Currently, we have native support for or can wrap many processor core types; however, there are some exceptions that prohibit the vectorization of the whole flow if the "bad" core is present in the physical plan. This issue tracks addressing all of those.

• LocalPlanNode
  • there is a WIP to address it, sql,colexec: add support for wrapping LocalPlanNode #55909
• MetadataTestSender and MetadataTestReceiver
  • I think we should remove these test processors entirely. Possibly colexec: add a way to propagate metadata in a streaming fashion #55758 is related.
• ChangeAggregator and ChangeFrontier
  • blocked by.
• InvertedFilterer
  • wrapping of the inverted filterer is unsupported because the cFetcher always decodes the data coming from disk, and the row-execution framework just happens to work due to EncDatum.encoded field. opt: change the key column of inverted indexes to use a new column id and type #50695 was the original issue filed for it, it is now closed, yet I think the problem is still present.
• Backfiller, CSVWriter, Sampler, SampleAggregator
  • these processors don't implement execinfra.RowSource interface, so they cannot be wrapped. They, however, use the DistSQL framework but are never - I think - planned in the flows containing other supported processors. Stats operators work is tracked in sql: implement vectorized operator for table statistics collection #41203.

Jira issue: CRDB-2842