Description

Right now keys to read are generated on random (random seed is const, so the problem is a bit theoretical) and it means the generated key may not exist. We should rather generate keys based on a list of existing keys (previously inserted).
It may not be trivial for benchmarks doing reads and writes simultaneously (like readrandomwriterandom).
And we have to do this will little-to-no impact on performance measurements.

We could also introduce a way to parametrize how many keys should exist (something like hit/miss ratio).

Rationale

We could test hit/miss rates on key reads. E.g. see different behavior for db performance, which returns NOT_FOUND / OK statuses in different ratios.

Sometimes it's required to read existing keys, otherwise the measurements may be e.g. unexpectedly high.

API Changes

rather no
perhaps a new benchmark type or new input parameter will be provided along this change

Implementation details

TBD