bug(sqlsmith): sqlsmith running in madsim is still not reproducible

[kwannoel]

Describe the bug

After #8068 , it was expected to sqlsmith should be able to reproduce results with each run.

The sequence of rng generated is still reproducible see:
local.txt
ci.txt

However it does not yield the same queries across different environments. (On the same machine and same build the same queries are returned)

I suspect this is because the sequence of query-gen can be out of order due to async execution.

Simple fix can be to ensure execution is synchronous, by refactoring to use postgres_client.

This pr: #7967 will help to stabilize main and PR workflow in the meantime.

To Reproduce

No response

Expected behavior

No response

Additional context

No response

[kwannoel]

cc @wangrunji0408

[wangrunji0408]

Yes, I also think async execution is the reason for irreproducible. If we could make all queries be generated sequentially at first, this problem might be resolved. BTW, we can't use postgres_client because it is not adapted to the simulator and will block the thread which is not allowed in the simulation. 🥲

[kwannoel]

Very hard to make it reproducible. Code changes can still lead to different code paths, and then it will no longer generate the same set of queries. Won't fix for now, since we already have snapshots which achieve the same thing.