We should deprecate mixed coordinator and worker clusters:
1) As a general best practice, one doesn't get great performance from such clusters, as the resource overhead from coordination (highly concurrent) reduces the performance of query execution (low concurrency, high CPU).
2) As we move into native clusters, such deployments become practically impossible, and deprecating this configuration will become necessary anyway.
Once this configuration is deprecated, we can also clean up the Presto Guice modules and draw a more clear separation between code that is coordinator-only and worker-only.
We should deprecate mixed coordinator and worker clusters:
1) As a general best practice, one doesn't get great performance from such clusters, as the resource overhead from coordination (highly concurrent) reduces the performance of query execution (low concurrency, high CPU). 2) As we move into native clusters, such deployments become practically impossible, and deprecating this configuration will become necessary anyway.
It's been documented that such configurations should be avoided for performance reasons, so it's likely this should cause minimal disruption. However, this will be a backwards incompatible change, and if this causes any great problems for the user community, please comment.
Once this configuration is deprecated, we can also clean up the Presto Guice modules and draw a more clear separation between code that is coordinator-only and worker-only.