phoenix与spark整合

目的是将phoenix做存储，spark做计算层。这样就结合了phoenix查询速度快和spark计算速度快的优点。
在这里将Phoenix的表作为spark的RDD或者DataFrames来操作，并且将操作的结果写回phoenix中。
这样做也扩大了两者的使用场景。

Phoenix 版本 4.4.0
Hbase版本 0.98
spark版本 spark-1.5.2-bin-hadoop2.6
首先配置 SPARK_CLASSPATH
要想在spark中操作phoenix，就必须让spark可以找到phoenix的相关类，所以我们把client放到spark_classpath中

export SPARK_CLASSPATH=$SPARK_CLASSPATH:/home/hadoop/phoenix/phoenix-spark-4.4.0-HBase-0.98-tests.jar
export SPARK_CLASSPATH=$SPARK_CLASSPATH:/home/hadoop/phoenix/phoenix-4.4.0-HBase-0.98-client.jar
export SPARK_CLASSPATH=$SPARK_CLASSPATH:/home/hadoop/phoenix/phoenix-server-client-4.4.0-HBase-0.98.jar

这样就可以在spark-shell中操作phoenix了

下来结合两者做下实验：
1> 在phoenix中创建几张表

[hadoop@10.10.113.45 ~/phoenix/bin]$>./sqlline.py 10.10.113.45:2181
0: jdbc:phoenix:10.10.113.45:2181> CREATE TABLE EMAIL_ENRON(
. . . . . . . . . . . . . . . . .> MAIL_FROM BIGINT NOT NULL,
. . . . . . . . . . . . . . . . .> MAIL_TO BIGINT NOT NULL
. . . . . . . . . . . . . . . . .> CONSTRAINT pk PRIMARY KEY(MAIL_FROM, MAIL_TO));
0: jdbc:phoenix:10.10.113.45:2181> CREATE TABLE EMAIL_ENRON_PAGERANK(
. . . . . . . . . . . . . . . . .> ID BIGINT NOT NULL,
. . . . . . . . . . . . . . . . .> RANK DOUBLE
. . . . . . . . . . . . . . . . .> CONSTRAINT pk PRIMARY KEY(ID));
No rows affected (0.52 seconds)

查看下是否创建成功

0: jdbc:phoenix:10.10.113.45:2181> !tables
+------------------------------------------+------------------------------------------+------------------------------------------+--------------+
| TABLE_CAT | TABLE_SCHEM | TABLE_NAME | |
+------------------------------------------+------------------------------------------+------------------------------------------+--------------+
| | SYSTEM | CATALOG | SYSTEM TABLE |
| | SYSTEM | FUNCTION | SYSTEM TABLE |
| | SYSTEM | SEQUENCE | SYSTEM TABLE |
| | SYSTEM | STATS | SYSTEM TABLE |
| | | EMAIL_ENRON | TABLE |
| | | EMAIL_ENRON_PAGERANK | TABLE |
+------------------------------------------+------------------------------------------+------------------------------------------+--------------+
0: jdbc:phoenix:10.10.113.45:2181>

2> 在将数据load到phoenix中，数据有40万行

[hadoop@10.10.113.45 ~/phoenix/bin]$>./psql.py -t EMAIL_ENRON 10.10.113.45:2181 /home/hadoop/sfs/enron.csv
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
15/12/03 10:06:37 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
csv columns from database.
CSV Upsert complete. 367662 rows upserted
Time: 21.783 sec(s)

数据来源：https://snap.stanford.edu/data/email-Enron.html
然后在查询下

0: jdbc:phoenix:10.10.113.45:2181> select count(*) from EMAIL_ENRON;
+------------------------------------------+
| COUNT(1) |
+------------------------------------------+
| 367662 |
+------------------------------------------+
1 row selected (0.289 seconds)

看37万数据，查询不到一秒！！！
下面进入到spark-shell 的交互模式，我们做一个PageRank 算法的例子

[hadoop@10.10.113.45 ~/spark/bin]$>./spark-shell
scala> import org.apache.spark.graphx._
import org.apache.spark.graphx._
scala> import org.apache.phoenix.spark._
import org.apache.phoenix.spark._
scala> val rdd = sc.phoenixTableAsRDD("EMAIL_ENRON", Seq("MAIL_FROM", "MAIL_TO"), zkUrl=Some("10.10.113.45"))
rdd: org.apache.spark.rdd.RDD[Map[String,AnyRef]] = MapPartitionsRDD[2] at map at SparkContextFunctions.scala:39
scala> val rawEdges = rdd.map{ e => (e("MAIL_FROM").asInstanceOf[VertexId], e("MAIL_TO").asInstanceOf[VertexId]) }
rawEdges: org.apache.spark.rdd.RDD[(org.apache.spark.graphx.VertexId, org.apache.spark.graphx.VertexId)] = MapPartitionsRDD[3] at map at <console>:29
scala> val graph = Graph.fromEdgeTuples(rawEdges, 1.0)
graph: org.apache.spark.graphx.Graph[Double,Int] = org.apache.spark.graphx.impl.GraphImpl@621bb3c3
scala> val pr = graph.pageRank(0.001)
pr: org.apache.spark.graphx.Graph[Double,Double] = org.apache.spark.graphx.impl.GraphImpl@55e444b1
scala> pr.vertices.saveToPhoenix("EMAIL_ENRON_PAGERANK", Seq("ID", "RANK"), zkUrl = Some("10.10.113.45"))（这一步会很耗内存，可能有的同学在测试的时候会报OOM，建议增大spark中executor memory，driver memory的大小）

我们在去phoenix中查看一下结果。

0: jdbc:phoenix:10.10.113.45:2181> select count(*) from EMAIL_ENRON_PAGERANK;
+------------------------------------------+
| COUNT(1) |
+------------------------------------------+
| 29000 |
+------------------------------------------+
1 row selected (0.113 seconds)
0: jdbc:phoenix:10.10.113.45:2181> SELECT * FROM EMAIL_ENRON_PAGERANK ORDER BY RANK DESC LIMIT 5;
+------------------------------------------+------------------------------------------+
| ID | RANK |
+------------------------------------------+------------------------------------------+
| 273 | 117.18141799210386 |
| 140 | 108.63091596789913 |
| 458 | 107.2728800448782 |
| 588 | 106.11840798585399 |
| 566 | 105.13932886531066 |
+------------------------------------------+------------------------------------------+
5 rows selected (0.568 seconds)