Apache Spark can work with multiple data sources that include various object stores like IBM Cloud Object Storage, OpenStack Swift and more. To access an object store, Apache Spark uses Hadoop modules that contain connectors to the various object stores.
Apache Spark needs only small set of object store functionalities. Specifically, Apache Spark requires object listing, objects creation, read objects, and getting data partitions. Hadoop connectors, however, must be compliant with the Hadoop ecosystem. This means they support many more operations, such as shell operations on directories, including move, copy, rename, etc. (these are not native object store operations). Moreover, Hadoop Map Reduce Client is designed to work with file systems and not object stores. The temp files and folders it uses for every write operation are renamed, copied, and deleted. This leads to dozens of useless requests targeted at the object store. It’s clear that Hadoop is designed to work with file systems and not object stores.
Stocator is implicitly designed for the object stores, it has very a different architecture from the existing Hadoop connector. It doesn’t depends on the Hadoop modules and interacts directly with object stores.
Stocator is a generic connector, that may contain various implementations for object stores. It shipped with OpenStack Swift and IBM Cloud Object Storage connectors. Stocator can be easily extended with more object store implementations.
- Hadoop ecosystem compliant. Implements Hadoop FileSystem interface
- No need to change or recompile Apache Spark
- Stocator doesn’t create any temporary folders or files for write operations. Each Spark's task generates only one object in the object store. Preserves existing Hadoop fault tolerance model
- Object's name may contain "/" and thus simulate directory structures
- Containers / buckets are automatically created
- Supports speculate mode
- Stocator uses Apache httpcomponents.httpclient.version version 4.5.2 and up
-
Checkout the master branch
https://github.com/SparkTC/stocator.git -
Change directory to
stocator -
Execute
mvn install -
If you want to build a jar with all the dependencies, please execute
mvn clean package -Pall-in-one
Stocator can be used easily with Apache Spark. There are different ways to use Stocator
Stocator deployed on spark-packages. This is the easiest form to integrate Stocator in Spark. Just follow stocator
It is possible to execute Apache Spark with Stocator, without compiling Apache Spark. Just make sure that Stocator jars on the class path. Build Stocator with
mvn clean package -Pall-in-one
Directory stocator/target contains standalone jar stocator-X.Y.Z-jar-with-dependencies.jar.
The best is to extend Spark's class path to include Stocator jar. Edit conf/spark-defaults.conf and add Stocator to the class path. For example
spark.driver.extraClassPath=/<PATH>/stocator/target/stocator-X-Y-Z-jar-with-dependencies.jar
spark.executor.extraClassPath=/<PATH>/stocator/target/stocator-X-Y-Z-jar-with-dependencies.jar
Another option is to run Apache Spark with
./bin/spark-shell --jars stocator-X.Y.Z-jar-with-dependencies.jar
However this is less adviced, as Spark will copy Stocator jar to the executors, which may consume time.
Less recommended, only for advanced users who want to recompile Spark. Both main pom.xml and core/pom.xml of Spark should be modified.
add to the <properties> of the main pom.xml
<stocator.version>X.Y.Z</stocator.version>
add stocator dependency to the main pom.xml
<dependency>
<groupId>com.ibm.stocator</groupId>
<artifactId>stocator</artifactId>
<version>${stocator.version}</version>
<scope>${hadoop.deps.scope}</scope>
</dependency>
modify core/pom.xml to include stocator
<dependency>
<groupId>com.ibm.stocator</groupId>
<artifactId>stocator</artifactId>
</dependency>
Compile Apache Spark with Haddop support as described here
Stocator verifies that
mapreduce.fileoutputcommitter.marksuccessfuljobs=true
If not modified, the default value of mapreduce.fileoutputcommitter.marksuccessfuljobs is true.
Stocator uses configuration keys that can be configured via spark's core-site.xml or provided in run time without using core-site.xml. To provide keys in run time use SparkContext variable with
sc.hadoopConfiguration.set("KEY","VALUE")
For usage with core-site.xml, see the configuration template located under conf/core-site.xml.template.
Stocator allows to access IBM Cloud Object Storage via cos:// schema. The general URI is the form
cos://<bucket>.<service>/object(s)
where bucket is object storage bucket and <service> identifies configuration group entry.
Each <service> may be any text, without special characters. Each service may use it's specific credentials and has different endpoint. By using multiple <service> allows to use different endpoints simultaneously.
For example, service=myObjectStore, then URI will be of the form
cos://<bucket>.myObjectStore/object(s)
and configuration keys will have prefix fs.cos.myObjectStore.
If none provided, the default value is service, e.g.
cos://<bucket>.service/object(s)
and configuration keys will have prefix fs.cos.service.
Configure Stocator in conf/core-site.xml
<property>
<name>fs.stocator.scheme.list</name>
<value>cos</value>
</property>
<property>
<name>fs.cos.impl</name>
<value>com.ibm.stocator.fs.ObjectStoreFileSystem</value>
</property>
<property>
<name>fs.stocator.cos.impl</name>
<value>com.ibm.stocator.fs.cos.COSAPIClient</value>
</property>
<property>
<name>fs.stocator.cos.scheme</name>
<value>cos</value>
</property>
Stocator COS connector expose "fs.cos." keys prefix. For backward compatibility Stocator also supports "fs.s3d" and "fs.s3a" prefix, where "fs.cos" has the highest priority and will overwrite other keys, if present.
To work with IAM and provide api key please switch to the relevant ibm-sdk branch depends on the Stocator version you need. For example for Stocator 1.0.24 release, switch to 1.0.24-ibm-sdk, for Stocator master 1.0.25-SNAPSHOT, switch to 1.0.25-SNAPSHOT-IBM-SDK and so on.
You will need to build Stocator manually, for example using 1.0.24-ibm-sdk branch:
git clone https://github.com/SparkTC/stocator
cd stocator
git fetch
git checkout -b 1.0.24-ibm-sdk origin/1.0.24-ibm-sdk
mvn clean install -Dmaven.test.skip=true
You now need to include the target/stocator-1.0.24-SNAPSHOT-IBM-SDK.jar into class path of Spark. Follow section Using Stocator without compiling Apache Spark
The next step if to configure Stocator with your COS credentials. The COS credentials is of the form
{
"apikey": "123",
"endpoints": "https://cos-service.bluemix.net/endpoints",
"iam_apikey_description": "Auto generated apikey during resource-key operation for Instance - abc",
"iam_apikey_name": "auto-generated-apikey-123",
"iam_role_crn": "role",
"iam_serviceid_crn": "identity-123::serviceid:ServiceId-XYZ",
"resource_instance_id": "abc"
}
The following is the list of the Stocator configuration keys. <service> can be any value, for example myCOS
| Key | Info | Mandatory | value |
|---|---|---|---|
fs.cos.<service>.iam.api.key |
API key | mandatory | value of apiKey |
fs.cos.<service>.iam.service.id |
Service ID | mandatory | Value of iam_serviceid_crn. In certain cases you need only value after :serviceid: |
fs.cos.<service>.endpoint |
COS endpoint | mandatory | Open link from endpoints and choose relevant endpoint. This endpoint should go here |
Example, configure <service> as myCOS:
<property>
<name>fs.cos.myCos.iam.api.key</name>
<value>123</value>
</property>
<property>
<name>fs.cos.myCos.endpoint</name>
<value>http://s3-api.us-geo.objectstorage.softlayer.net</value>
</property>
<property>
<name>fs.cos.myCos.iam.service.id</name>
<value>ServiceId-XYZ</value>
</property>
Now you can use URI
cos://mybucket.myCos/myobject(s)
An optional, it is possible to provide existing token instead of using API key.
Instead of providing fs.cos.myCos.iam.api.key, Stocator supports fs.cos.myCos.iam.api.token that may contain value of the existing token. When token is expired, Stocator will throw 403 exception. It's the user responsibility to provide long activation token or re-create token outside of Stocator.
The following is the list of the configuration keys. <service> can be any value, for example myCOS
| Key | Info | Mandatory |
|---|---|---|
fs.cos.<service>.access.key |
Access key | mandatory |
fs.cos.<service>.secret.key |
Secret key | mandatory |
fs.cos.<service>.session.token |
Session token | optional |
fs.cos.<service>.endpoint |
COS endpoint | mandatory |
fs.cos.<service>.v2.signer.type |
Signer type | optional |
Example, configure <service> as myCOS:
<property>
<name>fs.cos.myCos.access.key</name>
<value>ACCESS KEY</value>
</property>
<property>
<name>fs.cos.myCos.endpoint</name>
<value>http://s3-api.us-geo.objectstorage.softlayer.net</value>
</property>
<property>
<name>fs.cos.myCos.secret.key</name>
<value>SECRET KEY</value>
</property>
<property>
<name>fs.cos.myCos.session.token</name>
<value>SESSION TOKEN</value>
</property>
<property>
<name>fs.cos.myCos.v2.signer.type</name>
<value>false</value>
</property>
Now you can use URI
cos://mybucket.myCos/myobject(s)
| Key | Default | Info |
|---|---|---|
| fs.cos.socket.send.buffer | 8*1024 | socket send buffer to be used in the client |
| fs.cos.socket.recv.buffer | 8*1024 | socket send buffer to be used in the client |
| fs.cos.paging.maximum | 5000 | number of records to get while paging through a directory listing |
| fs.cos.threads.max | 10 | the maximum number of threads to allow in the pool used by TransferManager |
| fs.cos.threads.keepalivetime | 60 | the time an idle thread waits before terminating |
| fs.cos.signing-algorithm | override signature algorithm used for signing requests | |
| fs.cos.connection.maximum | 10000 | number of simultaneous connections to the object store |
| fs.cos.attempts.maximum | 20 | number of times we should retry errors |
| fs.cos.block.size | 128 | size of a block in MB |
| fs.cos.connection.timeout | 800000 | amount of time (in ms) until we give up on a connection to the object store |
| fs.cos.connection.establish.timeout | 50000 | amount of time (in ms) until we give up trying to establish a connection to the object store |
| fs.cos.client.execution.timeout | 500000 | amount of time (in ms) to allow a client to complete the execution of an API call |
| fs.cos.client.request.timeout | 500000 | amount of time to wait (in ms) for a request to complete before giving up and timing out |
| fs.cos.connection.ssl.enabled | true | Enables or disables SSL connections to COS |
| fs.cos.proxy.host | Hostname of the (optional) proxy server for COS connections | |
| fs.cos.proxy.port | Proxy server port. If this property is not set but fs.cos.proxy.host is, port 80 or 443 is assumed (consistent with the value of fs.cos.connection.ssl.enabled) | |
| fs.cos.proxy.username | Username for authenticating with proxy server | |
| fs.cos.proxy.password | Password for authenticating with proxy server | |
| fs.cos.proxy.domain | Domain for authenticating with proxy server | |
| fs.cos.user.agent.prefix | User agent prefix | |
| fs.cos.flat.list | false | In flat listing the result will include all objects under specific path prefix, for example bucket/a/b/data.txt, bucket/a/d.data. If listed bucket/a*, then result will include both objects. If flat list is set to flase, then it contains the same list behaviour as community s3a connector. |
| fs.stocator.cache.size | 2000 | The Guava cache size used by the COS connector |
| fs.cos.multipart.size | 8388608 | Size in bytes. Define multipart size |
| fs.cos.multipart.threshold | Max Integer | minimum size in bytes before we start a multipart uploads, default is max integer |
| fs.cos.fast.upload | false | enable or disable block upload |
| fs.stocator.glob.bracket.support | false | if true supports Hadoop string patterns of the form {ab,c{de, fh}}. Due to possible collision with object names, this mode prevents from create an object whose name contains {} |
| fs.cos.atomic.write | false | enable or disable atomic write to COS using conditional requests. When the flag is set to true and the operation is create with overwrite == false a conditional header will be used to handle race conditions for mutliple writers. If the path gets created between fs.create and stream.close by an external writer the close operation will fail and the object will not be written |
Stocator allows to access OpenStack Swift API based object stores via unique schema swift2d://.
-
Uses streaming for object uploads, without knowing object size. This is unique to Swift connector and removes the need to store object locally prior uploading it.
-
Supports Swiftauth, Keystone V2, Keystone V3 Password Scope Authentication
-
Supports any object store that expose Swift API and supports different authentication models
-
Supports access to public containers. For example
sc.textFile("swift2d://dal05.objectstorage.softlayer.net/v1/AUTH_ID/CONT/data.csv")
Add the dependence to Stocator in conf/core-site.xml
<property>
<name>fs.stocator.scheme.list</name>
<value>swift2d</value>
</property>
If Swift connector used concurrently with COS connector, then also configure
<property>
<name>fs.stocator.scheme.list</name>
<value>swift2d,cos</value>
</property>
Configure the rest of the keys
<property>
<name>fs.swift2d.impl</name>
<value>com.ibm.stocator.fs.ObjectStoreFileSystem</value>
</property>
<property>
<name>fs.stocator.swift2d.impl</name>
<value>com.ibm.stocator.fs.swift.SwiftAPIClient</value>
</property>
<property>
<name>fs.stocator.swift2d.scheme</name>
<value>swift2d</value>
</property>
The following is the list of the configuration keys
| Key | Info | Default value |
|---|---|---|
| fs.swift2d.service.SERVICE_NAME.auth.url | Mandatory | |
| fs.swift2d.service.SERVICE_NAME.public | Optional. Values: true, false | false |
| fs.swift2d.service.SERVICE_NAME.tenant | Mandatory | |
| fs.swift2d.service.SERVICE_NAME.password | Mandatory | |
| fs.swift2d.service.SERVICE_NAME.username | Mandatory | |
| fs.swift2d.service.SERVICE_NAME.block.size | Block size in MB | 128MB |
| fs.swift2d.service.SERVICE_NAME.region | Mandatory for Keystone | |
| fs.swift2d.service.SERVICE_NAME.auth.method | Optional. Values: keystone, swiftauth, keystoneV3 | keystoneV3 |
| fs.swift2d.service.SERVICE_NAME.nonstreaming.upload | Optional. If set to true then any object upload will be stored locally in the temp file and uploaded on close method. Disable stocator streaming mode | false |
<property>
<name>fs.swift2d.service.SERVICE_NAME.auth.url</name>
<value>http://IP:PORT/v2.0/tokens</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.public</name>
<value>true</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.tenant</name>
<value>TENANT</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.password</name>
<value>PASSWORD</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.username</name>
<value>USERNAME</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.auth.method</name>
<!-- swiftauth if needed -->
<value>keystone</value>
</property>
| Driver configuration key | Keystone V3 key |
|---|---|
| fs.swift2d.service.SERVICE_NAME.username | user id |
| fs.swift2d.service.SERVICE_NAME.tenant | project id |
<property>
<name>fs.swift2d.service.SERVICE_NAME.auth.url</name>
<value>https://identity.open.softlayer.com/v3/auth/tokens</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.public</name>
<value>true</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.tenant</name>
<value>PROJECTID</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.password</name>
<value>PASSWORD</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.username</name>
<value>USERID</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.auth.method</name>
<value>keystoneV3</value>
</property>
<property>
<name>fs.swift2d.service.SERVICE_NAME.region</name>
<value>dallas</value>
</property>
Below is the optional configuration that can be provided to Stocator and used internally to configure HttpClient.
| Configuration key | Default | Info |
|---|---|---|
| fs.stocator.MaxPerRoute | 25 | maximal connections per IP route |
| fs.stocator.MaxTotal | 50 | maximal concurrent connections |
| fs.stocator.SoTimeout | 1000 | low level socket timeout in milliseconds |
| fs.stocator.executionCount | 100 | number of retries for certain HTTP issues |
| fs.stocator.ReqConnectTimeout | 5000 | Request level connect timeout. Determines the timeout in milliseconds until a connection is established |
| fs.stocator.ReqConnectionRequestTimeout | 5000 | Request level connection timeout. Returns the timeout in milliseconds used when requesting a connection from the connection manager |
| fs.stocator.ReqSocketTimeout | 5000 | Defines the socket timeout (SO_TIMEOUT) in milliseconds, which is the timeout for waiting for data or, put differently, a maximum period inactivity between two consecutive data packets). |
| fs.stocator.joss.synchronize.time | false | Will disable JOSS to synchronize time with the server. Setting this value to 'false' will badly impact on temp url. However this will reduce HEAD on account which might be problematic if the user doesn't has access rights to HEAD an account |
| fs.stocator.tls.version | false | if not provided, then system choosen is the default. In certain cases user may setup custom value, like TLSv1.2 |
By default Stocator exposes swift2d:// for Swift API and cos:// for IBM Cloud Object Storage. It possible to configure Stocator to expose different schema.
The following example, will configure Stocator to respond to swift:// in addition to swift2d://. This is useful, so users doesn't need to modify existing jobs that already uses hadoop-openstack connector. Below the example, how to configure Stocator to respond both to swift:// and swift2d://
<property>
<name>fs.stocator.scheme.list</name>
<value>swift2d,swift</value>
</property>
<!-- configure stocator as swift2d:// -->
<property>
<name>fs.swift2d.impl</name>
<value>com.ibm.stocator.fs.ObjectStoreFileSystem</value>
</property>
<property>
<name>fs.stocator.swift2d.impl</name>
<value>com.ibm.stocator.fs.swift.SwiftAPIClient</value>
</property>
<property>
<name>fs.stocator.swift2d.scheme</name>
<value>swift2d</value>
</property>
<!-- configure stocator as swift:// -->
<property>
<name>fs.swift.impl</name>
<value>com.ibm.stocator.fs.ObjectStoreFileSystem</value>
</property>
<property>
<name>fs.stocator.swift.impl</name>
<value>com.ibm.stocator.fs.swift.SwiftAPIClient</value>
</property>
<property>
<name>fs.stocator.swift.scheme</name>
<value>swift</value>
</property>
val data = Array(1, 2, 3, 4, 5, 6, 7, 8)
val distData = sc.parallelize(data)
distData.saveAsTextFile("cos://mybucket.service/one1.txt")
Listing bucket mydata directly with a REST client will display
one1.txt
one1.txt/_SUCCESS
one1.txt/part-00000-taskid
one1.txt/part-00001-taskid
one1.txt/part-00002-taskid
one1.txt/part-00003-taskid
one1.txt/part-00004-taskid
one1.txt/part-00005-taskid
one1.txt/part-00006-taskid
one1.txt/part-00007-taskid
val squaresDF = spark.sparkContext.makeRDD(1 to 5).map(i => (i, i * i)).toDF("value","square")
squaresDF.write.format("parquet").save("cos://mybucket.service/data.parquet")
Follow
https://github.com/ehiggs/spark-terasort
Setup Stocator with COS as previosuly explained. In the example we use bucket teradata and service =service
Step 1:
export MAVEN_OPTS="-Xmx2g -XX:MaxPermSize=512M -XX:ReservedCodeCacheSize=512m"
Step 2:
./bin/spark-submit --driver-memory 2g --class com.github.ehiggs.spark.terasort.TeraGen /spark-terasort/target/spark-terasort-1.1-SNAPSHOT-jar-with-dependencies.jar 1g cos://teradata.service/terasort_in
Step 3:
./bin/spark-submit --driver-memory 2g --class com.github.ehiggs.spark.terasort.TeraSort /target/spark-terasort-1.1-SNAPSHOT-jar-with-dependencies.jar 1g cos://teradata.service/terasort_in cos://teradata.service/terasort_out
Step 4:
./bin/spark-submit --driver-memory 2g --class com.github.ehiggs.spark.terasort.TeraValidate /target/spark-terasort-1.1-SNAPSHOT-jar-with-dependencies.jar cos://teradata.service/terasort_out cos://teradata.service/terasort_validate
Copy
src/test/resources/core-site.xml.tempate to src/test/resources/core-site.xml
Edit src/test/resources/core-site.xml and configure Swift access details.
Functional tests will use container from fs.swift2d.test.uri. To use different container, change drivertest to different name. Container need not to be exists in advance and will be created automatically.
If you like to work on code, you can easily setup Eclipse project via
mvn eclipse:eclipse
and import it into Eclipse workspace.
To ease the debugging process, Please modify conf/log4j.properties to
log4j.logger.com.ibm.stocator=ALL
We ask that you include a line similar to the following as part of your pull request comments:
“DCO 1.1 Signed-off-by: Random J Developer“.
“DCO” stands for “Developer Certificate of Origin,”
and refers to the same text used in the Linux Kernel community. By adding this simple comment, you tell the community that you wrote the code you are contributing, or you have the right to pass on the code that you are contributing.
Join Stocator mailing list by sending email to stocator+subscribe@googlegroups.com.
Use stocator@googlegroups.com to post questions.
Please follow our wiki for more details. More information about Stocator can be find at
- Stocator: A High Performance Object Store Connector for Spark - SoCC '17 Proceedings of the 2017 Symposium on Cloud Computing
- MapReduce and object stores – How can we do it better?
- Advantages and complexities of integrating Hadoop with object stores
- Stocator on IBM Code
- Analyze data faster using Spark and IBM Cloud Object Storage
- Exabytes, Elephants, Objects and Apache Spark
- Simulating E.T. or: how to insert individual files into object storage from within a map function in Apache Spark
- Hadoop and object stores: Can we do it better? Strata Data Conference, 23-25 May 2017, London, UK
- VERY LARGE DATA FILES, OBJECT STORES, AND DEEP LEARNING—LESSONS LEARNED WHILE LOOKING FOR SIGNS OF EXTRA-TERRESTRIAL LIFE SPARK SUMMIT 2017 DATA SCIENCE AND ENGINEERING AT SCALE, JUNE 5 – 7, 2017 SAN FRANCISCO
- Using data connectors to work with IBM Cloud Object Storage in IBM Spectrum Conductor with Spark 2.2.1
This research was supported by IOStack, an H2020 project of the EU Commission