import java.io.InputStream;
import java.util.Arrays;
import java.util.Iterator;
import java.util.List;
import java.util.Properties;
import org.apache.log4j.Logger;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import org.json.JSONArray;
import org.json.JSONException;
import org.json.JSONObject;
public class MysqlSparkConnector {
private static Logger logger = Logger.getLogger(MysqlSparkConnector.class);
private static String confilePrefix = "druid.properties";
private static String driver = null;
private static String dbHost = null;
private static String dbPort = null;
private static String username = null;
private static String password = null;
private static String environmentArray[] = { "development", "query", "master", "sandbox" };
/**
* 用Spark从MySQL中查询数据
* 可在同一个Mysql数据库中联表查询
*
* @param spark
* @param dbEnv
* @param dbName
* @param tables
* @param sql
* @return
*/
public static Dataset<Row> queryBySparkFromMysql(SparkSession spark, String dbEnv, String dbName, String[] tables,
String sql) {
if (!readProperties(dbEnv, dbName)) {
logger.error("read properties error, please check configuration file: " + dbName + "." + dbEnv + "." + confilePrefix);
return null;
}
String url = "jdbc:mysql://" + dbHost + ":" + dbPort + "/" + dbName;
Properties connectionProperties = new Properties();
connectionProperties.put("driver", driver);
connectionProperties.put("user", username);
connectionProperties.put("password", password);
for (int i = 0; i < tables.length; i++) {
spark.read().jdbc(url, tables[i], connectionProperties).createOrReplaceTempView(tables[i]);
}
return spark.sql(sql);
}
/**
* 用Spark从MySQL中查询数据
* 分片查询数据量大的表
*
* @param spark
* @param dbEnv
* @param dbName
* @param table
* @param sql
* @param partitionColumn 用于分片的字段，必须是整数类型
* @param lowerBound 分片的下界
* @param upperBound 分片的上界
* @param numPartitions 要创建的分片数
* @return
*/
public static Dataset<Row> queryBySparkFromMysql(SparkSession spark, String dbEnv, String dbName, String[] tables,
String sql, String partitionColumn, long lowerBound, long upperBound, int numPartitions) {
if (!readProperties(dbEnv, dbName)) {
logger.error("read properties error, please check configuration file: " + dbName + "." + dbEnv + "." + confilePrefix);
return null;
}
String url = "jdbc:mysql://" + dbHost + ":" + dbPort + "/" + dbName;
Properties connectionProperties = new Properties();
connectionProperties.put("driver", driver);
connectionProperties.put("user", username);
connectionProperties.put("password", password);
/**
* Spark默认的jdbc并发度为1，所有的数据都会在一个partition中进行操作，无论你给的资源有多少，只有会有一个task在执行任务，如果查询的表比较大的话，很容易就会内存溢出
* 所以当数据量达到百万甚至千万以上时，就需要对数据源进行分片查询，即多个进程同时查询部分数据，避免一次性查出的数据过多导致内存溢出
* 如：
* partitionColumn : id
* lowerBound : 0
* upperBound : 10000
* numPartitions : 10
* 只会找出(upperBound-lowerBound)=10000条记录，根据id字段分numPartitions=10次查询，每次找出(upperBound-lowerBound)/numPartitions=1000条记录
* select * from table where id between 0 and 1000
* select * from table where id between 1000 and 2000
* ......
* select * from table where id between 9000 and 10000
*/
for (int i = 0; i < tables.length; i++) {
spark.read().jdbc(url, tables[i], partitionColumn, lowerBound, upperBound, numPartitions, connectionProperties).createOrReplaceTempView(tables[i]);
}
return spark.sql(sql);
}
/**
* 用Spark从MySQL中查询数据
* 可连接多个Mysql数据库后联表查询
*
* {
* "cets_swapdata": [
* {
* "isPartition": true,
* "partitionInfo": {
* "upperBound": 500000000,
* "numPartitions": 100,
* "partitionColumn": "id",
* "lowerBound": 0
* },
* "table": "fba_inventory"
* }
* ],
* "cets": [
* {
* "isPartition": false,
* "table": "stock_location"
* }
* ]
* }
*
* @param spark
* @param dbEnv
* @param dbTableJson
* @param sql
* @return
* @throws JSONException
*/
public static Dataset<Row> jointQueryBySparkFromMysql(SparkSession spark, String dbEnv, JSONObject dbTableJson,
String sql) {
String url = null;
Properties connectionProperties = null;
String dbName = null;
JSONArray tablesJSONArray = null;
JSONObject tableInfo = null;
String tableName = null;
boolean isPartition = false;
JSONObject partitionInfo = null;
String partitionColumn = null;
long lowerBound = 0;
long upperBound = 100;
int numPartitions = 1;
try {
@SuppressWarnings("unchecked")
Iterator<String> iterator = dbTableJson.keys();
while (iterator.hasNext()) {
dbName = iterator.next();
tablesJSONArray = dbTableJson.getJSONArray(dbName);
if (!readProperties(dbEnv, dbName)) {
logger.error("read properties error, please check configuration file: " + dbName + "." + dbEnv + "." + confilePrefix);
return null;
}
url = "jdbc:mysql://" + dbHost + ":" + dbPort + "/" + dbName;
connectionProperties = new Properties();
connectionProperties.put("driver", driver);
connectionProperties.put("user", username);
connectionProperties.put("password", password);
for (int i = 0; i < tablesJSONArray.length(); i++) {
tableInfo = tablesJSONArray.getJSONObject(i);
tableName = tableInfo.getString("table");
isPartition = tableInfo.getBoolean("isPartition");
if (isPartition) {
partitionInfo = tableInfo.getJSONObject("partitionInfo");
// 数据量大的表，分片读取
partitionColumn = partitionInfo.getString("partitionColumn");
lowerBound = partitionInfo.getLong("lowerBound");
upperBound = partitionInfo.getLong("upperBound");
numPartitions = partitionInfo.getInt("numPartitions");
spark.read().jdbc(url, tableName, partitionColumn, lowerBound, upperBound, numPartitions, connectionProperties).createOrReplaceTempView(tableName);
} else {
// 数据量小的表，一次读取
spark.read().jdbc(url, tableName, connectionProperties).createOrReplaceTempView(tableName);
}
}
}
} catch (Exception e) {
e.printStackTrace();
}
return spark.sql(sql);
}
/**
* 用Spark从MySQL中查询数据
* 可以根据任意字段分片查询
*
* @param spark
* @param dbEnv
* @param dbName
* @param tableName
* @param predicates 查询条件，每个条件都是一个task，即分片数==数组长度
* @return
*/
public static Dataset<Row> queryBySparkFromMysqlByConditions(SparkSession spark, String dbEnv, String dbName,
String tableName, String[] predicates) {
if (!readProperties(dbEnv, dbName)) {
logger.error("read properties error, please check configuration file: " + dbName + "." + dbEnv + "." + confilePrefix);
return null;
}
String url = "jdbc:mysql://" + dbHost + ":" + dbPort + "/" + dbName;
Properties connectionProperties = new Properties();
connectionProperties.put("driver", driver);
connectionProperties.put("user", username);
connectionProperties.put("password", password);
/**
* 根据查询条件predicates分片进行查询，如根据时间类型的字段modifiedOn，一个task查询一天的数据：
* String[] predicates = {
* "modifiedOn >= '2017-09-02' and modifiedOn < '2017-09-03'",
* "modifiedOn >= '2017-09-03' and modifiedOn < '2017-09-04'",
* "modifiedOn >= '2017-09-04' and modifiedOn < '2017-09-05'",
* ......
* "modifiedOn >= '2017-09-17' and modifiedOn < '2017-09-18'",
* "modifiedOn >= '2017-09-18' and modifiedOn < '2017-09-19'"
* }
*
* 也可以根据多个字段分片查询，就像是SQL语句中的where条件一样：
* String[] predicates = {
* "modifiedOn >= '2017-09-02' and modifiedOn < '2017-09-23'",
* "modifiedOn >= '2018' and modifiedOn < '2019'",
* "id < 998",
* "skuId = 'a17052200ux1459'"
* }
*
* 注意：如果条件重复的话，是会查出重复数据来的，如：
* String[] predicates = {"id=1", "id=2", "id=2"}
* 这样会找出三条记录，其中两条id=2的记录是一模一样的
*/
return spark.read().jdbc(url, tableName, predicates, connectionProperties);
}
/**
*
* 读取对应MySQL的配置信息
*
* @param environment
* @param dbName
* @return
*/
private static boolean readProperties(String environment, String dbName) {
String errorMsg = "";
boolean isContinue = true;
String confile = null;
InputStream is = null;
try {
// environment
List<String> environmentList = Arrays.asList(environmentArray);
if (!environmentList.contains(environment)) {
errorMsg = "environment must be one of: " + environmentList.toString();
isContinue = false;
}
// dbName
if (CommonFunction.isEmptyOrNull(dbName)) {
errorMsg = "dbName can not be null";
isContinue = false;
}
// 读取配置文件
if (isContinue) {
confile = dbName + "." + environment + "." + confilePrefix;
is = MysqlSparkConnector.class.getClassLoader().getResourceAsStream(confile);
if (is == null) {
errorMsg = "resource file: [" + confile + "] not find.";
isContinue = false;
}
if (isContinue) {
Properties properties = new Properties();
properties.load(is);
driver = properties.getProperty("driverClassName");
dbHost = properties.getProperty("dbHost");
dbPort = properties.getProperty("dbPort");
username = properties.getProperty("username");
password = properties.getProperty("password");
if (isContinue) {
if (CommonFunction.isEmptyOrNull(driver)) {
errorMsg = "the driver can not be null, please set in confile: " + confile;
isContinue = false;
}
}
if (isContinue) {
if (CommonFunction.isEmptyOrNull(dbHost)) {
errorMsg = "the dbHost can not be null, please set in confile: " + confile;
isContinue = false;
}
}
if (isContinue) {
if (CommonFunction.isEmptyOrNull(dbPort)) {
errorMsg = "the dbPort can not be null, please set in confile: " + confile;
isContinue = false;
}
}
if (isContinue) {
if (CommonFunction.isEmptyOrNull(username)) {
errorMsg = "the username can not be null, please set in confile: " + confile;
isContinue = false;
}
}
if (isContinue) {
if (CommonFunction.isEmptyOrNull(password)) {
errorMsg = "the password can not be null, please set in confile: " + confile;
isContinue = false;
}
}
}
}
if (!isContinue) {
logger.error(errorMsg);
}
} catch (Exception e) {
e.printStackTrace();
}
return isContinue;
}
}
扫一扫
430
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
