Verify distributed decimal behavior on real DM MPP

docs/README_zh.md

@@ -23,7 +23,7 @@ dmPython 是达梦数据库（DM8）的原生 Python 驱动程序，遵循 [Pyth
 - **Best-effort（尽力支持）**：尚未纳入 CI 覆盖的扩展使用场景。
 - **Not guaranteed（不保证）**：生产 SLA 承诺、厂商认证兼容性与闭源组件支持协议。
 - **连接安全**：`ssl_path` 支持使用客户端证书与私钥连接启用加密的 DM8。目录需包含 `ca-cert.pem`、`client-cert.pem`、`client-key.pem`；服务端证书没有 SAN 的旧版本还需提供与服务端完全一致的 `server-cert.pem`。指定 `ssl_path` 时，未协商加密的连接会报错。非空的 `ssl_pwd`、`ukey_name`、`ukey_pin` 仍不支持。
-- **主备路由**：读写分离模式 1 和 4 已在本机 DM8 主备环境、自动提交模式下验证；通过双端点服务名建立的新连接也通过了自动接管后的回归。MPP 集群路由仍需对应环境回归。
+- **主备与 MPP**：读写分离模式 1 和 4 已在本机 DM8 主备环境、自动提交模式下验证；通过双端点服务名建立的新连接也通过了自动接管后的回归。MPP 全局和本地登录已在本机两节点集群验证分布式读写。
 
 ## 路线图与状态
 

docs/api-reference.md

@@ -61,7 +61,7 @@ dmPython.connect(
 - `host` 与 `server` 互斥（只允许设置一个）。
 - IPv6 地址使用方括号，例如 `server="[::1]"`；`dsn` 可写为 `"[::1]:5236"`。
 - `dmsvc_path` 指向包含 `dm_svc.conf` 的目录；连接时可把 `server` 设为配置文件中的服务名。双端点服务名配置 `LOGIN_MODE=1` 后，已在本机 DM8 主备环境验证故障自动接管后的**新连接**会选择晋升的新主库；已有连接的自动恢复尚未验证。
-- `mpp_login` 接受 `DSQL_MPP_LOGIN_GLOBAL` 或 `DSQL_MPP_LOGIN_LOCAL`；`rwseparate` 接受 `DSQL_RWSEPARATE_OFF`、`DSQL_RWSEPARATE_ON` 或 `DSQL_RWSEPARATE_ON2`，`rwseparate_percent` 范围为 0–100。这些选项在建连时传给底层驱动。读写分离模式 1 和 4 已在本机 DM8 主备环境中验证自动提交模式下的查询路由；MPP 只验证了单机 LOCAL 握手，集群路由尚待验证。
+- `mpp_login` 接受 `DSQL_MPP_LOGIN_GLOBAL` 或 `DSQL_MPP_LOGIN_LOCAL`；`rwseparate` 接受 `DSQL_RWSEPARATE_OFF`、`DSQL_RWSEPARATE_ON` 或 `DSQL_RWSEPARATE_ON2`，`rwseparate_percent` 范围为 0–100。这些选项在建连时传给底层驱动。读写分离模式 1 和 4 已在本机 DM8 主备环境中验证自动提交模式下的查询路由；MPP 全局和本地登录已在本机两节点集群验证分布式读写。
 - `ssl_path` 指向含 `ca-cert.pem`、`client-cert.pem`、`client-key.pem` 的目录。服务端证书没有 SAN 时还需提供准确的 `server-cert.pem`，用于证书固定校验；有 SAN 的证书按 CA 链和主机名校验。设置后若服务端未协商加密，连接失败。非空的 `ssl_pwd`、`ukey_name`、`ukey_pin` 暂不支持。
 - `user` 支持 `user/password@server:port[/schema][?catalog=...]` 形式。
 - `login_timeout` 以毫秒为单位，默认 5000，限制首次建连握手；设为 0 表示不限制。`connection_timeout` 以秒为单位，默认 0 不限制，限制 SQL 执行时间。

docs/test-results/2026-09-27-mpp.md

@@ -0,0 +1,31 @@
+# Real DM8 MPP login and distributed decimal regression (2026-09-27)
+
+An isolated two-node DM8 MPP cluster ran in local Orb with the 2025-09-24
+ARM image. Each node had a separately initialized database (`EP01`, `EP02`),
+matching database creation parameters, `MAL_INI=1`, `MPP_INI=1`, a shared
+`dmmal.ini`, and the same `dmmpp.ctl` generated from `dmmpp.ini`. Both nodes
+reached `SYSTEM IS READY`, and their logs showed MAL links to each other.
+This follows the [DM MPP configuration model](https://eco.dameng.com/document/dm/zh-cn/pm/basic-concepts-principles.html).
+
+The ARM macOS Python 3.10 extension connected to both endpoints with
+`mpp_login=DSQL_MPP_LOGIN_GLOBAL` and `mpp_login=DSQL_MPP_LOGIN_LOCAL`.
+`scripts/verify_dm_mpp.py` created a hash-distributed table, batch-inserted
+12 `DECIMAL(30,8)` rows, and checked the exact values from both endpoints.
+The values included a 20-digit integer part with eight fractional digits,
+negative values, and `0.00000001`.
+
+| Query | EP01 endpoint | EP02 endpoint |
+| --- | ---: | ---: |
+| Global login | 12 exact rows | 12 exact rows |
+| Local login | 6 exact rows | 6 exact rows |
+
+The two local row ID sets were disjoint and their union equaled the 12 global
+rows, proving that the test exercised both physical EPs. The script removed
+its test table after success. It takes `DM_MPP_USER`, `DM_MPP_PASSWORD`,
+`DM_MPP_EP1_HOST`, and `DM_MPP_EP2_HOST`; optional endpoint ports default to
+5236. The account needs permission to create and drop a distributed table.
+
+The test was run locally because GitHub-hosted runners cannot reach the Orb
+cluster. The existing CI matrix uses one DM8 instance and does not exercise
+MPP. This result covers a two-EP cluster and one DM8 server build; larger
+clusters and MPP primary/standby combinations remain untested.

scripts/verify_dm_mpp.py

@@ -0,0 +1,89 @@
+"""Verify local and global login against a real two-node DM MPP cluster.
+
+Set DM_MPP_USER, DM_MPP_PASSWORD, DM_MPP_EP1_HOST, and DM_MPP_EP2_HOST.
+DM_MPP_EP1_PORT and DM_MPP_EP2_PORT default to 5236. The account must be
+able to create and drop a distributed table.
+"""
+
+from __future__ import annotations
+
+import os
+import uuid
+from decimal import Decimal
+
+import dmPython
+
+
+HIGH = Decimal("12345678901234567890.12345678")
+
+
+def connect(host: str, port: int, mode: int):
+    return dmPython.connect(
+        user=os.environ["DM_MPP_USER"],
+        password=os.environ["DM_MPP_PASSWORD"],
+        server=host,
+        port=port,
+        mpp_login=mode,
+        autoCommit=dmPython.DSQL_AUTOCOMMIT_ON,
+        login_timeout=3000,
+    )
+
+
+def rows(conn, table: str):
+    with conn.cursor() as cur:
+        cur.execute(f"SELECT ID, CAST(AMOUNT AS VARCHAR(80)) FROM {table} ORDER BY ID")
+        return {row_id: Decimal(amount) for row_id, amount in cur.fetchall()}
+
+
+def local_instance(conn):
+    with conn.cursor() as cur:
+        cur.execute("SELECT INSTANCE_NAME FROM V$INSTANCE")
+        return cur.fetchone()[0]
+
+
+def main():
+    endpoints = (
+        (os.environ["DM_MPP_EP1_HOST"], int(os.environ.get("DM_MPP_EP1_PORT", "5236"))),
+        (os.environ["DM_MPP_EP2_HOST"], int(os.environ.get("DM_MPP_EP2_PORT", "5236"))),
+    )
+    table = "DMPY_MPP_" + uuid.uuid4().hex[:8].upper()
+    expected = {i: HIGH if i % 2 else -HIGH for i in range(1, 13)}
+    expected[12] = Decimal("0.00000001")
+
+    with connect(*endpoints[0], dmPython.DSQL_MPP_LOGIN_GLOBAL) as global_conn:
+        with global_conn.cursor() as cur:
+            cur.execute(
+                f"CREATE TABLE {table} (ID INT PRIMARY KEY, AMOUNT DECIMAL(30,8)) "
+                "DISTRIBUTED BY HASH(ID)"
+            )
+        try:
+            with global_conn.cursor() as cur:
+                cur.executemany(
+                    f"INSERT INTO {table} VALUES (?, ?)", list(expected.items())
+                )
+
+            local_rows = []
+            instance_names = []
+            for host, port in endpoints:
+                with connect(host, port, dmPython.DSQL_MPP_LOGIN_GLOBAL) as conn:
+                    assert rows(conn, table) == expected
+                with connect(host, port, dmPython.DSQL_MPP_LOGIN_LOCAL) as conn:
+                    instance_names.append(local_instance(conn))
+                    local_rows.append(rows(conn, table))
+
+            assert instance_names[0] != instance_names[1], instance_names
+            assert local_rows[0] and local_rows[1], "both EPs must store rows"
+            assert local_rows[0].keys().isdisjoint(local_rows[1].keys())
+            assert local_rows[0] | local_rows[1] == expected
+            print(
+                f"MPP global: {len(expected)} exact rows from both endpoints; "
+                f"local: {instance_names[0]}={len(local_rows[0])}, "
+                f"{instance_names[1]}={len(local_rows[1])}"
+            )
+        finally:
+            with global_conn.cursor() as cur:
+                cur.execute(f"DROP TABLE {table}")
+
+
+if __name__ == "__main__":
+    main()