Use external volumes to process unstructured data

Use Spark on MaxCompute with external volumes

Spark on MaxCompute is a computing service provided by MaxCompute that is compatible with open source Spark. It provides a Spark computing framework on top of a unified system for computing resources, datasets, and permissions. This allows you to submit and run Spark jobs using familiar development methods to meet a wide range of data processing and analysis needs. If you require fine-grained permission control, use an external volume to manage access through the data warehouse's permission system. For more information, see Access OSS from Spark on MaxCompute.

-- Configurations
spark.hadoop.odps.cupid.volume.paths=odps://ms_proj1_dev/volume_yyy1/
spark.hadoop.odps.volume.common.filesystem=true
spark.hadoop.fs.odps.impl=org.apache.hadoop.fs.aliyun.volume.OdpsVolumeFileSystem
spark.hadoop.fs.AbstractFileSystem.odps.impl=org.apache.hadoop.fs.aliyun.volume.abstractfsimpl.OdpsVolumeFs

spark.hadoop.odps.access.id=xxxxxxxxx
spark.hadoop.odps.access.key=xxxxxxxxx
spark.hadoop.fs.oss.endpoint=oss-cn-beijing-internal.aliyuncs.com
spark.hadoop.odps.cupid.resources=ms_proj1_dev.jindofs-sdk-3.8.0.jar
spark.hadoop.fs.oss.impl=com.aliyun.emr.fs.oss.JindoOssFileSystem

spark.hadoop.odps.cupid.resources=public.python-2.7.13-ucs4.tar.gz
spark.pyspark.python=./public.python-2.7.13-ucs4.tar.gz/python-2.7.13-ucs4/bin/python
spark.hadoop.odps.spark.version=spark-2.4.5-odps0.34.0

-- Code
from numpy import array
from math import sqrt

from pyspark import SparkContext
from pyspark.mllib.clustering import KMeans, KMeansModel

if __name__ == "__main__":
    sc = SparkContext(appName="KMeansExample")  # SparkContext

    # Load and parse the data
    data = sc.textFile("odps://ms_proj1_dev/volume_yyy1/kmeans_data.txt")
    parsedData = data.map(lambda line: array([float(x) for x in line.split(' ')]))

    # Build the model (cluster the data)
    clusters = KMeans.train(parsedData, 2, maxIterations=10, initializationMode="random")

    # Evaluate clustering by computing Within Set Sum of Squared Errors
    def error(point):
        center = clusters.centers[clusters.predict(point)]
        return sqrt(sum([x**2 for x in (point - center)]))

    WSSSE = parsedData.map(lambda point: error(point)).reduce(lambda x, y: x + y)
    print("Within Set Sum of Squared Error = " + str(WSSSE))

    # Save and load model
    clusters.save(sc, "odps://ms_proj1_dev/volume_yyy1/target/PythonKMeansExample/KMeansModel")

    print(parsedData.map(lambda feature: clusters.predict(feature)).collect())

    sameModel = KMeansModel.load(sc, "odps://ms_proj1_dev/volume_yyy1/target/PythonKMeansExample/KMeansModel")
    
    print(parsedData.map(lambda feature: sameModel.predict(feature)).collect())
    sc.stop()

Use Proxima CE for vector computing

This section provides instructions and an example of how to use Proxima CE for vectorization in MaxCompute.

1. Install the Proxima CE resource package.

   For more information, see Installation and execution.

2. Run a job.

   • Limitations:

     • The Proxima SDK for Java currently only supports running task commands on a MaxCompute client that runs on Linux or macOS.

       Note

       Proxima CE jobs consist of two parts: local tasks and MaxCompute tasks. Local tasks are functional modules that do not involve MaxCompute SQL, MapReduce, or Graph jobs. MaxCompute tasks are jobs executed by MaxCompute engines such as SQL, MapReduce, and Graph. These two types of tasks are executed alternately. When a Proxima CE job starts, it first attempts to load the Proxima kernel on the local machine where the job is run by using the MaxCompute client. If the kernel loads successfully, certain modules run locally and call functions based on the Proxima kernel. If the kernel fails to load, an error is reported, but subsequent operations are not affected, and the modules call fallback functions instead. Because the job's JAR package contains Linux-related dependencies, it cannot be run on a MaxCompute client in a Windows operating system.

     • Running jobs from DataWorks MapReduce nodes is not currently supported. The underlying MaxCompute client version integrated into the MapReduce node is being upgraded, which causes jobs to fail. For now, submit jobs using the MaxCompute client.

   • Prepare data:

     -- Create input tables.
     CREATE TABLE doc_table_float_smoke(pk STRING, vector STRING) PARTITIONED BY (pt STRING);
     CREATE TABLE query_table_float_smoke(pk STRING, vector STRING) PARTITIONED BY (pt STRING);
     
     -- Insert data into the doc table (base table).
     ALTER TABLE doc_table_float_smoke add PARTITION(pt='20230116');
     INSERT OVERWRITE TABLE doc_table_float_smoke PARTITION (pt='20230116') VALUES
     ('1.nid','1~1~1~1~1~1~1~1'),
     ('2.nid','2~2~2~2~2~2~2~2'),
     ('3.nid','3~3~3~3~3~3~3~3'),
     ('4.nid','4~4~4~4~4~4~4~4'),
     ('5.nid','5~5~5~5~5~5~5~5'),
     ('6.nid','6~6~6~6~6~6~6~6'),
     ('7.nid','7~7~7~7~7~7~7~7'),
     ('8.nid','8~8~8~8~8~8~8~8'),
     ('9.nid','9~9~9~9~9~9~9~9'),
     ('10.nid','10~10~10~10~10~10~10~10');
     
     -- Insert data into the query table.
     ALTER TABLE query_table_float_smoke add PARTITION(pt='20230116');
     INSERT OVERWRITE TABLE query_table_float_smoke PARTITION (pt='20230116') VALUES
     ('q1.nid','1~1~1~1~2~2~2~2'),
     ('q2.nid','4~4~4~4~3~3~3~3'),
     ('q3.nid','9~9~9~9~5~5~5~5');

   • Sample job code:

     jar -libjars proxima-ce-aliyun-1.0.0.jar 
     -classpath proxima-ce-aliyun-1.0.0.jar com.alibaba.proxima2.ce.ProximaCERunner 
     -doc_table doc_table_float_smoke 
     -doc_table_partition 20230116 
     -query_table query_table_float_smoke 
     -query_table_partition 20230116 
     -output_table output_table_float_smoke 
     -output_table_partition 20230116 
     -data_type float 
     -dimension 8 
     -topk 1 
     -job_mode train:build:seek:recall 
     -external_volume shanghai_vol_ceshi
     -owner_id 1248953xxx
     ;

   • Sample result: Run the select * from output_table_float_smoke where pt='20230116'; command to query the result table.

     +------------+------------+------------+------------+
     | pk         | knn_result | score      | pt         |
     +------------+------------+------------+------------+
     | q1.nid     | 2.nid      | 4.0        | 20230116   |
     | q1.nid     | 1.nid      | 4.0        | 20230116   |
     | q1.nid     | 3.nid      | 20.0       | 20230116   |
     | q2.nid     | 4.nid      | 4.0        | 20230116   |
     | q2.nid     | 3.nid      | 4.0        | 20230116   |
     | q2.nid     | 2.nid      | 20.0       | 20230116   |
     | q3.nid     | 7.nid      | 32.0       | 20230116   |
     | q3.nid     | 8.nid      | 40.0       | 20230116   |
     | q3.nid     | 6.nid      | 40.0       | 20230116   |
     +------------+------------+------------+------------+