npgsql：如何在一个以集合为参数的命令中使用 npgsql 选择多行（具有多个列值）？

Question

I have two tables defined below, supplier_balances and supplier_balance_items (btw there is a 1[supplier_balance]:N[supplier_balance_items] relationship between the two):

CREATE TABLE IF NOT EXISTS sch_brand_payment_data_lake_proxy.supplier_balances (
/* id is here for joining purposes with items table, instead of joining with the 4 columns used for sake
   of making sure a record is deemed as unique */
  id                             bigserial NOT NULL,
  accounting_document            text      NOT NULL,
  accounting_document_type       text      NOT NULL,
  company_code                   text      NOT NULL,
  document_date_year             int4      NOT NULL,
  accounting_doc_created_by_user text,
  accounting_clerk               text,
  assignment_reference           text,
  document_reference_id          text,
  original_reference_document    text,
  payment_terms                  text,
  supplier                       text,
  supplier_name                  text,
  document_date                  timestamp,
  posting_date                   timestamp,
  net_due_date                   timestamp,
  created_on                     timestamp default NULL,
  modified_on                    timestamp default NULL,
  pushed_on                      timestamp default NULL,
  is_modified bool GENERATED ALWAYS AS (modified_on IS NOT NULL AND modified_on > created_on) STORED,
  is_pushed   bool GENERATED ALWAYS AS (pushed_on   IS NOT NULL AND pushed_on > modified_on)  STORED,
  CONSTRAINT supplier_balances_pkey   PRIMARY KEY (id),
  /* accounting_document being the field of the composite unique index -> faster querying */
  CONSTRAINT supplier_balances_unique UNIQUE (
     accounting_document,
     accounting_document_type,
     company_code,
     document_date_year)
);

/* Creating other indexes for querying of those as well */
CREATE INDEX IF NOT EXISTS supplier_balances_accounting_document_type_idx
ON sch_brand_payment_data_lake_proxy.supplier_balances (accounting_document_type);
CREATE INDEX IF NOT EXISTS supplier_balances_company_code_idx
ON sch_brand_payment_data_lake_proxy.supplier_balances (company_code);
CREATE INDEX IF NOT EXISTS supplier_balances_document_date_year_idx
ON sch_brand_payment_data_lake_proxy.supplier_balances (document_date_year);

CREATE TABLE IF NOT EXISTS sch_brand_payment_data_lake_proxy.supplier_balance_items
(
    supplier_balance_id             bigserial NOT NULL,
    posting_view_item               text      NOT NULL,
    posting_key                     text,
    amount_in_company_code_currency numeric,
    amount_in_transaction_currency  numeric,
    cash_discount_1_percent         numeric,
    cash_discount_amount            numeric,
    clearing_accounting_document    text,
    document_item_text              text,
    gl_account                      text,
    is_cleared                      bool,
    clearing_date                   timestamp,
    due_calculation_base_date       timestamp,
    /* uniqueness is basically the posting_view_item for a given supplier balance */
    CONSTRAINT supplier_balance_items_pkey PRIMARY KEY (supplier_balance_id, posting_view_item),
    /* 1(supplier balance):N(supplier balance items) */
    CONSTRAINT supplier_balance_items_fkey FOREIGN KEY (supplier_balance_id)
               REFERENCES sch_brand_payment_data_lake_proxy.supplier_balances (id)
               ON DELETE CASCADE
               ON UPDATE CASCADE
);

Note: I'm just filling up the columns that can't be NULL for the sake of simplicity.

INSERT INTO 
sch_brand_payment_data_lake_proxy.supplier_balances 
(accounting_document, accounting_document_type, company_code, document_date_year)
VALUES 
('A', 'B', 'C', 0),
('A', 'B', 'C', 1),
('A', 'B', 'C', 2),
('A', 'B', 'C', 3),
('A', 'B', 'C', 4),
('A', 'B', 'C', 5)
RETURNING id;

Output:

id
1
2
3
4
5
6

INSERT INTO 
sch_brand_payment_data_lake_proxy.supplier_balance_items 
(supplier_balance_id, posting_view_item)
VALUES 
(1, 'A'),
(1, 'B'),
(3, 'A'),
(3, 'B'),
(2, 'A'),
(1, 'C');

SELECT 
    accounting_document, 
    accounting_document_type, 
    company_code, 
    document_date_year
FROM sch_brand_payment_data_lake_proxy.supplier_balances;

Output:

id	accounting_document	accounting_document_type	company_code	document_date_year
1	A	B	C	0
2	A	B	C	1
3	A	B	C	2
4	A	B	C	3
5	A	B	C	4
6	A	B	C	5

SELECT 
    supplier_balance_id,
    posting_view_item
FROM sch_brand_payment_data_lake_proxy.supplier_balance_items;

Output:

supplier_balance_id	posting_view_item
1	A
1	B
3	A
3	B
2	A
1	C

Now if we want to select multiple values in a JOIN we can perform in a raw SQL:

SELECT 
    id,
    accounting_document, 
    accounting_document_type, 
    company_code, 
    document_date_year, 
    posting_view_item
FROM sch_brand_payment_data_lake_proxy.supplier_balances
LEFT OUTER JOIN sch_brand_payment_data_lake_proxy.supplier_balance_items
ON supplier_balances.id = supplier_balance_items.supplier_balance_id
WHERE (accounting_document, accounting_document_type, company_code, document_date_year)
IN  (('A', 'B', 'C', 1), ('A', 'B', 'C', 2))

Output:

id	accounting_document	accounting_document_type	company_code	document_date_year	posting_view_item
2	A	B	C	1	A
3	A	B	C	2	A

https://github.com/npgsql/npgsql/issues/1199

Now, when using npgsql in C#, reproducing the query just above is an easy feat:

using System.Data;

using Npgsql;

var connectionStringBuilder = new NpgsqlConnectionStringBuilder
{
    Host     = "localhost",
    Port     = 5432,
    Username = "brand_payment_migration",
    Password = "secret",
    Database = "brand_payment"
};
using var connection = new NpgsqlConnection(connectionStringBuilder.ToString());
connection.Open();
using var command = connection.CreateCommand();
command.CommandText = 
"SELECT id, accounting_document, accounting_document_type, company_code, document_date_year, posting_view_item " +
"FROM sch_brand_payment_data_lake_proxy.supplier_balances " +
"LEFT OUTER JOIN sch_brand_payment_data_lake_proxy.supplier_balance_items " +
"ON supplier_balances.id = supplier_balance_items.supplier_balance_id " +
"WHERE (accounting_document, accounting_document_type, company_code, document_date_year) " +
"IN (('A', 'B', 'C', 1), ('A', 'B', 'C', 2));";

using var reader = command.ExecuteReader();
using var dataTable = new DataTable();
dataTable.Load(reader);
var cols = dataTable.Columns.Cast<DataColumn>().ToArray();
Console.WriteLine(string.Join(Environment.NewLine, cols.Select((x, i) => $"Col{i} = {x}")));
Console.WriteLine(string.Join("\t", cols.Select((_, i) => $"Col{i}")));
foreach (var dataRow in dataTable.Rows.Cast<DataRow>())
{
    Console.WriteLine(string.Join("\t", dataRow.ItemArray));
}

which, as expected outputs:

Col0 = id
Col1 = accounting_document
Col2 = accounting_document_type
Col3 = company_code
Col4 = document_date_year
Col5 = posting_view_item
Col0    Col1    Col2    Col3    Col4    Col5
2       A       B       C       1       A
3       A       B       C       2       A
3       A       B       C       2       B

Now, what I would like to achieve is that, instead of passing a raw string for (('A', 'B', 'C', 1), ('A', 'B', 'C', 2)); , I would love to use a NpgSqlParameter with a collection of value set (ie for each column)).

So I've changed the C# snippet above and added the parameter

// ...
"WHERE (accounting_document, accounting_document_type, company_code, document_date_year) " +
"IN @values;";
var parameter = command.CreateParameter();
parameter.ParameterName = "@values";
parameter.NpgsqlDbType = NpgsqlDbType.Array;
parameter.NpgsqlValue = new object[,]
{
    { "A", "B", "C", 1 }, 
    { "A", "B", "C", 2 }
};
// Note: the same kind of issue arises when using tuples, i.e.
// ( "A", "B", "C", 1 )
// ( "A", "B", "C", 2 )
command.Parameters.Add(parameter);
using var reader = command.ExecuteReader();
// ...

Then I'm getting this exception:

Unhandled exception. System.ArgumentOutOfRangeException: Cannot set NpgsqlDbType to just Array, Binary-Or with the element type (e.g. Array of Box is NpgsqlDbType.Array | Npg
sqlDbType.Box). (Parameter 'value')
   at Npgsql.NpgsqlParameter.set_NpgsqlDbType(NpgsqlDbType value)
   at Program.<Main>$(String[] args) in C:\Users\natalie-perret\Desktop\Personal\playground\csharp\CSharpPlayground\Program.cs:line 25

Then I've tried to work around that error with:

parameter.NpgsqlDbType = NpgsqlDbType.Array | NpgsqlDbType.Unknown;

but then getting another exception:

Unhandled exception. System.ArgumentException: No array type could be found in the database for element .<unknown>
   at Npgsql.TypeMapping.ConnectorTypeMapper.ResolveByNpgsqlDbType(NpgsqlDbType npgsqlDbType)
   at Npgsql.NpgsqlParameter.ResolveHandler(ConnectorTypeMapper typeMapper)
   at Npgsql.NpgsqlParameterCollection.ValidateAndBind(ConnectorTypeMapper typeMapper)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior, Boolean async, CancellationToken cancellationToken)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior, Boolean async, CancellationToken cancellationToken)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior)
   at Program.<Main>$(String[] args) in C:\Users\natalie-perret\Desktop\Personal\playground\csharp\CSharpPlayground\Program.cs:line 32

Seems that the type needs to be registered for some reason, actually if I don't specify the type:

Unhandled exception. System.NotSupportedException: The CLR type System.Object isn't natively supported by Npgsql or your PostgreSQL. To use it with a PostgreSQL composite
 you need to specify DataTypeName or to map it, please refer to the documentation.
   at Npgsql.TypeMapping.ConnectorTypeMapper.ResolveByClrType(Type type)
   at Npgsql.TypeMapping.ConnectorTypeMapper.ResolveByClrType(Type type)
   at Npgsql.NpgsqlParameter.ResolveHandler(ConnectorTypeMapper typeMapper)
   at Npgsql.NpgsqlParameter.Bind(ConnectorTypeMapper typeMapper)
   at Npgsql.NpgsqlParameterCollection.ValidateAndBind(ConnectorTypeMapper typeMapper)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior, Boolean async, CancellationToken cancellationToken)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior, Boolean async, CancellationToken cancellationToken)
   at Npgsql.NpgsqlCommand.ExecuteReader(CommandBehavior behavior)
   at Program.<Main>$(String[] args) in C:\Users\natalie-perret\Desktop\Personal\playground\csharp\CSharpPlayground\Program.cs:line 31

---

[EDIT]

The temporary solution I've ended up with is to rely on the jsonb support and in particular the jsonb_to_recordset function (see the PostgreSQL documentation section about json functions ):

using System.Data;
using System.Text.Json;

using Npgsql;
using NpgsqlTypes;


var connectionStringBuilder = new NpgsqlConnectionStringBuilder
{
    Host     = "localhost",
    Port     = 5432,
    Username = "brand_payment_migration",
    Password = "secret",
    Database = "brand_payment"
};
using var connection = new NpgsqlConnection(connectionStringBuilder.ToString());
connection.Open();
using var command = connection.CreateCommand();
command.CommandText = 
"SELECT id, accounting_document, accounting_document_type, company_code, document_date_year, posting_view_item " +
"FROM sch_brand_payment_data_lake_proxy.supplier_balances " +
"LEFT OUTER JOIN sch_brand_payment_data_lake_proxy.supplier_balance_items " +
"ON supplier_balances.id = supplier_balance_items.supplier_balance_id " +
"WHERE (accounting_document, accounting_document_type, company_code, document_date_year) " +
"IN (SELECT * FROM jsonb_to_recordset(@values) " +
"AS params (accounting_document text, accounting_document_type text, company_code text, document_date_year integer));";
var parameter = command.CreateParameter();
parameter.ParameterName = "@values";
parameter.NpgsqlDbType = NpgsqlDbType.Jsonb;
parameter.NpgsqlValue = JsonSerializer.Serialize(new []
{
    new Params("A", "B", "C", 1), 
    new Params("A", "B", "C", 2)
});
command.Parameters.Add(parameter);
using var reader = command.ExecuteReader();
using var dataTable = new DataTable();
dataTable.Load(reader);
var cols = dataTable.Columns.Cast<DataColumn>().ToArray();
Console.WriteLine(string.Join(Environment.NewLine, cols.Select((x, i) => $"Col{i} = {x}")));
Console.WriteLine(string.Join("\t", cols.Select((_, i) => $"Col{i}")));
foreach (var dataRow in dataTable.Rows.Cast<DataRow>())
{
    Console.WriteLine(string.Join("\t", dataRow.ItemArray));
}


public Params(
    string accounting_document, 
    string accounting_document_type,
    string company_code,
    int document_date_year);

Output:

Col0 = id
Col1 = accounting_document
Col2 = accounting_document_type
Col3 = company_code
Col4 = document_date_year
Col5 = posting_view_item
Col0    Col1    Col2    Col3    Col4    Col5
2       A       B       C       1       A
3       A       B       C       2       A
3       A       B       C       2       B

But this comes at the cost of adding an additional step of json serialization when passing the parameters. So other than that and building an insanely long string, I'm kinda puzzled at the fact there is no way to pass directly the actual values, without extra steps, to the NpgsqlParameter.NpgsqlValue property.

[Edit 2]

Adding a DbFiddle

[Edit 3]

The same jsonb "trick" can be used for feeding the data (albeit, same hiccups I've already mentioned above):

INSERT INTO sch_brand_payment_data_lake_proxy.supplier_balances
    (accounting_document, accounting_document_type, company_code, document_date_year)
SELECT * FROM jsonb_to_recordset(
    '[{"accounting_document":"E","accounting_document_type":"B","company_code":"C","document_date_year":1},
      {"accounting_document":"E","accounting_document_type":"B","company_code":"C","document_date_year":2}]'::jsonb)
       AS params (accounting_document text, accounting_document_type text, company_code text, document_date_year integer)
RETURNING id;

Answer 1

[EDIT 4] Another way is to use jsonb_populate_recordset and pass the relevant NULL::table-full-name as the first parameter (which is gonna define the columns) and the relevant jsonb as the second parameter (akin to jsonb_to_recordset for the 1st parameter).

Basically the 3 main ways of achieving what I wanted (updated the DbFiddle accordingly ):

Note: things might get even easier with PostgreSQL 15 with the json_table feature .

[EDIT 3] This article makes a pretty good job a summing things up: https://dev.to/forbeslindesay/postgres-unnest-cheat-sheet-for-bulk-operations-1obg

[EDIT 2]

Following up the issue I've filed earlier today https://github.com/npgsql/npgsql/issues/4437#issuecomment-1113999994

I've settled for a solution / workaround mentioned by @dhedey in another, somehow, related issue :

If it helps anyone else, I have found quite a neat workaround for these types of queries using the UNNEST command, which can take multiple array parameters and zip them together into columns, which can be joined with the table to filter to the relevant columns.

The use of the join is also more performant than the ANY/IN pattern in some cases.

 SELECT * FROM table WHERE (itemauthor, itemtitle) = ANY (('bob', 'hello'), ('frank', 'hi')...)

Can be represented with:

 var authorsParameter = new NpgsqlParameter("@authors", NpgsqlDbType.Array | NpgsqlDbType.Varchar) { Value = authors.ToList() }; var titlesParameter = new NpgsqlParameter("@titles", NpgsqlDbType.Array | NpgsqlDbType.Varchar) { Value = titles.ToList() }; var results = dbContext.Set<MyRow>() .FromSqlInterpolated($@" SELECT t.* FROM UNNEST({authorsParameter}, {titlesParameter}) params (author, title) INNER JOIN table t ON t.author = params.author AND t.title = params.title ");

NB - the Varchar can be replaced by other types for parameters which are arrays of other types (eg Bigint) - check out the NpgsqlDbType enum for more details.

I've then rewritten some the code I posted originally and it seems that the unnest PostgreSQL function solution works like a charm. This is the answer I accept for the time being, it looks neater than the Json / JsonB which requires either further postgresql-json-specific mapping shenaniganeries or extract.

Though, I'm not yet exactly too sure about the performance implications:

• unnest involves that you map the difference
• jsonb_to_recordset requires an additional .NET Json serialization step, and in some instances, to explicitly map the output of jsonb_to_recordset to the relevant columns.

Both do not come for free. But I like that unnest makes it explicitly for each column (ie each set / collections of values of a bigger .NET type (tuples, records, classs, structs, etc.)) you're passing to the NpgsqlParameter.NpgsqlValue property which DB type is gonna be used via the NpgsqlDbType enum

using System.Data;

using Npgsql;
using NpgsqlTypes;


var connectionStringBuilder = new NpgsqlConnectionStringBuilder
{
    Host     = "localhost",
    Port     = 5432,
    Username = "brand_payment_migration",
    Password = "secret",
    Database = "brand_payment"
};
using var connection = new NpgsqlConnection(connectionStringBuilder.ToString());
connection.Open();

var selectStatement =
    "SELECT * FROM sch_brand_payment_data_lake_proxy.supplier_balances " +
    "WHERE (accounting_document, accounting_document_type, company_code, document_date_year) " +
    "IN (SELECT * FROM  unnest(" +
    "@accounting_document_texts, " +
    "@accounting_document_types, " +
    "@company_codes, " +
    "@document_date_years" +
    "))";

var insertStatement = 
    "INSERT INTO sch_brand_payment_data_lake_proxy.supplier_balances " +
    "(accounting_document, accounting_document_type, company_code, document_date_year) " + 
    "SELECT * FROM unnest(" +
    "@accounting_document_texts, " +
    "@accounting_document_types, " +
    "@company_codes, " +
    "@document_date_years" + 
    ") RETURNING id;";

var parameters = new (string Name, NpgsqlDbType DbType, object Value)[]
{
    ("@accounting_document_texts", NpgsqlDbType.Array | NpgsqlDbType.Text,    new[] {"G", "G", "G"}),
    ("@accounting_document_types", NpgsqlDbType.Array | NpgsqlDbType.Text,    new[] {"Y", "Y", "Y"}),
    ("@company_codes",             NpgsqlDbType.Array | NpgsqlDbType.Text,    new[] {"Z", "Z", "Z"}),
    ("@document_date_years",       NpgsqlDbType.Array | NpgsqlDbType.Integer, new[] {1, 2, 3})
};

connection.ExecuteNewCommandAndWriteResultToConsole(insertStatement, parameters);
connection.ExecuteNewCommandAndWriteResultToConsole(selectStatement, parameters);

public static class Extensions
{
    public static void AddParameter(this NpgsqlCommand command, string name, NpgsqlDbType dbType, object value)
    {
        var parameter = command.CreateParameter();
        parameter.ParameterName = name;
        parameter.NpgsqlDbType  = dbType;
        parameter.NpgsqlValue   = value;
        command.Parameters.Add(parameter);
    }

    public static NpgsqlCommand CreateCommand(this NpgsqlConnection connection, 
        string text, 
        IEnumerable<(string Name, NpgsqlDbType DbType, object Value)> parameters)
    {
        var command = connection.CreateCommand();
        command.CommandText = text;
        foreach (var (name, dbType, value) in parameters)
        {
            command.AddParameter(name, dbType, value);
        }

        return command;
    }
    public static void ExecuteAndWriteResultToConsole(this NpgsqlCommand command)
    {
        Console.WriteLine($"Executing command... {command.CommandText}");
        
        using var reader = command.ExecuteReader();
        using var dataTable = new DataTable();
        dataTable.Load(reader);
        var cols = dataTable.Columns.Cast<DataColumn>().ToArray();
        Console.WriteLine(string.Join(Environment.NewLine, cols.Select((x, i) => $"Col{i} = {x}")));
        Console.WriteLine(string.Join("\t", cols.Select((_, i) => $"Col{i}")));
        foreach (var dataRow in dataTable.Rows.Cast<DataRow>())
        {
            Console.WriteLine(string.Join("\t", dataRow.ItemArray));
        }
    }

    public static void ExecuteNewCommandAndWriteResultToConsole(this NpgsqlConnection connection, 
        string text,
        IEnumerable<(string Name, NpgsqlDbType DbType, object Value)> parameters)
    {
        using var command = connection.CreateCommand(text, parameters);
        command.ExecuteAndWriteResultToConsole();
    }
}

Output:

Executing command... INSERT INTO sch_brand_payment_data_lake_proxy.supplier_balances (accounting_document, accounting_document_type, company_code, document_date_year) SEL
ECT * FROM unnest(@accounting_document_texts, @accounting_document_types, @company_codes, @document_date_years) RETURNING id;
Col0 = id
Col0
28
29
30
Executing command... SELECT * FROM sch_brand_payment_data_lake_proxy.supplier_balances WHERE (accounting_document, accounting_document_type, company_code, document_date_y
ear) IN (SELECT * FROM  unnest(@accounting_document_texts, @accounting_document_types, @company_codes, @document_date_years))
Col0 = id
Col1 = accounting_document
Col2 = accounting_document_type
Col3 = company_code
Col4 = document_date_year
Col5 = accounting_doc_created_by_user
Col6 = accounting_clerk
Col7 = assignment_reference
Col8 = document_reference_id
Col9 = original_reference_document
Col10 = payment_terms
Col11 = supplier
Col12 = supplier_name
Col13 = document_date
Col14 = posting_date
Col15 = net_due_date
Col16 = created_on
Col17 = modified_on
Col18 = pushed_on
Col19 = is_modified
Col20 = is_pushed
Col0    Col1    Col2    Col3    Col4    Col5    Col6    Col7    Col8    Col9    Col10   Col11   Col12   Col13   Col14   Col15   Col16   Col17   Col18   Col19   Col20
28      G       Y       Z       1                                                                                                                       False   False
29      G       Y       Z       2                                                                                                                       False   False
30      G       Y       Z       3                                                                                                                       False   False

---

[EDIT 1]

Since @Charlieface pointed out that this isn't the appropriate answer, I figured it would be better to get the answer / information right from the npgsql maintainers / contributors.

Hence filing an issue on their GitHub repository: https://github.com/npgsql/npgsql/issues/4437

---

Original answer:

As of today, there is no way to pass among other things tuples or collections as composite "types" or via positional-slash-implicit "definitions", (which could be then used in a collection that would have been passed to the parameter value property), npgslq requires prior PostgreSQL type definitions (but still tuples and nested collections can't work out cause not regarded as safe enough by maintainers or at least one of them). https://github.com/npgsql/npgsql/issues/2154

As the exception says the corresponding composite is required in the database. This is because anonymous types are not mapped to records.

So, you should create a type and a struct which must be mapped to the type.

FYI, there is a similar issue #2097 to track mapping composites to value tuples.

But this would require a few other related devs for npgsql like #2097 which has been dropped the author / main contributed deemed as too brittle in https://github.com/dotnet/efcore/issues/14661#issuecomment-462440199

Note that after discussion in npgsql/npgsql#2097 we decided to drop this idea. C# value tuples don't have names, so any mapping to PostgreSQL composites would rely on field definition ordering, which seems quite dangerous/brittle.

I've finally decided to settle for the jsonb alternative, not a huge fan, but at least it allows to pass collections in a relatively secured way (as long as the serialization to pass the jsonb is under control).

But the initial way I envisioned doing is not something that can be done as of today.

---

Also one thing I've learnt along the way investigating while writing that post:

• There is a pretty good Slack server dedicated to PostgreSQL: postgresteam.slack.com
• A pretty nice guide about how to properly format SQL when asking for PostgreSQL-related help (albeit opinionated with the author opinions): https://www.depesz.com/2010/05/28/what-mistakes-you-can-avoid-when-looking-for-help-on-irc/
• A paste bin automatically formatting SQL following the author preferences: https://paste.depesz.com