I suppose it isn't that simple to
I suppose it isn't that simple to enforce idempotency
2 Replies
yeah, it's a bit tricky to automatically apply retries lower down in the stack. There was a suggestion that if you get a known-transient error from the DO you could parse the sql and decide if it's a read, and then retry it if it's a read. Requires parsing the sql statement in two places and a bit of guesswork though.
Not the worst idea, and I suppose the second SQL parsing could be done only if a transient error is encountered
Something else that could be a nice to have, just putting that out there, is a sort of easy way to check if a response is a known transient error
I guess there's already an issue about better documenting them, but a way to verify programmatically without having to maintain the list of strings in our code would be cool too!