From d59e97e99a8cf2dd0ec0251e8adff260ab8c5337 Mon Sep 17 00:00:00 2001
From: Will Donnelly <will@estuary.dev>
Date: Thu, 14 Mar 2024 14:30:35 -0500
Subject: [PATCH] source-mysql: Omit unexpected `DB_ROW_HASH_1` column values

See https://github.com/estuary/connectors/issues/1344 for the
full explanation, but this commit adds a special-case to our
document value translation logic to exclude columns by the name
of `DB_ROW_HASH_1` for which we have no type information.

There is no accompanying test because this can only be reproduced
on MariaDB and we use MySQL for our default test setup, but I have
tested this locally and confirmed that it fixes what would otherwise
be a capture error.
---
 source-mysql/discovery.go | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/source-mysql/discovery.go b/source-mysql/discovery.go
index 2a4c051c01..8fd678d07d 100644
--- a/source-mysql/discovery.go
+++ b/source-mysql/discovery.go
@@ -171,6 +171,15 @@ func (db *mysqlDatabase) translateRecordFields(columnTypes map[string]interface{
 		return nil
 	}
 	for id, val := range f {
+		// MariaDB versions 10.4 and up include a synthetic `DB_ROW_HASH_1` column in the
+		// binlog row change events for certain tables with unique hash indices (see issue
+		// https://github.com/estuary/connectors/issues/1344). In such cases we won't have
+		// any type information for the column, but we also don't want it anyway, so as a
+		// special case we just delete the 'DB_ROW_HASH_1' property from the document.
+		if id == "DB_ROW_HASH_1" && columnTypes[id] == nil {
+			delete(f, id)
+			continue
+		}
 		var translated, err = db.translateRecordField(columnTypes[id], val)
 		if err != nil {
 			return fmt.Errorf("error translating field %q value %v: %w", id, val, err)