Zhenghua Lyu

Long time no response. Close it. Feel free to re-open this if needed.

Seems the same fix as https://github.com/greenplum-db/gpdb/pull/16908/ but from a difference perspective.

Thanks so much for your findings and discussion here. I think this is introduced by commit https://github.com/greenplum-db/gpdb/commit/efb2777a84b255fa7bd3f82d7c532fc7caa7ddcd which remove the line of code ```c if (contain_vars_of_level((Node *) subselect, 1)) return...

Would you please share minirepro here? https://community.pivotal.io/s/article/How-to-Collect-DDL-and-Statistics-Information-Using-the-Minirepro-Utility?language=en_US

After many hours, now I am able to reproduce this bug in my local env with the help of GDB. I use `rr` to record the bug and then find...

I think this bug is introduced during merging PG9.6 of upstream code https://github.com/greenplum-db/gpdb-postgres-merge/commit/8a7d0701814a4e293efad22091d6f6fb441bbe1c In GP6 (pg94), it will always scan the shared memory procarray so that subtransaction is considered.

A very simple local reproduce is using isolation 2: ``` create table t1(a int); create table t2(a int); 1: begin; 1: select txid_current(); 1: end; 2: begin; 2: select txid_current();...

Fixed by https://github.com/greenplum-db/gpdb/pull/17327. I also add the case here into that PR.

Is it possible to have a core file of the process that have huge amount RSS?

> If the core file is generated, where will it be? I don’t see any core files 1. need set `ulimit` to allow generate core 2. core file location is...