Mark all input columns in LIMIT BY as required output

[kvap]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

Prevent removal of any columns in LimitBy by marking all input columns as required output.

Category

Bug Fix

Short description

The query analyzer only marks the actual arguments of LIMIT BY as required
output for the LimitBy step in the pipeline. This is fine, unless the query is
distributed, in which case the first stage might remove a column that is used
at the second stage (e.g. for ORDER BY) but is not part of the final select.

Detailed description

If t is a distributed table with columns a and b, then a query like this will not work:

:) SELECT a FROM t ORDER BY b LIMIT 1 BY a;
-- will raise DB::Exception: Not found column b in block. There are only columns: a.

That happens because on the remote side the pipeline looks like this:

LimitBy
 Expression
  MergeSorting
   PartialSorting
    Expression
     Log

Here the LimitBy is the last step in the pipeline, so finalize() feels it can remove all columns that are not part of the final projection and not required in LimitBy.

But then, on the initial node the pipeline looks like this:

Expression
 LimitBy
  Expression
   MergingSorted <-- the ORDER BY column is unavailable
    Asynchronous × 3
     Remote

Fix that by marking all inputs of LimitBy as required outputs.

[kvap]

cc @bocharov @twalwyn @Dorokhov @victor-perov

[victor-perov]

Looks good! Thanks, @kvap!