Flo
Flo
We need to improve the accuracy of our distributed rate limiting system, as it currently fails under high node and RPS scenarios, exceeding the 20% tolerance threshold when running `make...
We should add traces to all the ratelimit functions, so we can see whats taking the most time.
UsageLimiter.Service should be CreditLimiter.Service and so on.This should only be done once we've shipped identity based credits as the merge conflicts would be enormous otherwise. The v1 api can stay...
Once we merged a change the in the merge queue we should also only run the tests that we actually need to run.
When the parent ticket is done, we would still be running ALL tests for our go API even tho we might not need them.For example when we add a new...
Currently our docker compose builds the same go image multiple times for each service e.g krane api gw ctrl etc. It should be build only once and then re-used
@imeyer said: > all of this looks good! > > one thing i think we may want to do is not return 5xx for DB operation issues.. the 4xx class...
When a database operations fails we should record the span as replica so we can look at the error traces when something failed.This currently doesn't happen but should be done...
For speakeasy to know about paginated endpoints, we need to use the `x-speakeasy-pagination` directive.This is currently only used by the listIdentities endpoint, but should be used by all endpoints that...