PostgreSQLBulkCopy performance degradation for method BulkCopyType.ProviderSpecific

[AndreyShipunov]

Good afternoon!

After switching to .NET 8 bugs have been fixed in the ProviderSpecificCopySyncImpl method https://github.com/linq2db/linq2db/blob/master/Source/LinqToDB/DataProvider/PostgreSQL/PostgreSQLBulkCopy.cs#L139

Bugs have been fixed by adding lines:
https://github.com/linq2db/linq2db/blob/master/Source/LinqToDB/DataProvider/PostgreSQL/PostgreSQLBulkCopy.cs#L167
https://github.com/linq2db/linq2db/blob/master/Source/LinqToDB/DataProvider/PostgreSQL/PostgreSQLBulkCopy.cs#L170

Both lines use the GetNativeType method - this method is incredibly slow. And it is called in a loop for each column and for each row.

In the line https://github.com/linq2db/linq2db/blob/master/Source/LinqToDB/DataProvider/PostgreSQL/PostgreSQLBulkCopy.cs#L170 you can swap the check order to reduce the number of GetNativeType calls:
if (value is DateTimeOffset dto && _provider.GetNativeType(dataType.DbType) == NpgsqlProviderAdapter.NpgsqlDbType.TimeTZ)

The NormalizeTimeStamp method also calls `GetNativeType'.

At the moment, inserting 10,000 rows with 15 columns (four of them require calling NormalizeTimeStamp) via BulkCopyType.MultipleRows is about 3 times faster than BulkCopyType.ProviderSpecific. At the moment, BulkCopyType.ProviderSpecific cannot be used in production - it always works slower.

BulkCopyType.MultipleRows spends a lot of time and memory generating SQL strings, and BulkCopyType.ProviderSpecific spends 99% of the time calling GetNativeType 200,000 times.

As a good solution, you can calculate GetNativeType before loops, because it makes no sense to check the column type for each row.