Xiaoju Wu

Results 9 comments of Xiaoju Wu

BTW, memory usage of region cache could be tracked in case of risk of oom.

Another finding is, cached regions are much more than real live regions: ![image](https://user-images.githubusercontent.com/5151781/176081855-060ae98d-21a9-46a6-9ae4-d5a0328ad956.png) ![image](https://user-images.githubusercontent.com/5151781/176081861-25f75cec-09a0-438b-bd4a-15f8cef09b7b.png) This use case has many "truncate table". The eviction of cached regions could be enhanced.

``` mysql> set session tidb_enforce_mpp=ON; Query OK, 0 rows affected (0.04 sec) mysql> explain SELECT SUM(LO_EXTENDEDPRICE*LO_DISCOUNT) AS REVENUE FROM HAT.LINEORDER, HAT.DATE WHERE LO_ORDERDATE = D_DATEKEY AND D_YEAR = 1993 AND...

related with https://github.com/pingcap/tidb/pull/30659 and https://github.com/pingcap/tidb/pull/36771. The special virtual column tidb_shard('xxx'): 1. should not be considered as "used columns" in rule_partition_pruning 2. should not affect the access paths selection 3. should...

similar with https://github.com/pingcap/tidb/issues/36658

@Yisaer Can you help confirm with @mjonss if this issue can be closed since we have the analyze performance improved for partition table?