Eric Fu

@Li0k Is this done? Can we close this issue?

Strong +1 for optimizer hint / session variable. It's very common to try different execution plans or manually control the plans in practice. > how do we let users enable/disable...

See https://github.com/risingwavelabs/risingwave/issues/7491

@wcy-fdu is working on Memtable spill. Hopefully it can solve this issue. If not, let's see later.

`ENABLE_MEMORY_PROFILING` was true in last run. Let's analyze the heap dump to see what happened...

Ran again with memory-size-based back-pressure (#13775). Still OOMed... 🥹 https://buildkite.com/risingwave-test/ch-benchmark-pg-cdc/builds/103 [1701841399-2023-12-06-05-43-18.auto.heap.collapsed.zip](https://github.com/risingwavelabs/risingwave/files/13577722/1701841399-2023-12-06-05-43-18.auto.heap.collapsed.zip)

> change the flush threshold to 1mb or smaller may work I don't think so. Searching for `state_table`, the results show that the usage is less than 2GB, which is...

> https://github.com/risingwavelabs/risingwave/issues/14797#issuecomment-1963266578 Does this confirm that subquery unnesting is the major cause?

Running a test with `stream_hash_agg_max_dirty_groups_heap_size` = 2MB https://buildkite.com/risingwave-test/ch-benchmark-pg-cdc/builds/104 I think it doesn't make sense that `stream_hash_agg_max_dirty_groups_heap_size` is much bigger than `mem_table_spill_threshold`. They should be same, right?

> I guess Some AggGroup could be larger than the result row. Oh, do you mean min/max? Indeed. Let's wait for the test first. First we need to prove it's...