AccessViolationException and Heap Corruption in SQLGetData during ETL operations on Windows

# AccessViolationException and Heap Corruption in SQLGetData during ETL operations on Windows

## Environment

| Component | Version |
|-----------|---------|
| **ODBC Driver** | Amazon Redshift ODBC **v2.1.8.0** and **v2.1.12.0** (64-bit) - tested both |
| **aws-sdk-cpp** | 1.11.598 (v2.1.8) / 1.11.711 (v2.1.12) |
| **OpenSSL** | 1.1.1zb / 1.1.1zd |
| **OS** | Windows Server 2019 Standard |
| **.NET Runtime** | .NET 8.0.11, CoreCLR 8.0.1124.51707 |
| **Redshift endpoint** | `pennylane-external.csqwamh5pldr.eu-west-1.redshift.amazonaws.com:5439` |

### DSN Configuration

```
Database: Prod
Auth Type: Standard
SSL Mode: verify-ca
Min TLS: 1.2
Proxy: None
Cursor Size: 20000 (also tested with 1000 - same issue)
Advanced Options:
- Autofetch Refcursors: Enabled
- Rollback on Error: Enabled
- Database Metadata Current Database: Enabled
Retry Delay: 3
```

---

## Description

**Expected behavior:**
ODBC driver should reliably return data from Redshift queries without memory corruption during sustained ETL operations.

**Observed behavior:**
The ODBC driver causes the host application to crash with memory corruption errors:
1. `System.AccessViolationException` in `SQLGetData` native interop
2. `STATUS_HEAP_CORRUPTION` (0xc0000374) in `ntdll.dll`

The crashes are **intermittent** - ETL jobs sometimes complete successfully, other times crash mid-extraction on different tables. This suggests a race condition or memory management issue in the native driver code.

**Impact:**
Production ETL services crash unexpectedly, blocking data pipelines. Multiple independent customers are affected.

---

## Steps to reproduce

1. Configure ODBC DSN with Amazon Redshift driver v2.1.12.0 on Windows Server 2019
2. Connect to Redshift cluster using Standard authentication with SSL (verify-ca)
3. Execute multiple sequential SELECT queries extracting data from various tables
4. Process results using `SQLGetData` to retrieve column values (especially strings)
5. Run ETL jobs on schedule (e.g., every hour) with cursor size 20000

The crash typically occurs:
- During multi-table extraction workflows
- When reading string column data via `SQLGetData`
- Intermittently - not always on the same table

---

## Error messages / Stack traces

### Error 1: AccessViolationException (Windows Event Log)

```
Application: ServerMyReport.exe
CoreCLR Version: 8.0.1124.51707
.NET Version: 8.0.11
Description: The process was terminated due to an unhandled exception.

Exception Info: System.AccessViolationException: Attempted to read or write
protected memory. This is often an indication that other memory is corrupt.

Stack:
at Interop+Odbc.SQLGetData(OdbcStatementHandle, UInt16, SQL_C, CNativeBuffer, IntPtr, IntPtr ByRef)
at System.Data.Odbc.OdbcDataReader.GetData(Int32, SQL_C, Int32, Int32 ByRef)
at System.Data.Odbc.OdbcDataReader.internalGetString(Int32)
at System.Data.Odbc.OdbcDataReader.GetValue(Int32, TypeMap)
at System.Data.Odbc.OdbcDataReader.GetValue(Int32)
[...application code...]
```

### Error 2: Heap Corruption (Windows Event Log)

```
Exception Code: 0xc0000374 (STATUS_HEAP_CORRUPTION)
Faulting Module: ntdll.dll
```

### ODBC Driver Log (redshift_odbc.log)

Recurring empty error messages observed across 6+ months:

```
[ERROR] 2026-01-27 16:39:59.496 [RSLIBPQ:rslibpq.c:472] [11896] pError=
[ERROR] 2026-01-27 16:39:41.242 [RSLIBPQ:rslibpq.c:472] [6540] pError=
[ERROR] 2026-01-27 15:43:50.676 [RSLIBPQ:rslibpq.c:472] [6244] pError=
...
```

Note: The error message after `pError=` is empty, which appears to be a logging issue in the driver.

Also observed streaming cursor operations before crashes:

```
[INFO] 2026-01-27 11:18:33.614 [RSUTIL:rsutil.c:10574] Skiping current result for streaming cursor...
[INFO] 2026-01-27 11:18:33.614 [RSUTIL:rsutil.c:10584] Skiping current result for streaming cursor done.iSocketError=0
```

(Note: "Skiping" appears to be a typo in the driver source code)

---

## Workarounds tried (all unsuccessful)

| Workaround | Result |
|------------|--------|
| Upgraded driver from v2.1.8.0 to v2.1.12.0 | Still crashes |
| Reduced cursor size from 20000 to 1000 | Still crashes |
| Limited to 1 CPU / 1 concurrent ETL query | Still crashes |
| Ensured proper Dispose of ODBC resources in .NET code | Still crashes |

---

## Additional context

- **Two independent customers** experiencing identical symptoms on different infrastructures
- The crash occurs in **native ODBC driver code** (`SQLGetData`), not in the managed .NET layer
- Both `AccessViolationException` and `STATUS_HEAP_CORRUPTION` indicate memory corruption in native code
- Issue persists across driver versions (v2.1.8.0 and v2.1.12.0)
- The intermittent nature and different failing tables suggest a race condition, buffer overflow, or use-after-free bug
- ODBC log shows empty `pError=` messages - possible logging bug masking the actual error

---

## Attachments

- `redshift_odbc.log` - Driver trace log spanning July 2025 to January 2026

[redshift_odbc.log](https://github.com/user-attachments/files/24961862/redshift_odbc.log)

Start by reproducing the intermittent SQLGetData failure during sequential string extraction on Windows, then inspect the native paths referenced in redshift_odbc.log, including rslibpq.c:472 and rsutil.c:10574-10584. Done means identifying the memory-corruption cause, preventing the AccessViolationException and heap corruption, and ensuring the relevant driver errors are no longer empty.

Written by the indexing model from the issue text.