在开发中积攒下来的业务工具类,方便快速编写业务。
maven
<dependency>
<groupId>com.github.hwywl</groupId>
<artifactId>business-tools</artifactId>
<version>1.0.3-RELEASE</version>
</dependency>
Gradle
implementation 'com.github.hwywl:business-tools:1.0.3-RELEASE'
通过反射自动生成MessageType,自动拼装数据结构,遵循简单的原则,你给我数据我给你parquet文件。 目前支持的数据类型如下:
- int、Integer
- long、Long
- float、Float
- double、Double
- String
可以查看github中的源码,不看也行,很简单的,使用如下:
public class ParquetUtilTest {
String filePath = "F:\\a.parquet";
/**
* 写入一个数据到parquet文件
*
* @throws IOException
* @throws CustomException
*/
@Test
public void ParquetWriteTest() throws IOException, CustomException, IntrospectionException {
TestModel model = getModel();
ParquetUtil.writerParquet(filePath, model);
}
/**
* 写入多个数据到parquet文件
*
* @throws IOException
* @throws CustomException
*/
@Test
public void ParquetWriteListTest() throws IOException, CustomException, IntrospectionException {
List<TestModel> models = getModels();
ParquetUtil.writerParquet(filePath, models);
}
/**
* 开放原始API,可以用于非一次性写入
*
* @throws IOException
* @throws CustomException
* @throws IntrospectionException
*/
@Test
public void ParquetWriteGroupListTest() throws IOException, CustomException, IntrospectionException {
ParquetWriter<Group> writer = ParquetUtil.getParquetWriter(filePath, TestModel.class);
Group group = ParquetUtil.getGroup(getModel());
writer.write(group);
// close写出文件
writer.close();
}
/**
* 将parquet文件转为对象集合
*
* @throws IOException
*/
@Test
public void ParquetReadBeanTest() throws IOException {
List<TestModel> models = ParquetUtil.readParquetBean(filePath, TestModel.class);
System.out.println(models);
}
/**
* 将parquet文件转为Map集合
*
* @throws IOException
*/
@Test
public void ParquetReadMapTest() throws IOException {
List<Map<String, Object>> maps = ParquetUtil.readParquetMap(filePath);
System.out.println(maps);
}
/**
* 将parquet文件转为对象集合,并只拿2条数据
*
* @throws IOException
*/
@Test
public void ParquetReadBeanMaxLineTest() throws IOException {
List<TestModel> models = ParquetUtil.readParquetBean(filePath, 2, TestModel.class);
System.out.println(models);
}
/**
* 测试数据
*
* @return
*/
private TestModel getModel() {
return new TestModel(2, 3, 6L, "校花", 10);
}
/**
* 测试数据
*
* @return
*/
private List<TestModel> getModels() {
List<TestModel> arrayList = new ArrayList<>();
for (int i = 0; i < 10; i++) {
arrayList.add(new TestModel(2, 3, 6L, "校花", 10));
}
return arrayList;
}
}
public class HtmlTest {
@Test
public void htmlReadTest() {
String template = EmailTemplateUtil.htmlMapTemplate("测试", getModels());
System.out.println(template);
}
/**
* 测试数据
*
* @return 测试数据
*/
private List<Map<String, Object>> getModels() {
List<Map<String, Object>> arrayList = new ArrayList<>();
for (int i = 0; i < 10; i++) {
Map<String, Object> map = new LinkedHashMap<>();
map.put("校花", "张花");
map.put("美女", "李花");
map.put("小萝莉", "茉莉花");
map.put("御姐", null);
map.put("班花", 111);
arrayList.add(map);
}
return arrayList;
}
}
public class HtmlTest {
@Test
public void htmlReadBeanTest() {
String template = EmailTemplateUtil.htmlBeanTemplate("测试", getBeanModels());
System.out.println(template);
}
/**
* 测试数据
*
* @return
*/
private List<HtmlModel> getBeanModels() {
List<HtmlModel> arrayList = new ArrayList<>();
for (int i = 0; i < 10; i++) {
arrayList.add(new HtmlModel(2, 3, 6L, "校花", 10));
}
return arrayList;
}
}
HtmlModel类
public class HtmlModel {
@HtmlAnnotation("游戏id")
int app_id;
@HtmlAnnotation("VIP等级")
Integer svip_level;
@HtmlAnnotation("VIP剩余时间")
Long svip_remain;
@HtmlAnnotation("游戏名称")
String name;
@HtmlAnnotation("充值金额")
double money;
}
在我们业务中很多场景为了速度是需要多线程并行运算的,最后再把各个结果集合起来返回。 示例:
public class ParallelThreadTest {
/**
* 测试并发执行运算
*/
@Test
public void parallelTest() throws InterruptedException {
List<HtmlModel> list = new ArrayList<>();
// 模拟四个线程并发执行,最后得到一个结果集
Runnable runnable1 = () -> list.addAll(getBeanModels());
Runnable runnable2 = () -> list.addAll(getBeanModels());
Runnable runnable3 = () -> list.addAll(getBeanModels());
Runnable runnable4 = () -> list.addAll(getBeanModels());
List<Runnable> tasks = Arrays.asList(runnable1, runnable2, runnable3, runnable4);
int secs = ParallelThread.parallelThreadTask(tasks);
// 这句代码在正式环境中没有意义,这里是为了防止主线程结束
// Junit不支持多线程
Thread.sleep(3000);
System.out.println("线程耗时:" + secs + "毫秒");
System.out.println(list);
}
/**
* 测试数据
*
* @return
*/
private List<HtmlModel> getBeanModels() {
List<HtmlModel> arrayList = new ArrayList<>();
for (int i = 0; i < 2; i++) {
arrayList.add(new HtmlModel(2, 3, 6L, "校花", 10));
}
return arrayList;
}
}
- 增加一个简单的多线程并发执行工具
- 升级开源工具类
- 开放parquet文件写入的原始API
- 修复写入parquet文件乱序问题
- 增加parquet文件的读写
- 增加html表格的生成。
- 工具还有不足之处,请大家Issues ヾ(๑╹◡╹)ノ"
- 我那么可爱你不点个star吗 φ(>ω<*)
- 联系我的邮箱:ilovey_hwy@163.com
- 我的博客:https://www.hwy.ac.cn
- GitHub:https://github.com/HWYWL