Skip to content

NestedEnumerator uses stale cursor if data changes between job runs #615

@liamstewart

Description

@liamstewart

Cursors store boundary values (> cursor_value). This is fine for normal enumerations but can lead to problems with NestedEnumerators. Nested cursors depend on the exact values of outer items, but that state is never explicitly stored. If data changes between runs, say a new outer value becomes eligible for processing, the NestedEnumerator doesn't recognize that and uses cursor state for inner layers from the old value.

test "nested cursors depend on outer values" do
  outer = [10, 20, 30]
  middle = [1, 2, 3]
  inner = [:a, :b, :c]

  # next- [20, 2, :b]
  cursor = [0, 0, 0]

  enums = [
    ->(cursor) { enumerator_builder.build_array_enumerator(outer, cursor: cursor) },
    ->(_outer_item, cursor) { enumerator_builder.build_array_enumerator(middle, cursor: cursor) },
    ->(outer_item, middle_item, cursor) {
      v = inner.map { |i| [outer_item, middle_item, i] }
      enumerator_builder.build_array_enumerator(v, cursor: cursor)
    },
  ]

  enum = NestedEnumerator.new(enums, cursor: cursor).each
  value_and_cursor = enum.take(1)[0]
  assert_equal([[20, 2, :b], [0, 0, 1]], value_and_cursor)

  # next should be [20, 2, :c] but we insert a new item in between so next is [15, 2, :c].
  # an argument could be made that next should be [15, 1, :a]
  outer = [10, 15, 20, 30]
  enum = NestedEnumerator.new(enums, cursor: value_and_cursor[1]).each
  value_and_cursor = enum.take(1)[0]
  assert_equal([[20, 2, :c], [0, 0, 2]], value_and_cursor)
end

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions