Open aarongable opened 1 month ago
Note that replacing the Auto-Increment behavior will break the current partition management scheme.
Step 1 here is:
We also probably want to reduce the configured parallelism per RPC, as we believe that may have been a contributing factor in the SA<->ProxySQL connection churn we've historically had problems with.
From there, we can flip the flag in staging/prod, and easily turn it off if it causes problems. If it doesn't cause problems, we're good here -- if it does cause problems, we can dive into batch inserts, random ID generation, and all the knock-on effects of that.
This doesn't work in MySQL, so we need to remove it before we migrate.
We may want to replace it with random ID generation rather than auto-increment, if we're worried about the efficiency of inserting 100 rows for a maximally-sized newOrder request.