在线时间:8:00-16:00
迪恩网络APP
随时随地掌握行业动态
扫描二维码
关注迪恩网络微信公众号
软件名:DataPie 功能:支持SQL server、SQLite、ACCESS数据库的导入、导出、存储过程调用,支持EXCEL2007、EXCEL2003、ACCESS2007、 CSV文件导入数据库,支持EXCEL、CSV、ZIP、ACCESS文件方式导出,支持数据拆分导出及自定义SQL查询与导出。 开发背景:作者从事财务管理工作,主要是出具集团的内部财务报表,随着公司精细化管理的需求,管理报表的数据量急速增长, 依赖EXCEL加工处理数据已经变得极为困难,因此团队全面转向关系数据库进行数据处理,为减少财务人员使用数据库的难度,因此专门针对财务报表核算需要,开发了该工具。目前,我月度报表处理的数据量超过5G,最大的单次运算量记录接近千万,该工具主要发挥的作用就是将收集到的数据, 导入SQL SERVER数据库,进行报表运算,并且输出各类财务报表,对于几十万级的数据输入、输出基本上能够轻松应付。 源码下载地址:https://github.com/yfl8910/DataPie
.NET 6 版本:https://github.com/yfl8910/DataPie2
软件界面及主要代码: 1.导入界面,支持EXCEL2007、EXCEL2003、ACCESS2007、 CSV文件导入数据库,支持CSV文件夹整体导入相应表。财务工作的数据源较多,有业务提供的excel数据,也有从系统中取出的csv、excel或者text文档,为满足多种数据源的输入,所以多种数据源导入数据库。目前SQL SERVER导入整体效率较高,ACCESS下,excel文件导入效率最高(几十万的数据量可以很快导入),其他方式效率还有待提升。 涉及核心代码如下 public static DataTable GetDataTableFromFile(string path, string tname) { string ace = "Microsoft.ACE.OLEDB.12.0"; string jet = "Microsoft.Jet.OLEDB.4.0"; string xl2007 = "Excel 12.0 Xml"; string xl2003 = "Excel 8.0"; string imex = "IMEX=1"; string hdr = "Yes"; string conn = "Provider={0};Data Source={1};Extended Properties=\"{2};HDR={3};{4}\";"; string select = ""; string ext = Path.GetExtension(path); OleDbDataAdapter oda; DataTable dt = new DataTable(tname); switch (ext.ToLower()) { case ".xlsx": conn = String.Format(conn, ace, Path.GetFullPath(path), xl2007, hdr, imex); select = string.Format("SELECT * FROM [{0}$]", tname); break; case ".xls": conn = String.Format(conn, jet, Path.GetFullPath(path), xl2003, hdr, imex); select = string.Format("SELECT * FROM [{0}$]", tname); break; case ".accdb": conn = String.Format("Provider=Microsoft.ACE.OLEDB.12.0;Data Source= {0};Persist Security Info=False;", path); select = string.Format("SELECT * FROM [{0}]", tname); break; case ".csv": conn = String.Format(conn, ace, path.Substring(0, path.LastIndexOf('\\')), "text;Excel 12.0", hdr, imex); select = string.Format("SELECT * FROM [{0}]", Path.GetFileName(path)); break; default: throw new Exception("File Not Supported!"); } OleDbConnection con = new OleDbConnection(conn); con.Open(); oda = new OleDbDataAdapter(select, con); oda.Fill(dt); con.Close(); return dt; }
2.数据导出界面,支持EXCEL2007、CSV、ACCESS、zip压缩文件、分拆导出几个功能。我目前从事的岗位,报表整体数据量接近千万级,其中单表需要导出分发到业务手中的数据量最大也超过了百万级,目前达到百万级数据量的表导出,一般采用CSV文件或者压缩包的形式,或者分多个表导出。 涉及核心代码。excel导出主要源码 : public static void SaveExcel(string FileName, string sql, string SheetName) { FileInfo newFile = new FileInfo(FileName); if (newFile.Exists) { newFile.Delete(); newFile = new FileInfo(FileName); } using (ExcelPackage package = new ExcelPackage(newFile)) { try { ExcelWorksheet ws = package.Workbook.Worksheets.Add(SheetName); IDataReader reader = DBConfig.db.DBProvider.ExecuteReader(sql); ws.Cells["A1"].LoadFromDataReader(reader, true); } catch (Exception ex) { throw ex; } package.Save(); } } /// <summary> /// 工作簿中添加新的SHEET /// </summary> public static bool SaveExcel(ExcelPackage package, string sql, string SheetName) { try { ExcelWorksheet ws = package.Workbook.Worksheets.Add(SheetName); IDataReader reader = DBConfig.db.DBProvider.ExecuteReader(sql); ws.Cells["A1"].LoadFromDataReader(reader, true); return true; } catch (Exception ex) { throw ex; } } /// <summary> /// 单表格导出到一个EXCEL工作簿 /// </summary> public static int ExportExcel(string FileName, string sql, string SheetName) { if (FileName != null) { Stopwatch watch = Stopwatch.StartNew(); watch.Start(); SaveExcel(FileName, sql, SheetName); watch.Stop(); return Convert.ToInt32(watch.ElapsedMilliseconds / 1000); } return -1; } public static async Task<int> ExportExcelAsync(string FileName, string sql, string SheetName) { return await Task.Run( () => { return ExportExcel(FileName,sql,SheetName); } ); } /// <summary> /// 多表格导出到一个EXCEL工作簿 /// </summary> public static int ExportExcel(string[] TabelNameArray, string filename) { if (filename != null) { Stopwatch watch = Stopwatch.StartNew(); watch.Start(); FileInfo newFile = new FileInfo(filename); if (newFile.Exists) { newFile.Delete(); newFile = new FileInfo(filename); } using (ExcelPackage package = new ExcelPackage(newFile)) { for (int i = 0; i < TabelNameArray.Length; i++) { string sql = "select * from [" + TabelNameArray[i] + "]"; IDataReader reader = DBConfig.db.DBProvider.ExecuteReader(sql); SaveExcel(package, sql, TabelNameArray[i]); } package.Save(); } watch.Stop(); return Convert.ToInt32(watch.ElapsedMilliseconds / 1000); } return -1; } public static async Task<int> ExportExcelAsync(string[] TabelNameArray, string filename) { return await Task.Run( () => { return ExportExcel( TabelNameArray,filename); } ); } /// <summary> /// 分拆导出 /// </summary> public static int ExportExcel(string[] TabelNameArray, string filename, string[] whereSQLArr) { if (filename != null) { Stopwatch watch = Stopwatch.StartNew(); watch.Start(); FileInfo file = new FileInfo(filename); int wherecount = whereSQLArr.Length; int count = TabelNameArray.Length; for (int i = 0; i < wherecount; i++) { string s = filename.Substring(0, filename.LastIndexOf("\\")); StringBuilder newfileName = new StringBuilder(s); int index = whereSQLArr[i].LastIndexOf("="); string sp = whereSQLArr[i].Substring(index + 2, whereSQLArr[i].Length - index - 3); newfileName.Append("\\" + file.Name.Substring(0, file.Name.LastIndexOf(".")) + "_" + sp + ".xlsx"); FileInfo newFile = new FileInfo(newfileName.ToString()); if (newFile.Exists) { newFile.Delete(); newFile = new FileInfo(newfileName.ToString()); } for (int j = 0; j < count; j++) { string sql = "select * from [" + TabelNameArray[j] + "]" + whereSQLArr[i]; IDataReader reader = DBConfig.db.DBProvider.ExecuteReader(sql); SaveExcel(newfileName.ToString(), sql, TabelNameArray[j]); } } watch.Stop(); return Convert.ToInt32(watch.ElapsedMilliseconds / 1000); } return -1; } public static async Task<int> ExportExcelAsync(string[] TabelNameArray, string filename, string[] whereSQLArr) { return await Task.Run( () => { return ExportExcel(TabelNameArray,filename,whereSQLArr); } ); } csv导出主要源码 (其他代码请参github上的源码): private static void WriteHeader(IDataReader reader, StreamWriter sw) { for (int i = 0; i < reader.FieldCount; i++) { if (i > 0) sw.Write(','); sw.Write(reader.GetName(i)); } sw.Write(Environment.NewLine); } private static void WriteContent(IDataReader reader, StreamWriter sw) { for (int i = 0; i < reader.FieldCount; i++) { if (i > 0) sw.Write(','); String v = reader[i].ToString(); if (v.Contains(',') || v.Contains('\n') || v.Contains('\r') || v.Contains('"')) { sw.Write('"'); sw.Write(v.Replace("\"", "\"\"")); sw.Write('"'); } else { sw.Write(v); } } sw.Write(Environment.NewLine); } public static int SaveCsv(IDataReader reader, string filename) { Stopwatch watch = Stopwatch.StartNew(); watch.Start(); StreamWriter sw = new StreamWriter(filename, false, Encoding.GetEncoding("gb2312")); WriteHeader(reader, sw); while (reader.Read()) { WriteContent(reader, sw); } sw.Flush(); sw.Close();//释放资源 watch.Stop(); return Convert.ToInt32(watch.ElapsedMilliseconds / 1000); } public static async Task<int> ExportCsvAsync(IDataReader reader, string filename) { return await Task.Run( () => { return SaveCsv( reader, filename);} ); } public static StreamWriter GetStreamWriter(string filename, int outCount) { string s = filename.Substring(0, filename.LastIndexOf(".")); StringBuilder newfileName = new StringBuilder(s); newfileName.Append(outCount + 1 + ".csv"); FileInfo newFile = new FileInfo(newfileName.ToString()); if (newFile.Exists) { newFile.Delete(); newFile = new FileInfo(newfileName.ToString()); } StreamWriter sw = new StreamWriter(newfileName.ToString(), false, Encoding.GetEncoding("gb2312")); return sw; } public static int SaveCsv(IDataReader reader, string filename, int pagesize) { int innerCount = 0, outCount = 0; Stopwatch watch = Stopwatch.StartNew(); watch.Start(); StreamWriter sw = GetStreamWriter(filename, outCount); WriteHeader(reader, sw); while (reader.Read()) { if (innerCount < pagesize) { WriteContent(reader, sw); innerCount++; } else { innerCount = 0; outCount++; sw.Flush(); sw.Close(); sw = GetStreamWriter(filename, outCount); WriteHeader(reader, sw); WriteContent(reader, sw); innerCount++; } } sw.Flush(); sw.Close(); watch.Stop(); return Convert.ToInt32(watch.ElapsedMilliseconds / 1000); } public static async Task<int> ExportCsvAsync(IDataReader reader, string filename, int pagesize) { return await Task.Run(() => { return SaveCsv(reader, filename,pagesize); }); }
3.设置拆分导出条件。财务工作中,涉及向业务单位分发各自的财务数据,如果采用手工筛选,然后分发效率较低,所以开发此功能,快速实现数据的拆分发送。
4.自定义SQL代码导出。通过双击左边的表名即可自动生成导出SQL脚本,点击添加条件,还可以增加导出的条件。该工具在数据核查时,使用比较方便,虽然SQL server数据库也自带相应的功能,但是导出功能缺无法满足我的需求,所以开发一个方便自己使用的界面还是很有必要的。
其他各项功能,及代码请参考github上的源码。
|
2023-10-27
2022-08-15
2022-08-17
2022-09-23
2022-08-13
请发表评论