of supposedly unnecessary queries and relying heavily on erroring as default behavior
I believe there should be ways to improve this. Have a couple ideas but mostly just documenting for now.