diff --git a/MesETL.App/Services/ETL/ZstReader.cs b/MesETL.App/Services/ETL/ZstReader.cs
index c9704f3..f3435c0 100644
--- a/MesETL.App/Services/ETL/ZstReader.cs
+++ b/MesETL.App/Services/ETL/ZstReader.cs
@@ -1,4 +1,6 @@
-using Microsoft.Extensions.Logging;
+using System.Runtime.InteropServices;
+using System.Text;
+using Microsoft.Extensions.Logging;
 using ZstdSharp;
 
 namespace MesETL.App.Services.ETL;
@@ -11,7 +13,12 @@ public class ZstReader : CsvReader
     protected new readonly Lazy<StreamReader> Reader;
     private Stream? _stream;
 
-
+    private readonly List<char> _str = new(1024);
+    private readonly char[] _charBuffer = new char[1024];
+    private int _charLen = 0;
+    private int _charPos = 0;
+    
+    
     public ZstReader(string filePath, string tableName, string[] headers, string delimiter = ",", char quoteChar = '\"', ILogger? logger = null) 
         : base(filePath, tableName, headers, delimiter, quoteChar, logger)
     {
@@ -20,6 +27,7 @@ public class ZstReader : CsvReader
              _stream = new DecompressionStream(File.OpenRead(filePath));
             return new StreamReader(_stream);
         }, false);
+        ReadBuffer();
     }
         
     public ZstReader(Stream stream, string tableName, string[] headers, string delimiter = ",", char quoteChar = '\"', ILogger? logger = null) 
@@ -27,16 +35,56 @@ public class ZstReader : CsvReader
     {
         var ds = new DecompressionStream(stream);
         Reader = new Lazy<StreamReader>(() => new StreamReader(ds), false);
+        ReadBuffer();
+    }
+
+    private int ReadBuffer()
+    {
+        _charLen = _charPos = 0;
+        _charLen = Reader.Value.ReadBlock(_charBuffer);
+        return _charLen;
     }
 
     public override async ValueTask<bool> ReadAsync()
     {
-        var str = await Reader.Value.ReadLineAsync();
-        if (string.IsNullOrWhiteSpace(str))
+        // 缓冲区已经读取完毕，并且流状态为EOF
+        if (_charPos == _charLen && ReadBuffer() == 0)
             return false;
 
-        var fields = ParseRowFaster(str, QuoteChar, Delimiter[0]);
-        Current = new DataRecord(fields, TableName, Headers);
+        do
+        {
+            // 读取缓冲区
+            var span = _charBuffer.AsSpan(_charPos, _charLen - _charPos);
+            var newLineIdx = span.IndexOfAny('\r', '\n');
+            // 读取到行，结合当前构建字符串转换进行转换
+            if (newLineIdx >= 0)
+            {
+                if (_str.Count == 0) // => 可以提高一点性能...
+                {
+                    var fields = ParseRowFaster(span[..newLineIdx], QuoteChar, Delimiter[0]);
+                    Current = new DataRecord(fields, TableName, Headers);
+                }
+                else
+                {
+                    _str.AddRange(span[..newLineIdx]);
+                    var fields = ParseRowFaster(CollectionsMarshal.AsSpan(_str), QuoteChar, Delimiter[0]);
+                    Current = new DataRecord(fields, TableName, Headers);
+                }
+                _str.Clear();
+
+                var ch = span[newLineIdx];
+                _charPos += newLineIdx + 1;
+                if (ch == '\r' && (_charPos < _charLen || ReadBuffer() > 0) && _charBuffer[_charPos] == '\n') // 跳过CRLF
+                    ++_charPos;
+                return true;
+            }
+            // 未读取到行，将缓冲区插入构建字符串
+            _str.AddRange(span);
+        } while (ReadBuffer() > 0);
+
+        var f = ParseRowFaster(CollectionsMarshal.AsSpan(_str), QuoteChar, Delimiter[0]);
+        Current = new DataRecord(f, TableName, Headers);
+        _str.Clear();
         return true;
     }