不使用Sqoop流程，利用CacheManager直接完成SparkSQL数据流直接回写Oracle
Posted
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了不使用Sqoop流程，利用CacheManager直接完成SparkSQL数据流直接回写Oracle相关的知识，希望对你有一定的参考价值。
以前都是使用Sqoop来完成数据从生成的hdfs数据存储上来抽取至oracle的数据库：sqoop抽取语句：
sqoop export --connect "jdbc:oracle:thin:@ip:port:sid" --username 用户名 --password 密码 --table sid.表名 --export-dir hdfs://nameservice1/user/XXX(hdfs地址) --fields-terminated-by "\001" --null-non-string ‘‘ --null-string ‘‘ -m 10;
由于项目需求我们现在要完成在代码中省城所需字段之后，直接回写到oracle中，因为数据量每天都很大，用实例或者List存有很大的局限性，可能会出现内存异常等不可预料的东西，所以我通过缓存器机制来存储数据，然后进行生成结果的临时表直接回写（后面做的hbase接口封装批量提交也比较类似）
废话不多说直接上代码：
1、建立缓存实体
package usi.java.oracle;/**
- @author HK
-
@date 2011-2-15 下午06:45:57
*/
public class Cache {
private String key;
private Object value;
private long timeOut;
private boolean expired;
public Cache() {
super();
}public Cache(String key, String value, long timeOut, boolean expired) {
this.key = key;
this.value = value;
this.timeOut = timeOut;
this.expired = expired;
}public String getKey() {
return key;
}public long getTimeOut() {
return timeOut;
}public Object getValue() {
return value;
}public void setKey(String string) {
key = string;
}public void setTimeOut(long l) {
timeOut = l;
}public void setValue(Object object) {
value = object;
}public boolean isExpired() {
return expired;
}public void setExpired(boolean b) {
expired = b;
}
}
2、建立缓存控制器
package usi.java.oracle;
import java.util.Date;
import java.util.HashMap;
/**
- @author HK
-
@date 2011-2-15 下午09:40:00
*/
public class CacheManager {private static HashMap cacheMap = new HashMap();
/**
-
This class is singleton so private constructor is used.
*/
private CacheManager() {
super();
}/**
- returns cache item from hashmap
- @param key
-
@return Cache
*/
private synchronized static Cache getCache(String key) {
return (Cache)cacheMap.get(key);
}/**
- Looks at the hashmap if a cache item exists or not
- @param key
-
@return Cache
*/
private synchronized static boolean hasCache(String key) {
return cacheMap.containsKey(key);
}/**
-
Invalidates all cache
*/
public synchronized static void invalidateAll() {
cacheMap.clear();
}/**
- Invalidates a single cache item
-
@param key
*/
public synchronized static void invalidate(String key) {
cacheMap.remove(key);
}/**
- Adds new item to cache hashmap
- @param key
-
@return Cache
*/
private synchronized static void putCache(String key, Cache object) {
cacheMap.put(key, object);
}/**
- Reads a cache item‘s content
- @param key
-
@return
*/
public static Cache getContent(String key) {
if (hasCache(key)) {
Cache cache = getCache(key);
if (cacheExpired(cache)) {
cache.setExpired(true);
}
return cache;
} else {
return null;
}
}/**
- @param key
- @param content
-
@param ttl
*/
public static void putContent(String key, Object content, long ttl) {
Cache cache = new Cache();
cache.setKey(key);
cache.setValue(content);
cache.setTimeOut(ttl + new Date().getTime());
cache.setExpired(false);
putCache(key, cache);
}/* @modelguid {172828D6-3AB2-46C4-96E2-E72B34264031} /
private static boolean cacheExpired(Cache cache) {
if (cache == null) {
return false;
}
long milisNow = new Date().getTime();
long milisExpire = cache.getTimeOut();
if (milisExpire < 0) { // Cache never expires
return false;
} else if (milisNow >= milisExpire) {
return true;
} else {
return false;
}
}
-
}
3、建立需要导出数据对象
package usi.java.oracle;
public class TaskAll {
private String mme_eid;
private String mme_editor;
private String entitytype_eid;
private String project_eid;
private String resource_eid;
public String getMme_eid() {
return mme_eid;
}
public void setMme_eid(String mme_eid) {
this.mme_eid = mme_eid;
}
public String getMme_editor() {
return mme_editor;
}
public void setMme_editor(String mme_editor) {
this.mme_editor = mme_editor;
}
public String getEntitytype_eid() {
return entitytype_eid;
}
public void setEntitytype_eid(String entitytype_eid) {
this.entitytype_eid = entitytype_eid;
}
public String getProject_eid() {
return project_eid;
}
public void setProject_eid(String project_eid) {
this.project_eid = project_eid;
}
public String getResource_eid() {
return resource_eid;
}
public void setResource_eid(String resource_eid) {
this.resource_eid = resource_eid;
}
}
5、执行逻辑主体，回写数据，批量提交
package usi.java.oracle;
import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
//import java.sql.ResultSet;
import java.util.List;
import org.apache.spark.SparkConf;
import org.apache.spark.SparkContext;
import org.apache.spark.sql.DataFrame;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.hive.HiveContext;
public class redict_to_171ora {
public static void main(String[] args) {
SparkConf sc = new SparkConf().setAppName("redict_to_171ora");
SparkContext jsc = new SparkContext(sc);
HiveContext hc = new HiveContext(jsc);
String hivesql1="select t.mme_eid,t.mme_editor,t.entitytype_eid,t.project_eid,t.resource_eid from usi_odso.c_taskall t limit 150000";
DataFrame redict_to_171ora= hc.sql(hivesql1);
//redict_to_171ora.registerTempTable("hivesql1");
List<Row> collect=redict_to_171ora.javaRDD().collect();
int o=0;
for (Row lists: collect){
TaskAll task=new TaskAll();
task.setMme_eid(lists.getString(0));
task.setMme_editor(lists.getString(1));
task.setEntitytype_eid(lists.getString(2));
task.setProject_eid(lists.getString(3));
task.setResource_eid(lists.getString(4));
CacheManager.putContent(o+"", task, 30000000);
o++;
/* System.out.println(lists.size());
System.out.println(lists.getString(0));
System.out.println(lists.getString(1));
System.out.println(lists.getString(2));
System.out.println(lists.getString(3));
System.out.println(lists.getString(4));*/
}
System.out.println(o);
Connection con = null;// 创建一个数据库连接
PreparedStatement pre = null;// 创建预编译语句对象，一般都是用这个而不用Statement
//ResultSet result = null;// 创建一个结果集对象
try
{
Class.forName("oracle.jdbc.driver.OracleDriver");// 加载Oracle驱动程序
System.out.println("开始尝试连接数据库！");
String url = "jdbc:oracle:" + "thin:@ip:1521:sid";// 127.0.0.1是本机地址，XE是精简版Oracle的默认数据库名
String user = "user";// 用户名,系统默认的账户名
String password = "password";// 你安装时选设置的密码
con = DriverManager.getConnection(url, user, password);// 获取连接
System.out.println("连接成功！");
String sql = "insert into c_taskall_test(mme_eid,mme_editor,entitytype_eid,project_eid,resource_eid) values(?,?,?,?,?)";// 预编译语句，“？”代表参数
pre = con.prepareStatement(sql);// 实例化预编译语句
for(int i=0;i<o;i++){
// for (Row lists: collect){
// String sql = "insert into c_taskall_test(mme_eid,mme_editor,entitytype_eid,project_eid,resource_eid) values(‘"+task.getMme_eid()+"‘,‘"+task.getMme_editor()+"‘,‘"+task.getEntitytype_eid()+"‘,‘"+task.getProject_eid()+"‘,‘"+task.getResource_eid()+"‘)";// 预编译语句，“？”代表参数
// pre.setString(1, "三星");// 设置参数，前面的1表示参数的索引，而不是表中列名的索引
TaskAll task=(TaskAll) CacheManager.getContent(""+i).getValue();
pre.setString(1, task.getMme_eid());
pre.setString(2, task.getMme_editor());
pre.setString(3, task.getEntitytype_eid());
pre.setString(4, task.getProject_eid());
pre.setString(5, task.getResource_eid());
pre.addBatch();
if(i%20000==0){//可以设置不同的大小；如50，100，500，1000等等
pre.executeBatch();
con.commit();
pre.clearBatch();
// System.out.println("i的值"+i);
}
// result = pre.executeQuery();// 执行查询，注意括号中不需要再加参数
}
pre.executeBatch();
con.commit();
pre.clearBatch();
// System.out.println("i的值"+i);
/* if (result != null)
result.close();*/
if (pre != null)
pre.close();
/* while (result.next())
// 当结果集不为空时
System.out.println("usernum:" + result.getString("usernum") + "flow:"
+ result.getString("flow"));*/
}
catch (Exception e)
{
e.printStackTrace();
}
finally
{
try
{
// 逐一将上面的几个对象关闭，因为不关闭的话会影响性能、并且占用资源
// 注意关闭的顺序，最后使用的最先关闭
/* if (result != null)
result.close();*/
if (pre != null)
pre.close();
if (con != null)
con.close();
//System.out.println("数据库连接已关闭！");
}
catch (Exception e)
{
e.printStackTrace();
}
}
}}
以上是关于不使用Sqoop流程，利用CacheManager直接完成SparkSQL数据流直接回写Oracle的主要内容，如果未能解决你的问题，请参考以下文章