Ken, Wang comments

Results 66 comments of


                                            Ken, Wang

Task level retry and Stage level retry

> I tested this locally, and it worked really well! > > I ran one scheduler, and two executors and I could kill one executor and still see a query...

Need to add an unique id to MetricsSet or ExecutionPlanMetricsSet

@alamb A single session could have multiple statements run concurrently. And a statement_id is not enough to differ metrics either. For example a single query statement might contain multiple join...

Need to add an unique id to MetricsSet or ExecutionPlanMetricsSet

If we do not have unique ids to represent the MetricsSet, we have to use some other implicit ways to correlate the metrics. One solution is to leverage the array...

Need to add an unique id to MetricsSet or ExecutionPlanMetricsSet

https://github.com/apache/arrow-ballista/issues/116

Need to add an unique id to MetricsSet or ExecutionPlanMetricsSet

> job_id Yes, the job_id in Ballista is similar to statement_id. In current Ballista implementation, each SQL statement will just trigger one job.

Add Caller Context to the ObjectStore Trait for tracing and diagnostic

@crepererum I think maybe we just need to add a parameter of type String `` ext: Option``. Users of the ObjectStore can feel free to encode any tracing info to...

Is there a way to set retry and max retry times for client request ?

In Netty, there is an IdleStateHandler which can trigger event to close the idle tcp connections. Not sure whether Hyper has a way to achieve the same effects or not.

Support Selective Aggregations: FILTER clause to aggregate Functions

https://modern-sql.com/feature/filter

Support Grouping functions with Group By CUBE/ROLLUP/GROUPING SETS

I'm working on it now.

Memory Limited GroupBy (Externalized / Spill)

@alamb @yjshen Can we make the `GroupState` and the Accumulator states serializable ? With this approach, we do not need to do any sort when spiiling data to disks. And...