이 콘텐츠는 선택한 언어로 제공되지 않습니다.
OpenStack Data Processing
Manually provisioning and scaling Hadoop clusters in Red Hat OpenStack Platform
Abstract
1. Overview
OpenStack Data Processing is provided by the OpenStack Sahara project, which provides a robust interface to easily provision and scale Hadoop clusters. Such clusters can then be used to run resource-intensive jobs, typically for processing large data sets. As an OpenStack component, OpenStack Data Processing is fully integrated into the OpenStack ecosystem; for example, users can administer the entire Hadoop data processing workflow through the OpenStack dashboard — from configuring clusters, all the way to launching and running jobs on them.
For more information about Hadoop, see http://hadoop.apache.org/.
OpenStack Data Processing uses different plug-ins for provisioning specific clusters of each Hadoop distribution. The Hadoop parameters available for configuring clusters differ depending on the Hadoop distribution (and, by extension, the plug-in used). As of this release (Red Hat OpenStack Platform 10), OpenStack Data Processing supports the following plug-ins:
- Cloudera (CDH) 5.5.0.
Hortonworks Data Platform
- MapR 5.1
OpenStack Data Processing also includes a Guides tab. This tab features wizards that will help you create the templates necessary in order to launch clusters and run jobs on them. However, you will still need to register the components necessary for using OpenStack Data Processing, such as Hadoop images and job binaries. As such, if you intend to use the Guides feature, we recommend you read Section 5, “Register the Required Components” first.
2. Installation
This section assumes that you are deploying OpenStack Data Processing on the overcloud, through the director. It also assumes that you created a director installation user named stack, identical to Creating a Director Installation User (from Director Installation and Usage).
The director uses environment files to configure the overcloud during deployment. These environment files are stored in /usr/share/openstack-tripleo-heat-templates/, along with the Puppet scripts and heat templates used to orchestrate each service.
The OpenStack Data Processing service is disabled on the overcloud by default. This is because the OpenStack Data Processing components are registered as null operations (OS::Heat::None) in the main overcloud environment file (/usr/share/openstack-tripleo-heat-templates/overcloud-resource-registry-puppet.yaml):
OS::TripleO::Services::SaharaApi: OS::Heat::None OS::TripleO::Services::SaharaEngine: OS::Heat::None
To enable these services, include an environment file that links these resources to their respective Heat templates in the /usr/share/openstack-tripleo-heat-templates/puppet/services directory. Some services have predefined environment files in the environments directory. For example, the Data Processing service (sahara) use the /usr/share/openstack-tripleo-heat-templates/environments/services/sahara.yaml file, which contains the following:
resource_registry: OS::TripleO::Services::SaharaApi: ../../puppet/services/sahara-api.yaml OS::TripleO::Services::SaharaEngine: ../../puppet/services/sahara-engine.yaml
This overrides the default null operation resources and enables the services. Include this environment file when running the openstack overcloud deploy command.
$ openstack overcloud deploy --templates -e \ /usr/share/openstack-tripleo-heat-templates/environments/services/sahara.yaml
You can also enable instance locality to force clusters to use volumes local to the same node. Doing so could improve the performance of your cluster. See Appendix A, Enabling Instance Locality for related infrormation.
If you passed any extra environment files when you created the overcloud, pass them again here using the -e option to avoid making undesired changes to the overcloud. For more information, see Modifying the Overcloud Environment (from Director Installation and Usage).
3. Workflow
OpenStack Data Processing provisions and scales Hadoop clusters using pre-configured cluster templates that define specifically designed instances. These instances form the individual nodes that make up Hadoop clusters; you can then use these Hadoop clusters to run the jobs/binaries that will process your data.
If you intend to use OpenStack Data Processing, you should already be familiar with the necessary components for working within the Hadoop framework. As such, the general workflow described in this section assumes that you already have the following components prepared:
- A Hadoop image; specifically, a Red Hat Enterprise Linux image containing a Hadoop data processing plug-in. See Section 1, “Overview” for a list of supported plug-ins.
- The input data you wish to process, preferably uploaded to the Object Storage service.
- The job binaries and libraries you will use to process the input data, preferably uploaded to the Object Storage service.
For details on how to upload content to the Object Storage service, see Upload an Object.
In addition, you should also have a general idea of the computational resources required to run the job. This will help you determine what type of nodes (and how many of each) you will need.
The following high-level workflow describes how to configure and use the OpenStack Data Processing service to launch clusters and run jobs on those clusters:
Create an image containing the necessary plug-in components for OpenStack Data Processing (Section 4, “Create Hadoop Image”). This will be your Hadoop image.
The procedure for creating this image differs depending on your chosen Hadoop plug-in.
Register the following required components to the OpenStack Data Processing service:
- Hadoop image
- Data sources (namely, your input data and where the output data should go)
Create node group templates. Each template defines many useful Hadoop-specific settings for any given node, most notably:
- What Hadoop plug-in and version should the node group use?
- Which processes should run on the node?
Create or upload cluster templates. A cluster template defines, among other things:
- Node group composition: namely, how many nodes of each node group should make up the cluster.
- Cluster-scoped Hadoop configurations: specific parameters you need to set for each Hadoop component (HIVE, AMBARI, HDFS, and the like).
- Launch a Hadoop cluster (using a cluster template), and run a job on the cluster (namely, running a registered job binary on a data source). You can also scale the cluster (as in, add or remove nodes of any type) as needed.
- Register job binaries, scripts, or libraries to the OpenStack Data Processing Service, create jobs, and launch them on Hadoop clusters. Jobs define which job binaries, scripts, or libraries should be used to process registered data sources.
The next few sections describe each workflow step in greater detail.
4. Create Hadoop Image
This release supports the following plug-ins:
- Cloudera (CDH) 5.5.0.
Hortonworks Data Platform
- MapR 5.1
To create a Hadoop image based on any of these plug-ins, you need to first download a Red Hat Enterprise Linux 7 image. You can get one from the following link (requires a Red Hat subscription):
https://access.redhat.com/downloads/content/69/ver=/rhel---7/7.2/x86_64/product-software
After downloading the image, configure your system to use your Red Hat subscription details. Doing so will allow your system to further download the tools necessary for creating a Hadoop image from your downloaded Red Hat Enterprise Linux image. To do so, enter the following commands from a terminal (as root):
# export REG_USER=USERNAME # export REG_PASSWORD=PASSWORD # export REG_POOL_ID=POOLID
Where:
- USERNAME and PASSWORD are your Red Hat subscription credentials.
- POOLID is your Red Hat subscription’s Pool ID. To determine your subscription’s Pool ID, enter following command:
# subscription-manager list --available
This command will display important details about your subscription, including the Pool ID value.
For more information about manually configuring your Red Hat subscription, see Using and Configuring Red Hat Subscription Manager.
Next, install the sahara-image-elements package:
# yum install sahara-image-elements -y
Finally, create the Hadoop image. To do this, go to the same directory as your Red Hat Enterprise Linux image and run the following command:
# sahara-image-create -p PLUGINReplace PLUGIN with the plug-in that the Hadoop image should use. Supported values are:
- cloudera (Cloudera (CDH) 5.5.0)
- mapr (MapR 5.1)
ambari (Hortonworks Data Platform)
- OpenStack Data Processing supports two versions of the Hortonworks Data Platform plug-in. Include -v 2.3 (for version 2.3) or -v 2.4 (for version 2.4) when you run the command.
The sahara-image-create command will create an image containing the name of the plug-in (for example, sahara-rhel-7-2-cdh-5-5.qcow2 for cloudera).
5. Register the Required Components
OpenStack Data Processing requires a Hadoop image containing the necessary elements to launch and use Hadoop clusters. Specifically, Red Hat OpenStack Platform requires an image containing Red Hat Enterprise Linux with the necessary data processing plug-in.
Once you have a Hadoop image suitable for the jobs you wish to run, register it to the OpenStack Data Processing service. To do so:
- Upload the image to the Image service. For instructions on how to do so, see Upload an Image.
- After uploading the image, select Project > Data Processing > Image Registry in the dashboard.
- Click Register Image, and select the Hadoop image from the Image drop-down menu.
- Enter the user name that the OpenStack Data Processing service should use to apply settings and manage processes on each instance/node. The user name set for this purpose on the official images provided by Red Hat Enterprise Linux (which you used in Section 4, “Create Hadoop Image”) is cloud-user.
By default, the OpenStack Data Processing service will add the necessary plug-in and version tags in the plug-in and Version drop-down menu. Verify that the tag selection is correct, then click Add plugin tags to add them. The OpenStack Data Processing service also allows you to use custom tags to either differentiate or group registered images. Use the Add custom tag button to add a tag; tags appear in the box under the Description field.
To remove a custom tag, click the x beside its name.
- Click Done. The image should now appear in the Image Registry table.
5.1. Register Input and Output Data Sources
After registering an image, register your data input source and output destination. You can register both as objects from the Object Storage service; as such, you need to upload both as objects first. For instructions on how to do so, see Upload an Object.
You can also register data objects straight from another Hadoop-compatible distributed file system (for example, HDFS). For information on how to upload data to your chosen distributed file system, see its documentation.
- In the dashboard, select Project > Data Processing > Data Sources.
- Click Create Data Source. Enter a name for your data source in the Name field.
- Use the Description field to describe the data source (optional).
Select your data source’s type and URL. The procedure for doing so depends on your source’s location:
If your data is located in the Object Storage service, select Swift from the Data Source Type drop-down menu. Then:
- Provide the container and object name of your data source as swift://CONTAINER/OBJECT in the URL field.
- If your data source requires a login, supply the necessary credentials in the Source username and Source password fields.
If your data is located in a Hadoop Distributed File System (HDFS), select the corresponding source from the Data Source Type drop-down menu. Then, enter the data source’s URL in the URL field as hdfs://HDFSHOST:PORT/OBJECTPATH, where:
- HDFSHOST is the host name of the HDFS host.
- PORT is the port on which the data source is accessible.
- OBJECTPATH is the available path to the data source on HDFSHOST.
- Click Done. The data source should now be available in the Data Sources table.
Perform this procedure for each data input/output object required for your jobs.
6. Configure Node Group Templates
Data processing through Hadoop involves running jobs on clusters. Each cluster is made up of specially-configured nodes (or instances). OpenStack Data Processing allows you to configure templates defining the different types of nodes required for each cluster. These templates are called node group templates.
A node group templates defines the settings required for each node in a Hadoop cluster. The following procedure describes how to configure a node group template:
- In the dashboard, select Project > Data Processing > Node Group Templates.
- Click Create Template.
- Use the Plugin Name and Version drop-down menus to select the name and version of the Hadoop plug-in that the node will use.
- Click Create.
- Enter a name for your template in the Template Name field.
- Use the Description field to describe the node group template you are creating (optional).
- Select a flavor that the nodes should use from the OpenStack Flavor drop-down menu. The flavor should be appropriate to the computing, memory, and storage needs of the node. For more information about flavors, see Manage Flavors.
Select a storage location for your node from the Storage location drop-down menu:
- Ephemeral Drive: the OpenStack Data Processing service will generate the necessary ephemeral storage for the node.
Cinder Volume: with this option, you can configure the Block Storage service to create a set number of volumes for each node. Use the Volumes per node field to set the number of volumes, and the Volumes size (GB) field to specify the size of each volume.
For most deployments, we recommend the Cinder Volume option. Data on ephemeral drives are dependent on the integrity of the host, and therefore vulnerable to host failures.
- Use the Floating IP pool drop-down menu to set whether the node should use floating IPs; and if so, which one. For more information about floating IPs, see Configure IP Addressing.
Next, configure the node’s security. To create a security group for the node group, select the Auto Security Group check box. You can also launch the node in existing security groups by selecting their corresponding check boxes from the Security Groups list.
See Project Security Management for more details about security groups.
- Finally, choose which Hadoop processes should be launched in the node group. To do so, select each chosen processes’ check box from the Processes list. Consult your chosen Hadoop plug-in’s documentation for more information about each process.
- Click Create. The template should now appear in the Node Group Template table.
The Node Group Template table lists all available node group templates, their respective Hadoop plug-ins, and versions. This table also lists which Hadoop processes will run on each template’s nodes.
7. Configure Cluster Templates
After registering the required components (]) and configuring the types of nodes you need (xref:confignodetemplates[), you can now configure cluster templates. Cluster templates define the following settings for a cluster:
- The node composition of the cluster (as in, how many of each node type). Available node types are listed Project > Data Processing > Node Group Templates.
- The Hadoop plug-in and version used by the cluster.
- A list of processes that should only be launched once on a single host (anti-affinity).
- Cluster-level parameters for Hadoop services.
To create and configure a cluster template:
- In the dashboard, select Project > Data Processing > Cluster Templates.
- Use the Plugin Name and Version drop-down menus to select the name and version of the Hadoop plug-in that the cluster will use.
- Click Create.
- Enter a name for your template in the Template Name field.
- Use the Description field to describe the cluster template you are creating (optional).
- From the Use anti-affinity groups for list, check the box of each process that should not be launched more than once on a host.
Click the Node Groups tab. From here, configure how many nodes of each node type should run in the cluster. To do this:
- Use the Select a Node Group Template to add: drop-down menu to select a template (see Section 6, “Configure Node Group Templates” for related information).
- Click + to create a node group entry based on the template. Do this for each node group template you want to use in the cluster template.
For each node group entry, set how many instances of that node type should launch in the cluster.
You can also enter a name for each node group entry (optional).
- Configure Hadoop cluster settings by component as needed. Each component (for example, ZOOKEEPER, AMBARI, HUE) has its own parameters tab. Each tab contains related parameters, most of which also have brief contextual help. For more information on these parameters, consult your selected Hadoop plug-in’s documentation on supported functions for each component.
- Once you are satisfied with the configuration of your cluster template, click Create. Your template should now appear in the Cluster Templates table.
Alternatively, you can also use the Upload Template button to upload an existing cluster template from your local file system.
You can also launch a cluster using any existing templates from the Cluster Templates table. To do this, select Launch Cluster from the Actions drop-down menu of a cluster template. To view launched clusters, select Project > Data Processing > Clusters. For more information about launching a cluster, see Section 8, “Launch a Cluster”.
8. Launch a Cluster
Once you have a cluster template (see Section 7, “Configure Cluster Templates”), you can launch a cluster. To do so, open the Launch Cluster wizard. There are several ways to open this wizard:
- Select Project > Data Processing > Cluster Templates. From there, select Launch Cluster from the Action drop-down menu of a template.
- Select Project > Data Processing > Clusters. From there, click Launch Cluster; then, use the Plugin Name and Hadoop Version drop-down menus to select the name and version of the Hadoop plug-in that the cluster will use.
Afterwards, click Create. This will open the Launch Cluster wizard:
- In the Launch Cluster wizard, enter a name for the cluster in the Cluster Name field.
- Use the Description field to describe the cluster you are launching (optional).
- If you opened the Launch Cluster wizard through the Cluster table in Project > Data Processing > Clusters, you can use the Cluster Template drop-down menu to select a cluster template. You will only be able to select templates that are compatible with the Hadoop plug-in and version you selected earlier.
- Select a Hadoop image that the cluster should use from the Base Image drop-down menu. For details on creating and registering a Hadoop image, see ] and xref:registercomp[.
If needed, select a key pair from the Keypair drop-down menu. You can also click + beside this menu to create a new key pair. While key pairs are not required to launch a cluster, you will need them to log into cluster nodes (for example, through SSH).
For information on key pairs, see Manage Key Pairs.
- Select which network the cluster should use from the Neutron Management Network drop-down menu. For more details on adding and managing networks in OpenStack, see Common Administrative Tasks.
- Click Create to launch the cluster.
To view launched clusters, select Project > Data Processing > Clusters.
8.1. Scale or Delete a Cluster
OpenStack Data Processing allows you to easily scale an existing cluster to suit your resourcing needs. Scaling allows you to add or remove nodes of any type (or number) from a running cluster.
To view launched clusters, select Project > Data Processing > Clusters. You can scale or delete clusters from this page. For instructions on how to launch a cluster, see Section 8, “Launch a Cluster”.
To scale a cluster:
On the Clusters table, choose a cluster to scale. Then, select Scale Cluster from that cluster’s Action drop-down menu. .Scale the cluster as necessary:
- To add a new node type to the cluster, select its template from the Select a Node Group Template to add drop-down menu. Then, click the + button to add the node type to the cluster; you can then set how many nodes of that type should be added.
- To add or remove nodes from an existing node group, use the + or - buttons on the node group’s row. Alternatively, you can simply set the number of nodes in the node group’s Count field.
- Click Scale.
To delete a cluster, select Delete Cluster from its Action drop-down menu. You can also delete multiple clusters by selecting their check boxes and clicking the Delete Clusters button.
9. Configure Jobs
With the OpenStack Data Processing service, jobs define the actual data processing tasks. Each job specifies a job type (for example, Pig, Hive, or MapReduce), binary, script, and library. A job can only use binaries, scripts, or libraries that are registered with the OpenStack Data Processing service.
After creating a job, you can then launch it on a cluster and run it against an input data source. Input and output data sources, like job binaries, must also be registered first with the OpenStack Data Processing service (see Section 5.1, “Register Input and Output Data Sources”).
9.1. Register Job Binaries, Scripts, or Libraries
The process for registering job binaries, scripts, and libraries is similar to image and data source registration. You can register them directly from the Object Storage service; for instructions on how to upload objects to the Object Storage service, see Upload an Object. Alternatively, you can also upload binaries and libraries directly from your local file system directly into the OpenStack Data Processing service.
- In the dashboard, select Project > Data Processing > Job Binaries.
- Click Create Job Binary.
- Enter a name for the object (namely, your script, binary, or library). This name will be used when selecting the object. If your object requires a particular name or extension (for example, .jar), include it here.
- Use the Description field to describe the script, binary, or library you are registering (optional).
Configure the object depending on its storage type.
If the object is available through the Object Storage service, select Swift from the Storage type drop-down menu. Then:
- Provide the container and object name of your script, binary, or library as swift://CONTAINER/OBJECT in the URL field.
- If your script, binary, or library requires a login, supply the necessary credentials in the Username and Password fields
Otherwise, select Internal database from the Storage type drop-down menu. Then, use the Internal binary drop-down menu to either:
- Select an available binary, library, or script from the OpenStack Data Processing service, or
- Input a script directly into the dashboard (Create a script), or
- Upload a binary, library, or script directly from your local file system (Upload a new file).
- Click Create. The binary, library, or script should now be available in the Job Binaries table.
9.2. Create a Job Template
Once the required binaries, scripts, and libraries are registered with OpenStack Data Processing, perform the following steps:
- In the dashboard, select Project > Data Processing > Job Templates.
- Click Create Job Template.
- Enter a name for your job in the Name field.
- Select the correct type from the Job Type drop-down menu. For more information about job types, consult your chosen plug-in’s documentation regarding supported job types.
- Select the binary that should be used for this job from the Choose a main binary drop-down menu. The options in this menu are populated with job binaries and scripts registered with the OpenStack Data Processing service; for more information, see Section 9.1, “Register Job Binaries, Scripts, or Libraries”.
- Use the Description field to describe the job you are creating (optional).
- If the job binary you specified requires libraries, add them. To do so, click the Libs tab and select a library from the Choose libraries drop-down menu. Then, click Choose to add the library to the job; the library should be included in the Chosen libraries list. Repeat this for every library required by the job binary. Like binaries, the options in the Choose libraries drop-down menu are populated with libraries registered with the OpenStack Data Processing service. For more information, see Section 9.1, “Register Job Binaries, Scripts, or Libraries”.
- Click Create. The job should now be available in the Jobs table.
10. Launch Jobs
After creating a job, you can launch it to process data registered with the OpenStack Data Processing service (see ]. Jobs require a cluster; you can launch the job on an existing cluster (xref:launchexistingcluster[) or an entirely new one (Section 10.2, “Launch a Job on a New Cluster”).
Launching a job involves specifying input data sources and output data destination. Both objects must first be registered with the OpenStack Data Processing service. For more information, see Section 5.1, “Register Input and Output Data Sources”.
10.1. Launch a Job on an Existing Cluster
To view a list of existing clusters in the dashboard, select Project > Data Processing > Clusters. For information on how to launch a cluster, see Section 8, “Launch a Cluster”.
To launch a job on an existing cluster:
- In the dashboard, select Project > Data Processing > Jobs. The Jobs table displays all available job templates; see Section 9.2, “Create a Job Template” for details on creating new job templates.
- Choose which job template to use; then, select Launch On Existing Cluster from the job template’s Actions drop-down menu.
On the Launch Job wizard, select your input data source from the Input drop-down menu. Then, select your output destination from the Output drop-down menu.
If needed, you can also register your input data source or output destination from here. To do so, click the + on either Input or Output drop-down menus. Doing so will open the Create Data Source wizard; for more information, see Section 5.1, “Register Input and Output Data Sources”.
- From the Cluster drop-down menu, select which cluster the job should run on.
If you need to set any special job properties for this job, click the Configure tab. From there, click Add under either Configuration or Parameters to specify any special name/value pairs. You can specify multiple name/value pairs through this tab.
For more information about supported job properties, consult your chosen Hadoop plug-in’s documentation.
- Click Launch.
To view the status of launched jobs, select Project > Data Processing > Jobs. See Section 10.3, “Delete or Re-Launch Launched Jobs” for instructions on how to re-launch or delete a launched job.
10.2. Launch a Job on a New Cluster
After creating a job template, you can also use it to launch a job on an entirely new cluster. Doing so gives you the option to automatically kill the cluster after the job is finished.
- In the dashboard, select Project > Data Processing > Jobs. The Jobs table displays all available jobs; see Section 9.2, “Create a Job Template” for details on creating new jobs.
- Choose which job template to use; then, select Launch On New Cluster from the job template’s Actions drop-down menu.
- Use the plug-in Name and Version drop-down menus to select the name and version of the Hadoop plug-in that the job will use.
- Click Create.
- Enter a name for your cluster in the Cluster Name field.
- Select a Hadoop image that the cluster should use from the Base Image drop-down menu. For details on creating and registering a Hadoop image, see ] and xref:registercomp[.
If needed, select a key pair from the Keypair drop-down menu. You can also click + beside this menu to create a new key pair. While key pairs are not required to launch a cluster, you will need them to log into cluster nodes (for example, through SSH).
For information on key pairs, see Manage Key Pairs.
- Select which network the cluster should use from the Neutron Management Network drop-down menu. For more details on adding and managing networks in OpenStack, see Common Administrative Tasks.
- By default, the OpenStack Data Processing service will delete the cluster as soon as the job finishes. To prevent this from happening, select the Persist cluster after job exit check box.
- Next, click the Job tab. From there, select your input data source from the Input drop-down menu. Then, select your output destination from the Output drop-down menu.
- If needed, you can also register your input data source or output destination from here. To do so, click the + on either Input or Output drop-down menus. Doing so will open the Create Data Source wizard; for more information, see Section 5.1, “Register Input and Output Data Sources”.
If you need to set any special job properties for this job, click the Configure tab. From there, click Add under either Configuration or Parameters to specify any special name/value pairs. You can specify multiple name/value pairs through this tab.
For more information about supported job properties, consult your chosen Hadoop plug-in’s documentation.
- Click Launch.
To view the status of launched jobs, select Project > Data Processing > Jobs. See Section 10.3, “Delete or Re-Launch Launched Jobs” for instructions on how to re-launch or delete a launched job.
10.3. Delete or Re-Launch Launched Jobs
To view the status of launched jobs, select Project > Data Processing > Jobs. From here, you can delete or re-launch a job.
To delete a launched job, select Delete job execution from its Action drop-down menu. You can also delete multiple launched jobs by selecting their check boxes and clicking the Delete job executions button.
To re-launch a job on an existing cluster, select Relaunch on Existing Cluster from its Action drop-down menu. For instructions on how to continue, see Section 10.1, “Launch a Job on an Existing Cluster”.
Alternatively, you can re-launch a job execution on a completely new cluster. To do so, select Relaunch on New Cluster from its Action drop-down menu. For instructions on how to continue, see Section 10.2, “Launch a Job on a New Cluster”.
