DDtKey opened a new issue, #9317:
URL: https://github.com/apache/arrow-datafusion/issues/9317

   ### Describe the bug
   
   I see a lot of cases of using `tokio::spawn` without wrapping `JoinHandle`s 
correctly.
   So if we stop execution (e.g `tokio::select!`) - we can find some tasks that 
are still alive.
   We can even see panic after canceling an execution (at least I did)
   
   Examples:
   - `demux_task`: [a 
function](https://github.com/apache/arrow-datafusion/blob/02c948d91b82f3e8988fafd12a072a476500c13d/datafusion/core/src/datasource/file_format/write/demux.rs#L73-L113)
 and then just [await on JoinHandle 
](https://github.com/apache/arrow-datafusion/blob/02c948d91b82f3e8988fafd12a072a476500c13d/datafusion/core/src/datasource/file_format/arrow.rs#L298-L308)
   - parquet 
[spawn_rg_join_and_finalize_task](https://github.com/apache/arrow-datafusion/blob/02c948d91b82f3e8988fafd12a072a476500c13d/datafusion/core/src/datasource/file_format/parquet.rs#L822-L847)
   - parquet 
[spawn_column_parallel_row_group_writer](https://github.com/apache/arrow-datafusion/blob/02c948d91b82f3e8988fafd12a072a476500c13d/datafusion/core/src/datasource/file_format/parquet.rs#L778-L779)
   - orchestration 
[serialize_rb_stream_to_object_store](https://github.com/apache/arrow-datafusion/blob/02c948d91b82f3e8988fafd12a072a476500c13d/datafusion/core/src/datasource/file_format/write/orchestration.rs#L61-L65)
   
   
   ### To Reproduce
   
   Easiest way, is to create a long-running query which probably may even fail, 
and track the tokio tasks with tokio console after cancelling the execution.
   
   ### Expected behavior
   
   I believe execution should be cancel-safe, and I think we need to care more 
about spawned tasks.
   We already had such issues before, e.g: 
https://github.com/apache/arrow-datafusion/issues/5178
   
   But I think even `AbortOnDropMany` doesn't solve all edge-cases, for example 
there are places when we do return `Vec<JoinHandle>` or just `JoinHandle` from 
an `async` function, but actually the execution can be canceled before wrapping 
into `AbortOnDropMany`. 
   So I'd say https://github.com/apache/arrow-datafusion/issues/6513 is more 
relevant now, I guess even single task should be spawned with `JoinSet`
   
   ### Additional context
   
   How can we protect of introducing this? 
   I think we have to pay attention to any `spawn` during review at least


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to