From c8cfbe28cf1eadb7357bf13eaf610f2de98800e7 Mon Sep 17 00:00:00 2001
From: taiyang-li <654010905@qq.com>
Date: Wed, 29 Mar 2023 16:51:05 +0800
Subject: [PATCH] allow map key to be optional

---
 cpp/src/parquet/arrow/schema.cc | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/cpp/src/parquet/arrow/schema.cc b/cpp/src/parquet/arrow/schema.cc
index 267b892e4b40d..9a4081dbf1343 100644
--- a/cpp/src/parquet/arrow/schema.cc
+++ b/cpp/src/parquet/arrow/schema.cc
@@ -558,10 +558,23 @@ Status MapToSchemaField(const GroupNode& group, LevelInfo current_levels,
     return Status::Invalid("Key-value map node must have 1 or 2 child elements. Found: ",
                            key_value.field_count());
   }
+
+  /*
+   * If Parquet file was written by Flink, key type of map column is allowed to be optional, like this:
+   * optional group event_info (MAP) {
+   *   repeated group key_value {
+   *     optional binary key (UTF8);
+   *     optional binary value (UTF8);
+   *   }
+   * }
+   *
+   * Refer to: https://nightlies.apache.org/flink/flink-docs-release-1.16/docs/dev/table/types/#constructured-data-types
   const Node& key_node = *key_value.field(0);
   if (!key_node.is_required()) {
     return Status::Invalid("Map keys must be annotated as required.");
   }
+  */
+
   // Arrow doesn't support 1 column maps (i.e. Sets).  The options are to either
   // make the values column nullable, or process the map as a list.  We choose the latter
   // as it is simpler.