This currently relies on a pool of workers and endpoints; ideally we would be able to share a worker or even better multiplex multiple calls over a single endpoint (this would require wire protocol changes, however!). Care should be taken not to hurt performance if we do enable a multithreaded worker (which would be necessary, unless we switch to a model where all threads send work to a single worker thread).
Reporter: David Li / @lidavidm
Related issues:
Note: This issue was originally created as ARROW-16127. Please see the migration documentation for further details.