[Bug]: txn need retry in rc mode during CREATE INDEX

[cpegeric]

Is there an existing issue for the same bug?

• I have checked the existing issues.

Branch Name

main

Commit ID

ec44079

Other Environment Information

- Hardware parameters:
- OS type:
- Others:

Actual Behavior

When I create an IVF index with 1 million vector with dimension 3072, create index is very slow and take around 10 mins to finish. At the end of the create index, insert into mo_catalog.mo_indexes failed with error "txn need retry in rc mode, def changed." Actually, only one session is running. As a result, create index re-run again and run successfully second time.

Expected Behavior

Create index should be finished in one time without retry.

Steps to Reproduce

1 million vector with dimension 3072.

sql> create index idx using ivfflat on tbl(embed) lists=1000 op_type 'vector_l2_ops';

Additional information

No response

[cpegeric]

cannot reproduce any more. closed.

[cpegeric]

you can test the repo with commit 9c15844

[ouyuanning]

从日志来看，大概的原因是：
1、有个SQL需要调用内部执行器执行一条子SQL
2、子SQL在开始lock meta table的时候，发现schema change，于是报错。但是这个报错没有引起retry，而是往上抛出了，
初步怀疑是这个地方，内部执行器和普通执行路径有差异：

[cpegeric]

To reproduce the bug,

you can test the repo with commit 9c15844

1. create table ivftable (id bigint primary key auto_increment, embed vecf64(3072));

2. load 1 million vector with dimension 3072

3. create index idx using ivfflat on tbl(embed) lists=1000 op_type 'vector_l2_ops';