Enhancement
task executor manager is run on all tidb node, and it will run a complex join every 300ms. This might make tidb takes some mount of CPU even on idle state, and produce some slow query when the cluster has a large number of nodes and heavy workload.
we pick 300ms to make DXF more responsive to make small tasks runs faster as we don't have a mechanism(except PD) to send events across tidb now, and tidb doesn't support prepared stmt for internal statement, so the SQL is compiled every time.
maybe we can change the join to two separate single table query to make it cost less.
Enhancement
task executor manager is run on all tidb node, and it will run a complex join every 300ms. This might make tidb takes some mount of CPU even on idle state, and produce some slow query when the cluster has a large number of nodes and heavy workload.
we pick 300ms to make DXF more responsive to make small tasks runs faster as we don't have a mechanism(except PD) to send events across tidb now, and tidb doesn't support prepared stmt for internal statement, so the SQL is compiled every time.
maybe we can change the join to two separate single table query to make it cost less.