Connecting Hadoop to OBS

Overview

Hadoop provides a distributed resource scheduling engine for processing and analyzing large-scale data sets. OBS effects the Hadoop HDFS protocol. It can replace HDFS in the Hadoop system to connect to big data components such as Spark, MapReduce, and Hive, serving as the data lake storage for big data computing.

The HDFS protocol is defined through the FileSystem abstract class in Hadoop, which can be effected by different storage systems, such as the HDFS service built in Hadoop and Huawei Cloud OBS.

Constraints

The following HDFS semantics are not supported:

Lease
Symbolic link operations
Proxy users
File concat
File checksum
File replication factor
Extended attributes (Xattrs) operations
Snapshot operations
Storage policy
Quota
POSIX ACL
Delegation token operations

Precautions

To reduce output logs, add the following configuration to the /opt/hadoop-3.1.1/etc/hadoop/log4j.properties file:

log4j.logger.com.obs=ERROR

Procedure

Hadoop 3.1.1 is used here as an example. You are advised to use the latest version. You are not advised to use a Hadoop version earlier than 2.8.3.

Download hadoop-3.1.1.tar.gz and decompress it to the /opt/hadoop-3.1.1 directory.

Add the following content to the /etc/profile file:

export HADOOP_HOME=/opt/hadoop-3.1.1
export PATH=$HADOOP_HOME/bin:$HADOOP_HOME/sbin:$PATH

Install hadoop-huaweicloud.
1. Download it from GitHub.
  
  If no JAR package of the required version is available, modify the Hadoop version in the POM file under the hadoop-huaweicloud directory and then compile the file again.
2. Copy the hadoop-huaweicloud-x.x.x-hw-y.jar package to the /opt/hadoop-3.1.1/share/hadoop/tools/lib and /opt/hadoop-3.1.1/share/hadoop/common/lib directories.
  
  In a hadoop-huaweicloud-x.x.x-hw-y.jar package name, x.x.x indicates the Hadoop version number, and y indicates the OBSA version number. For example, in the hadoop-huaweicloud-3.1.1-hw-40.jar package name, 3.1.1 is the Hadoop version number, and 40 is the OBSA version number.

Configure Hadoop.

Add OBS configurations to the /opt/hadoop-3.1.1/etc/hadoop/core-site.xml file:

      
       
         
         
           <property>
<name>fs.obs.impl</name>
<value>org.apache.hadoop.fs.obs.OBSFileSystem</value>
</property>
<property>
<name>fs.AbstractFileSystem.obs.impl</name>
<value>org.apache.hadoop.fs.obs.OBS</value>
</property>
<property>
<name>fs.obs.access.key</name>
<value>xxx</value>
<description>HuaweiCloud Access Key Id</description>
</property>
<property>
<name>fs.obs.secret.key</name>
<value>xxx</value>
<description>HuaweiCloud Secret Access Key</description>
</property>
<property>
<name>fs.obs.endpoint</name>
<value>xxx</value>
<description>HuaweiCloud Endpoint</description>
</property>

          

        

      
     

Check whether the connection is successful.
You can use a CLI or MapReduce program for verification. Examples are provided as follows:
- CLI
  hadoop fs -ls obs://obs-bucket/
  
  Command output:
```
-rw-rw-rw- 1 root root 1087 2018-06-11 07:49 obs://obs-bucket/test1
-rw-rw-rw- 1 root root 1087 2018-06-11 07:49 obs://obs-bucket/test2
```
- MapReduce program
  hadoop jar share/hadoop/mapreduce/hadoop-mapreduce-examples-3.1.1.jar wordcount obs://example-bucket/input/test.txt obs://obs-bucket/output

[Appendix] hadoop-huaweicloud Configurations

Configuration Item	Default Value	Mandatory	Description
fs.obs.impl	org.apache.hadoop.fs.obs.OBSFileSystem	Yes	-
fs.AbstractFileSystem.obs.impl	org.apache.hadoop.fs.obs.OBS	Yes	-
fs.obs.endpoint	N/A	Yes	Endpoint of Huawei Cloud OBS
fs.obs.access.key	N/A	Yes	Huawei Cloud's access key ID (AK) for accessing the corresponding OBS bucket.
fs.obs.secret.key	N/A	Yes	Huawei Cloud's secret access key (SK) for accessing the corresponding OBS bucket.
fs.obs.session.token	N/A	No	Huawei Cloud's security token for accessing the corresponding OBS bucket. This token is required when a temporary AK/SK pair is used.
fs.obs.security.provider	N/A	No	Class for calling the com.obs.services.IObsCredentialsProvider API. This API is used to obtain the credentials for accessing OBS.
fs.obs.connection.ssl.enabled	FALSE	No	Specifies whether to access OBS through HTTPS.
fs.obs.threads.keepalivetime	60	No	Parameter keepAliveTime is used to control the read and write thread pool.
fs.obs.threads.max	20	No	Parameters corePoolSize and maximumPoolSize are used to control the read and write thread pool.
fs.obs.max.total.tasks	20	No	Parameter BlockingQueue is used to control the capacity of the read and write thread pool. Its value is the sum of the values of fs.obs.threads.max and fs.obs.max.total.tasks.
fs.obs.multipart.size	104857600	No	Size of a multipart upload.
fs.obs.fast.upload.buffer	disk	No	Specifies a cache method. All written data is cached and then uploaded to OBS. The options are as follows: disk: Data is cached on a disk. array: Data is cached in the JVM on-heap memory. bytebuffer: Data is cached in the JVM off-heap memory.
fs.obs.buffer.dir	${hadoop.tmp.dir}	No	Specifies the cache directory when fs.obs.fast.upload.buffer is set to disk. In such cases, multiple directories are supported and separated by commas (,).
fs.obs.bufferdir.verify.enable	FALSE	No	Specifies whether to verify the existence of the cache directory and whether the directory has the write permissions, when fs.obs.fast.upload.buffer is set to disk.
fs.obs.fast.upload.active.blocks	4	No	Specifies the maximum number of caches allowed by each stream operation (the maximum number of thread tasks that can be submitted through a multipart upload thread pool). This limits the maximum cache space (calculated from fs.obs.fast.upload.active.blocks x fs.obs.multipart.size) that can be used by each stream operation.
fs.obs.fast.upload.array.first.buffer	1048576	No	When fs.obs.fast.upload.buffer is set to array, this parameter is used to control the initial size of JVM on-heap memory.
fs.obs.readahead.range	1048576	No	Size of the part that will be read ahead.
fs.obs.multiobjectdelete.enable	TRUE	No	Specifies whether to enable a batch deletion when directories are deleted.
fs.obs.delete.threads.max	20	No	Parameters maximumPoolSize and corePoolSize are used to control the thread pool.
fs.obs.multiobjectdelete.maximum	1000	No	Specifies the maximum number of objects that can be deleted in a batch deletion request. The maximum value is 1000.
fs.obs.multiobjectdelete.threshold	3	No	Specifies the minimum number of objects in a batch deletion. If the number of objects to be batch deleted is less than this parameter value, batch deletion will not be started.
fs.obs.list.threads.core	30	No	Parameter corePoolSize is used to control the thread pool.
fs.obs.list.threads.max	60	No	Parameter maximumPoolSize is used to control the thread pool.
fs.obs.list.workqueue.capacity	1024	No	Capacity of the parameter BlockingQueue that is used to control the thread pool.
fs.obs.list.parallel.factor	30	No	This parameter is used to control concurrency factors.
fs.obs.paging.maximum	1000	No	Specifies the maximum number of objects that can be returned in a list request. The maximum value is 1000.
fs.obs.copy.threads.max	40	No	When a bucket renames a directory, parameters maximumPoolSize and corePoolSize are used to control the thread copy. Their value is half of the value of this parameter. The capacity of BlockingQueue is 1024.
fs.obs.copypart.size	104857600	No	Specifies the size of a single part in a multipart copy. If the size of an object to be copied exceeds this parameter value, multipart copy is performed, and the size of a single part is set to this parameter value. Otherwise, simple copy is performed.
fs.obs.copypart.threads.max	5368709120	No	If multipart copy is performed during the copy of a single object, maximumPoolSize and corePoolSize are used to configure the multipart copy thread pool. Their value is half of the value of this parameter. The capacity of BlockingQueue is 1024.
fs.obs.getcanonicalservicename.enable	FALSE	No	Controls the return value of API getCanonicalServiceName(). TRUE: obs://bucketname FALSE: null
fs.obs.multipart.purge	FALSE	No	Specifies whether to clear multipart upload tasks in a bucket when OBSFilesystem is initialized.
fs.obs.multipart.purge.age	86400	No	Time before which multipart upload tasks in a bucket will be cleared when OBSFilesystem is initialized.
fs.obs.trash.enable	FALSE	No	Specifies whether to enable the trash feature.
fs.obs.trash.dir	N/A	No	Directory for storing deleted files.
fs.obs.block.size	134217728	No	Block size.