apacheGH-34437: [R] Use FetchNode and OrderByNode (apache#34685)

### Rationale for this change

See also apache#32991. By using the new nodes, we're closer to having all dplyr query business happening inside the ExecPlan. Unfortunately, there are still two cases where we have to apply operations in R after running a query:

* apache#34941: Taking head/tail on unordered data, which has non-deterministic results but that should be possible, in the case where the user wants to see a slice of the result, any slice
* apache#34942: Implementing tail in the FetchNode or similar would enable removing more hacks and workarounds.

Once those are resolved, we can simply further and then move to the new Declaration class.

### What changes are included in this PR?

This removes the use of different SinkNodes and many R-specific workarounds to support sorting and head/tail, so *almost* 
everything we do in a query should be represented in an ExecPlan. 

### Are these changes tested?

Yes. This is mostly an internal refactor, but behavior changes are accompanied by test updates.

### Are there any user-facing changes?

The `show_query()` method will print slightly different ExecPlans. In many cases, they will be more informative. 

`tail()` now actually returns the tail of the data in cases where the data has an implicit order (currently only in-memory tables). Previously it was non-deterministic (and would return the head or some other slice of the data).

When printing query objects that include `summarize()` when the `arrow.summarize.sort = TRUE` option is set, the sorting is correctly printed.

It's unclear if there should be changes in performance; running benchmarks would be good but it's also not clear that our benchmarks cover all affected scenarios. 

* Closes: apache#34437
* Closes: apache#31980
* Closes: apache#31982

Authored-by: Neal Richardson <neal.p.richardson@gmail.com>
Signed-off-by: Nic Crane <thisisnic@gmail.com>

Author: nealrichardson

r/R/arrowExports.R

@@ -287,6 +287,11 @@ do_arrow_summarize <- function(.data, ..., .groups = NULL) {
       stop(paste("Invalid .groups argument:", .groups))
     }
     out$drop_empty_groups <- .data$drop_empty_groups
+    if (getOption("arrow.summarise.sort", FALSE)) {
+      # Add sorting instructions for the rows to match dplyr
+      out$arrange_vars <- .data$selected_columns[.data$group_by_vars]
+      out$arrange_desc <- rep(FALSE, length(.data$group_by_vars))
+    }
   }
   out
 }

r/R/dplyr-summarize.R

@@ -284,17 +284,7 @@ tail.arrow_dplyr_query <- function(x, n = 6L, ...) {
 #'   mutate(x = gear / carb) %>%
 #'   show_exec_plan()
 show_exec_plan <- function(x) {
-  adq <- as_adq(x)
-
-  # do not show the plan if we have a nested query (as this will force the
-  # evaluation of the inner query/queries)
-  # TODO see if we can remove after ARROW-16628
-  if (is_collapsed(x) && has_head_tail(x$.data)) {
-    warn("The `ExecPlan` cannot be printed for a nested query.")
-    return(invisible(x))
-  }
-
-  result <- as_record_batch_reader(adq)
+  result <- as_record_batch_reader(as_adq(x))
   plan <- result$Plan()
   on.exit({
     plan$.unsafe_delete()
@@ -419,6 +409,25 @@ query_can_stream <- function(x) {
 
 is_collapsed <- function(x) inherits(x$.data, "arrow_dplyr_query")
 
-has_head_tail <- function(x) {
-  !is.null(x$head) || !is.null(x$tail) || (is_collapsed(x) && has_head_tail(x$.data))
+has_unordered_head <- function(x) {
+  if (is.null(x$head %||% x$tail)) {
+    # no head/tail
+    return(FALSE)
+  }
+  !has_order(x)
+}
+
+has_order <- function(x) {
+  length(x$arrange_vars) > 0 ||
+    has_implicit_order(x) ||
+    (is_collapsed(x) && has_order(x$.data))
+}
+
+has_implicit_order <- function(x) {
+  # Approximate what ExecNode$has_ordered_batches() would return (w/o building ExecPlan)
+  # An in-memory table has an implicit order
+  # TODO(GH-34698): FileSystemDataset and RecordBatchReader will have implicit order
+  inherits(x$.data, "ArrowTabular") &&
+    # But joins, aggregations, etc. will result in non-deterministic order
+    is.null(x$aggregations) && is.null(x$join) && is.null(x$union_all)
 }

r/R/dplyr.R

@@ -74,11 +74,11 @@ ExecPlan <- R6Class("ExecPlan",
 
       if (is_collapsed(.data)) {
         # We have a nested query.
-        if (has_head_tail(.data$.data)) {
-          # head and tail are not ExecNodes; at best we can handle them via
-          # SinkNode, so if there are any steps done after head/tail, we need to
-          # evaluate the query up to then and then do a new query for the rest.
-          # as_record_batch_reader() will build and run an ExecPlan
+        if (has_unordered_head(.data$.data)) {
+          # TODO(GH-34941): FetchNode should do non-deterministic fetch
+          # Instead, we need to evaluate the query up to here,
+          # and then do a new query for the rest.
+          # as_record_batch_reader() will build and run an ExecPlan and do head() on it
           reader <- as_record_batch_reader(.data$.data)
           on.exit(reader$.unsafe_delete())
           node <- self$SourceNode(reader)
@@ -126,15 +126,6 @@ ExecPlan <- R6Class("ExecPlan",
           options = .data$aggregations,
           key_names = group_vars
         )
-
-        if (grouped && getOption("arrow.summarise.sort", FALSE)) {
-          # Add sorting instructions for the rows too to match dplyr
-          # (see below about why sorting isn't itself a Node)
-          node$extras$sort <- list(
-            names = group_vars,
-            orders = rep(0L, length(group_vars))
-          )
-        }
       } else {
         # If any columns are derived, reordered, or renamed we need to Project
         # If there are aggregations, the projection was already handled above.
@@ -166,82 +157,81 @@ ExecPlan <- R6Class("ExecPlan",
         }
       }
 
-      # Apply sorting: this is currently not an ExecNode itself, it is a
-      # sink node option.
-      # TODO: handle some cases:
-      # (1) arrange > summarize > arrange
-      # (2) ARROW-13779: arrange then operation where order matters (e.g. cumsum)
+      # Apply sorting and head/tail
+      head_or_tail <- .data$head %||% .data$tail
       if (length(.data$arrange_vars)) {
-        node$extras$sort <- list(
+        if (!is.null(.data$tail)) {
+          # Handle tail first: Reverse sort, take head
+          # TODO(GH-34942): FetchNode support for tail
+          node <- node$OrderBy(list(
+            names = names(.data$arrange_vars),
+            orders = as.integer(!.data$arrange_desc)
+          ))
+          node <- node$Fetch(.data$tail)
+        }
+        # Apply sorting
+        node <- node$OrderBy(list(
           names = names(.data$arrange_vars),
-          orders = .data$arrange_desc,
-          temp_columns = names(.data$temp_columns)
-        )
-      }
-      # This is only safe because we are going to evaluate queries that end
-      # with head/tail first, then evaluate any subsequent query as a new query
-      if (!is.null(.data$head)) {
-        node$extras$head <- .data$head
-      }
-      if (!is.null(.data$tail)) {
-        node$extras$tail <- .data$tail
+          orders = as.integer(.data$arrange_desc)
+        ))
+
+        if (length(.data$temp_columns)) {
+          # If we sorted on ad-hoc derived columns, Project to drop them
+          temp_schema <- node$schema
+          cols_to_keep <- setdiff(names(temp_schema), names(.data$temp_columns))
+          node <- node$Project(make_field_refs(cols_to_keep))
+        }
+
+        if (!is.null(.data$head)) {
+          # Take the head now
+          node <- node$Fetch(.data$head)
+        }
+      } else if (!is.null(head_or_tail)) {
+        # Unsorted head/tail
+        # Handle a couple of special cases here:
+        if (node$has_ordered_batches()) {
+          # Data that has order, even implicit order from an in-memory table, is supported
+          # in FetchNode
+          if (!is.null(.data$head)) {
+            node <- node$Fetch(.data$head)
+          } else {
+            # TODO(GH-34942): FetchNode support for tail
+            # FetchNode currently doesn't support tail, but it has limit + offset
+            # So if we know how many rows the query will result in, we can offset
+            data_without_tail <- .data
+            data_without_tail$tail <- NULL
+            row_count <- nrow(data_without_tail)
+            if (!is.na(row_count)) {
+              node <- node$Fetch(.data$tail, offset = row_count - .data$tail)
+            } else {
+              # Workaround: non-deterministic tail
+              node$extras$slice_size <- head_or_tail
+            }
+          }
+        } else {
+          # TODO(GH-34941): non-deterministic FetchNode
+          # Data has non-deterministic order, so head/tail means "just show me any N rows"
+          # FetchNode does not support non-deterministic scans, so we have to handle outside
+          node$extras$slice_size <- head_or_tail
+        }
       }
       node
     },
     Run = function(node) {
       assert_is(node, "ExecNode")
-
-      # Sorting and head/tail (if sorted) are handled in the SinkNode,
-      # created in ExecPlan_build
-      sorting <- node$extras$sort %||% list()
-      select_k <- node$extras$head %||% -1L
-      has_sorting <- length(sorting) > 0
-      if (has_sorting) {
-        if (!is.null(node$extras$tail)) {
-          # Reverse the sort order and take the top K, then after we'll reverse
-          # the resulting rows so that it is ordered as expected
-          sorting$orders <- !sorting$orders
-          select_k <- node$extras$tail
-        }
-        sorting$orders <- as.integer(sorting$orders)
-      }
-
       out <- ExecPlan_run(
         self,
         node,
-        sorting,
-        prepare_key_value_metadata(node$final_metadata()),
-        select_k
+        prepare_key_value_metadata(node$final_metadata())
       )
 
-      if (!has_sorting) {
-        # Since ExecPlans don't scan in deterministic order, head/tail are both
+      if (!is.null(node$extras$slice_size)) {
+        # For non-deterministic scans, head/tail are
         # essentially taking a random slice from somewhere in the dataset.
         # And since the head() implementation is way more efficient than tail(),
         # just use it to take the random slice
-        # TODO(ARROW-16628): handle limit in ExecNode
-        slice_size <- node$extras$head %||% node$extras$tail
-        if (!is.null(slice_size)) {
-          out <- head(out, slice_size)
-        }
-      } else if (!is.null(node$extras$tail)) {
-        # TODO(ARROW-16630): proper BottomK support
-        # Reverse the row order to get back what we expect
-        out <- as_arrow_table(out)
-        out <- out[rev(seq_len(nrow(out))), , drop = FALSE]
-        out <- as_record_batch_reader(out)
-      }
-
-      # If arrange() created $temp_columns, make sure to omit them from the result
-      # We can't currently handle this in ExecPlan_run itself because sorting
-      # happens in the end (SinkNode) so nothing comes after it.
-      # TODO(ARROW-16631): move into ExecPlan
-      if (length(node$extras$sort$temp_columns) > 0) {
-        tab <- as_arrow_table(out)
-        tab <- tab[, setdiff(names(tab), node$extras$sort$temp_columns), drop = FALSE]
-        out <- as_record_batch_reader(tab)
+        out <- head(out, node$extras$slice_size)
       }
-
       out
     },
     Write = function(node, ...) {
@@ -272,13 +262,8 @@ ExecNode <- R6Class("ExecNode",
   inherit = ArrowObject,
   public = list(
     extras = list(
-      # `sort` is a slight hack to be able to keep around arrange() params,
-      # which don't currently yield their own ExecNode but rather are consumed
-      # in the SinkNode (in ExecPlan$run())
-      sort = NULL,
-      # Similar hacks for head and tail
-      head = NULL,
-      tail = NULL,
+      # Workaround for non-deterministic head/tail
+      slice_size = NULL,
       # `source_schema` is put here in Scan() so that at Run/Write, we can
       # extract the relevant metadata and keep it in the result
       source_schema = NULL
@@ -295,6 +280,7 @@ ExecNode <- R6Class("ExecNode",
       old_meta$r <- get_r_metadata_from_old_schema(self$schema, old_schema)
       old_meta
     },
+    has_ordered_batches = function() ExecNode_has_ordered_batches(self),
     Project = function(cols) {
       if (length(cols)) {
         assert_is_list_of(cols, "Expression")
@@ -336,6 +322,16 @@ ExecNode <- R6Class("ExecNode",
     },
     Union = function(right_node) {
       self$preserve_extras(ExecNode_Union(self, right_node))
+    },
+    Fetch = function(limit, offset = 0L) {
+      self$preserve_extras(
+        ExecNode_Fetch(self, offset, limit)
+      )
+    },
+    OrderBy = function(sorting) {
+      self$preserve_extras(
+        ExecNode_OrderBy(self, sorting)
+      )
     }
   ),
   active = list(