前段时间需要进行大批量数据导入,DBA给提供的是CVS文件,但是每个CVS文件都好几个GB大小,直接进行load,数据库很慢还会产生内存不足的问题,为了实现这个功能,写了个快速切分文件的程序。
[Java]代码
import org.apache.log4j.LogManager; import org.apache.log4j.Logger; import java.io.*; import java.util.*; import java.util.concurrent.*; public class FileSplitUtil { private final static Logger log = LogManager.getLogger(FileSplitUtil.class); private static final long originFileSize = 1024 * 1024 * 100;// 100M private static final int blockFileSize = 1024 * 1024 * 64;// 防止中文乱码,必须取2的N次方 /** * CVS文件分隔符 */ private static final char cvsSeparator = '^'; public static void main(String args[]){ long start = System.currentTimeMillis(); try { String fileName = "D://csvtest//aa.csv"; File sourceFile = new File(fileName); if (sourceFile.length() >= originFileSize) { String cvsFileName = fileName.replaceAll("////", "/"); FileSplitUtil fileSplitUtil = new FileSplitUtil(); List<String> parts=fileSplitUtil.splitBySize(cvsFileName, blockFileSize); for(String part:parts){ System.out.println("partName is:"+part); } } System.out.println("总文件长度"+sourceFile.length()+",拆分文件耗时:" + (System.currentTimeMillis() - start) + "ms."); }catch (Exception e){ log.info(e.getStackTrace()); } } /** * 拆分文件 * * @param fileName 待拆分的完整文件名 * @param byteSize 按多少字节大小拆分 * @return 拆分后的文件名列表 */ public List<String> splitBySize(String fileName, int byteSize) throws IOException, InterruptedException { List<String> parts = new ArrayList<String>(); File file = new File(fileName); int count = (int) Math.ceil(file.length() / (double) byteSize); int countLen = (count + "").length(); RandomAccessFile raf = new RandomAccessFile(fileName, "r"); long totalLen = raf.length(); CountDownLatch latch = new CountDownLatch(count); for (int i = 0; i < count; i++) { String partFileName = file.getPath() + "." + leftPad((i + 1) + "", countLen, '0') + ".cvs"; int readSize=byteSize; long startPos=(long)i * byteSize; long nextPos=(long)(i+1) * byteSize; if(nextPos>totalLen){ readSize= (int) (totalLen-startPos); } new SplitRunnable(readSize, startPos, partFileName, file, latch).run(); parts.add(partFileName); } latch.await();//等待所有文件写完 //由于切割时可能会导致行被切断,加工所有的的分割文件,合并行 mergeRow(parts); return parts; } /** * 分割处理Runnable * * @author supeidong */ private class SplitRunnable implements Runnable { int byteSize; String partFileName; File originFile; long startPos; CountDownLatch latch; public SplitRunnable(int byteSize, long startPos, String partFileName, File originFile, CountDownLatch latch) { this.startPos = startPos; this.byteSize = byteSize; this.partFileName = partFileName; this.originFile = originFile; this.latch = latch; } public void run() { RandomAccessFile rFile; OutputStream os; try { rFile = new RandomAccessFile(originFile, "r"); byte[] b = new byte[byteSize]; rFile.seek(startPos);// 移动指针到每“段”开头 int s = rFile.read(b); os = new FileOutputStream(partFileName); os.write(b, 0, s); os.flush(); os.close(); latch.countDown(); } catch (IOException e) { log.error(e.getMessage()); latch.countDown(); } } } /** * 合并被切断的行 * * @param parts */ private void mergeRow(List<String> parts) { List<PartFile> partFiles = new ArrayList<PartFile>(); try { //组装被切分表对象 for (int i=0;i<parts.size();i++) { String partFileName=parts.get(i); File splitFileTemp = new File(partFileName); if (splitFileTemp.exists()) { PartFile partFile = new PartFile(); BufferedReader reader=new BufferedReader(new InputStreamReader(new FileInputStream(splitFileTemp),"gbk")); String firstRow = reader.readLine(); String secondRow = reader.readLine(); String endRow = readLastLine(partFileName); partFile.setPartFileName(partFileName); partFile.setFirstRow(firstRow); partFile.setEndRow(endRow); if(i>=1){ String prePartFile=parts.get(i - 1); String preEndRow = readLastLine(prePartFile); partFile.setFirstIsFull(getCharCount(firstRow+preEndRow)>getCharCount(secondRow)); } partFiles.add(partFile); reader.close(); } } //进行需要合并的行的写入 for (int i = 0; i < partFiles.size() - 1; i++) { PartFile partFile = partFiles.get(i); PartFile partFileNext = partFiles.get(i + 1); StringBuilder sb = new StringBuilder(); if (partFileNext.getFirstIsFull()) { sb.append("/r/n"); sb.append(partFileNext.getFirstRow()); } else { sb.append(partFileNext.getFirstRow()); } writeLastLine(partFile.getPartFileName(),sb.toString()); } } catch (Exception e) { log.error(e.getMessage()); } } /** * 得到某个字符出现的次数 * @param s * @return */ private int getCharCount(String s) { int count = 0; for (int i = 0; i < s.length(); i++) { if (s.charAt(i) == cvsSeparator) { count++; } } return count; } /** * 采用BufferedInputStream方式读取文件行数 * * @param filename * @return */ public int getFileRow(String filename) throws IOException { InputStream is = new BufferedInputStream(new FileInputStream(filename)); byte[] c = new byte[1024]; int count = 0; int readChars = 0; while ((readChars = is.read(c)) != -1) { for (int i = 0; i < readChars; ++i) { if (c[i] == '/n') ++count; } } is.close(); return count; } /** * 读取最后一行数据 * @param filename * @return * @throws IOException */ private String readLastLine(String filename) throws IOException { // 使用RandomAccessFile , 从后找最后一行数据 RandomAccessFile raf = new RandomAccessFile(filename, "r"); long len = raf.length(); String lastLine = ""; if(len!=0L) { long pos = len - 1; while (pos > 0) { pos--; raf.seek(pos); if (raf.readByte() == '/n') { lastLine = raf.readLine(); lastLine=new String(lastLine.getBytes("8859_1"), "gbk"); break; } } } raf.close(); return lastLine; } /** * 修改最后一行数据 * @param fileName * @param lastString * @return * @throws IOException */ private void writeLastLine(String fileName,String lastString){ try { // 打开一个随机访问文件流,按读写方式 RandomAccessFile randomFile = new RandomAccessFile(fileName, "rw"); // 文件长度,字节数 long fileLength = randomFile.length(); //将写文件指针移到文件尾。 randomFile.seek(fileLength); //此处必须加gbk,否则会出现写入乱码 randomFile.write(lastString.getBytes("gbk")); randomFile.close(); } catch (IOException e) { log.error(e.getMessage()); } } /** * 左填充 * * @param str * @param length * @param ch * @return */ public static String leftPad(String str, int length, char ch) { if (str.length() >= length) { return str; } char[] chs = new char[length]; Arrays.fill(chs, ch); char[] src = str.toCharArray(); System.arraycopy(src, 0, chs, length - src.length, src.length); return new String(chs); } /** * 合并文件行内部类 */ class PartFile { private String partFileName; private String firstRow; private String endRow; private boolean firstIsFull; public String getPartFileName() { return partFileName; } public void setPartFileName(String partFileName) { this.partFileName = partFileName; } public String getFirstRow() { return firstRow; } public void setFirstRow(String firstRow) { this.firstRow = firstRow; } public String getEndRow() { return endRow; } public void setEndRow(String endRow) { this.endRow = endRow; } public boolean getFirstIsFull() { return firstIsFull; } public void setFirstIsFull(boolean firstIsFull) { this.firstIsFull = firstIsFull; } } }
原创文章,作者:奋斗,如若转载,请注明出处:https://blog.ytso.com/10279.html