Managed Applications on Clusters

Prev Next

Overview

You can now run certain third-party applications on CNodes in your VAST Cluster. Specifically, you can run applications such as the Spark database query engine on designated CNodes, and share compute resources on these CNodes between VAST storage and the applications.

The applications use the VAST connector to optimize access to a VAST DataBase.

Viewing Applications

The Applications tab in the Data Engine page of the VAST Web UI shows the applications that are running on application designated CNodes.

Creating Applications

You can create Spark applications on CNodes. Spark applications consist of Master and Worker nodes, both of which are configured in the settings below.

  1. On the VAST Web UI navigate to the Data Engine page, and select the Application tab.

  2. Click Create Application.

  3. Enter details for the application In the General section. In this version, only Spark can be selected as the application.

    Field

    Description

    App Name

    The name of the Application (example: Spark)

    Application

    Spark (only option in this version)

    Image Tag

    Select the application image from the dropdown list (Spark 3.4).

  4. In the Resource Selection section, select the CNodes on which the application will run from the All Possible CNodes box (on the right) , and move them to the Selected CNodes box (on the left). The selected CNodes are the application CNodes.

    Note

    It is not recommended to select all CNodes for Applications. This could impair Cluster performance.

  5. In the Resources Limitation section, optionally set limits on the CNode resources the application can use. In the Use up to field, set the maximum CPU percentage the app can use (between 20% and 60%, in increments of 5%)

  6. In the Network section enter details for the Spark Master and Worker nodes. Each node requires a Virtual IP (VIP) address. Select these from the Virtual IP pools allocated for the Cluster (in the Virtual IP tab of the Network Access page).

    For the Master node, enter a VIrtual IP address.

    Field

    Description

    Virtual IP

    The Virtual IP address of the Spark Master node (from the Cluster VIP pool).

    For the Worker nodes, enter a list of Virtual IP addresses.

    Field

    Description

    Virtual IP

    A list of Virtual IP addresses for the Worker nodes (form the Cluster VIP Pool).

    Optionally, set advanced network details for the Master and Worker nodes:

    Field

    Description

    Netmask

    The subnet mask of the Virtual IP assigned to the Master and Worker nodes.

    Gateway IP

    The IP address of the gateway of the Virtual IP assigned to the  Master and Worker nodes.

    VLAN

    If you want to tag the Virtual IP pool with a specific VLAN on the data network, enter the VLAN number (0-4096). See Tagging Virtual IP Pools with VLANs.Tagging Virtual IP Pools with VLANs

  7. Click Create. The application is created on the selected CNodes. The application images are loaded on the selected CNodes, and then started. This can take some time.

    Monitor progress on the Activities page of the VAST Web UI: events appear indicating that the application creation has completed (event name: create_managed_application). When this process is complete, the application appears in the list of applications in the Applications page.

Initializing and Starting the Spark Application Cluster on the CNode

After the Spark application is created and deployed on the CNode, it must be initialized and started.

  1. On the Data Engine page, select the Applications tab.

  2. Select the application in the list of applications deployed on the Cluster. The status for the applications is INIT after it is created.

  3. Right-click on the application, and click Start. The status for the application changes to RUN as it starts to run. Right-click on the application and click View CNode State, to monitor the status of the application.

Accessing the Spark Master UI

The Spark Master and Worker UIs are available on the VIPs at port 9292 (for the Master) and 9293 (Workers).

Setup Spark Clients to Access the CNode Spark Cluster

Follow these steps to configure a client host to connect to the Spark Cluster running on the VAST Cluster CNode.

Using the Standalone Spark Driver

  1. Download the Spark driver (3.4.1):

    wget -q -4 https://archive.apache.org/dist/spark/spark-3.4.1/spark-3.4.1-bin-hadoop3-scala2.13.tgz
    tar xvf spark-3.4.1-bin-hadoop3-scala2.13.tgz
    mv spark-3.4.1-bin-hadoop3-scala2.13 /opt/spark

    Note

    The version should match the version running on the VAST Cluster (spark 3.4.1 with Scala-13).

  2. Install the Spark driver (and include the connector jars in the Spark location for the application):

    curl -fsSL -o spark-vast-plugin.zip "https://github.com/vast-data/vast-db-connectors/releases/download/spark3-vast-3.4.1-f93839bfa38a/spark3-vast-3.4.1-f93839bfa38a.zip"
    unzip spark-vast-plugin.zip
    mv spark3-vast* /opt/spark/vast
  3. Run these commands for the S3A Hive interface to include jars in the Spark vanilla jars library:

    wget -q https://repo1.maven.org/maven2/com/amazonaws/aws-java-sdk-bundle/1.11.1026/aws-java-sdk-bundle-1.11.1026.jar
    wget -q https://repo1.maven.org/maven2/org/apache/hadoop/hadoop-aws/3.3.2/hadoop-aws-3.3.2.jar
    mv hadoop-aws-3.3.2.jar /opt/spark/jars/
    mv aws-java-sdk-bundle-1.11.1026.jar /opt/spark/jars/

Using the Docker Spark Driver

You can avoid setting up Spark env on your host, by deploying a pre-built Spark-Vast image, packed up with the VastDB connector and ready to use.

Expose the driver port and host network and deploy the driver image:

  • Run this docker command:

    docker run -d --name spark-client \
    --net=host \
    -p 4040:4040 \
    -v $PWD/spark-scripts:/tmp/spark-scripts \
    vastdataorg/spark-vast /bin/sleep infinity

    Note

    This is an example, Using Docker volume mounts (spark-scripts), you can expose to the docker driver your scripts/applications to submit on the spark cluster.

Configuring Spark Executors

Optimizing the allocation of executors on Spark can significantly improve the workload performance.

There are two primary approaches to setting up executors for a Spark job:

  • Static Allocation

  • Dynamic Allocation

Static Allocation

Static allocation involves setting specific numbers of executors, cores, and memory ahead of time. To calculate these values effectively, compare the total resources on each worker (CPU cores and memory) to the resources requested per executor. For example, if each worker has 16 cores and 64 GB RAM, and you allocate 5 cores and 20 GB RAM per executor, you can run three executors per worker without over-allocating. This approach ensures each executor has sufficient resources without straining the system. These parameters remain fixed for the lifetime of the application.

Key Configuration Parameters

  • spark.executor.cores. The number of cores per executor.

  • spark.executor.memory. The amount of memory allocated to each executor.

  • spark.executor.instances. The number of executors.

Example:

--conf spark.executor.cores=5 \
--conf spark.executor.memory=16g \
--conf spark.executor.instances=3 \

Note

Setting spark.executor.instances is optional because you can indirectly control the number of executors with spark.executor.cores and spark.executor.memory. By configuring these, you define the size of each executor. The number that fit into your cluster depends on the total resources available on each worker node.

Advantages & Disadvantages of Static Allocation

Advantages
  • Resource allocation is predictable, which can be easier to manage in environments with fixed capacities.

  • Simple to configure as it doesn't require additional overhead to manage scaling.

Disadvantages
  • May lead to underutilization or overutilization of resources depending on the workload, as the number of executors does not change in response to the job's demands.

  • Not ideal for varying workloads where the processing needs can change dramatically over time.

Dynamic Allocation

Dynamic allocation enables Spark to add or remove executors dynamically based on the workload. This means Spark can request more executors when there is a high demand for processing and can release executors when the demand decreases.

Key Configuration Parameters

  • spark.dynamicAllocation.enabled. Enables dynamic allocation.

  • spark.dynamicAllocation.minExecutors. Minimum number of executors Spark will maintain.

  • spark.dynamicAllocation.maxExecutors. Maximum number of executors Spark can allocate.

  • spark.dynamicAllocation.initialExecutors. Initial number of executors Spark should start with.

  • spark.dynamicAllocation.executorIdleTimeout. Duration after which an idle executor is removed.

Advantages & Disadvantages of Dynamic Allocation

Advantages
  • Improved Resource Utilization: Adjusts the number of executors based on the workload, potentially leading to better utilization of cluster resources.

Disadvantages
  • Complexity: Requires a more sophisticated setup, including proper configuration of the Spark cluster manager and possibly fine-tuning the parameters for optimal performance.

  • Potential for Latency: Scaling decisions aren't instantaneous, which can introduce delays in executor provisioning, impacting job start times or scaling reactions to workload changes.

Choosing Between Static and Dynamic Allocation

Use Static Allocation when you have a predictable workload, or when operating in a static cluster environment.

Use Dynamic Allocation for jobs with varying workloads, where resource utilization efficiency is critical, It's also beneficial when cluster workloads are unpredictable and diverse, making it hard to manually determine the optimal number of executors.

Starting Spark Sessions on a Client

Submitting a Spark Session using PySpark or Spark-Submit

Using Dynamic Allocation

Example of submitting a PySpark script to the spark master with dynamic executors allocation:

/opt/spark/bin/pyspark --master spark://<master-ip>:2424 \
--conf spark.dynamicAllocation.enabled=true \
--conf spark.dynamicAllocation.minExecutors=3 \
--conf spark.dynamicAllocation.maxExecutors=6 \
--conf spark.dynamicAllocation.initialExecutors=4 \
--conf spark.dynamicAllocation.executorIdleTimeout=120s \
--conf spark.driver.maxResultSize=4g \
--driver-class-path $(echo /opt/spark/vast/*.jar | tr ' ' ':') \
--conf spark.executor.extraClassPath=$(echo /opt/spark/vast/*.jar | tr ' ' ':') \
--jars $(echo /opt/spark/vast/*.jar | tr ' ' ',') \
--conf spark.executor.userClassPathFirst=true \
--conf spark.driver.userClassPathFirst=true \
--conf spark.driver.host=10.71.16.91 \
--driver-memory 32g < pyspark_app.py

Spark-submit can also be used in place of pyspark in the example above.

Using Static Allocation

Example of submitting a PySpark script to the spark master with Static executors allocation:

/opt/spark/bin/pyspark --master spark://<master-ip>:2424 \
--conf spark.executor.memory=10g \
--conf spark.executor.cores=3 \
--conf spark.driver.maxResultSize=4g \
--driver-class-path $(echo /opt/spark/vast/*.jar | tr ' ' ':') \
--conf spark.executor.extraClassPath=$(echo /opt/spark/vast/*.jar | tr ' ' ':') \
--jars $(echo /opt/spark/vast/*.jar | tr ' ' ',') \
--conf spark.executor.userClassPathFirst=true \
--conf spark.driver.userClassPathFirst=true \
--conf spark.driver.host=10.71.16.91 \
--driver-memory 32g < pyspark_app.py

Spark-submit can also be used in place of pyspark in the example above.

Running a Spark-SQL Session to Query the VAST Database

  1. Run the following command to start spark-sql:

    /opt/spark/bin/spark-sql --master spark://<master-vip>:2424 \
      --driver-class-path $(echo /opt/spark/vast/*.jar | tr ' ' ':') \
      --conf spark.executor.extraClassPath=$(echo /opt/spark/vast/*.jar | tr ' ' ':') \
      --jars $(echo /opt/spark/vast/*.jar | tr ' ' ',') \
      --conf spark.executor.userClassPathFirst=true \
      --conf spark.driver.userClassPathFirst=true \
      --conf spark.driver.maxResultSize=4g \
      --conf spark.driver.memory=16g \
      --conf spark.executor.cores=2 \
      --conf spark.executor.memory=4g \
      --conf spark.ndb.endpoint=http://172.19.197.1 \
      --conf spark.ndb.data_endpoints=http://127.0.0.1 \
      --conf spark.ndb.access_key_id=Q00Q0.. \
      --conf spark.ndb.secret_access_key=IpKLQnvx.. \
      --conf spark.ndb.num_of_splits=64 \
      --conf spark.ndb.num_of_sub_splits=8 \
      --conf spark.ndb.rowgroups_per_subsplit=1 \
      --conf spark.ndb.query_data_rows_per_split=4000000 \
      --conf spark.ndb.retry_max_count=3 \
      --conf spark.ndb.retry_sleep_duration=1 \
      --conf spark.ndb.parallel_import=true \
      --conf spark.ndb.dynamic_filter_compaction_threshold=100 \
      --conf spark.ndb.dynamic_filtering_wait_timeout=2 \
      --conf spark.sql.catalog.ndb=spark.sql.catalog.ndb.VastCatalog \
      --conf spark.sql.extensions=ndb.NDBSparkSessionExtension \
      --conf spark.python.authenticate.socketTimeout=1m \
      --conf spark.driver.host=10.71.16.91 \
      --conf spark.sql.catalogImplementation=in-memory

    where <master-vip> is the VIP of the Master node.

  2. Run queries from spark-sql:

    spark-sql> select * from `ndb`.`vastdb1`.`schema1`.`customer` limit 10;
Use spark-sql with Hive to access a Vast Database with S3 data

You can use Hive to access S3 data in the VAST Database.

  • Run the following command (which includes --conf spark.sql.catalogImplementation=hive, for Hive):

    /opt/spark/bin/spark-sql --master spark://172.19.197.21:2424 \
      --driver-class-path $(echo /opt/spark/vast/*.jar | tr ' ' ':') \
      --conf spark.executor.extraClassPath=$(echo /opt/spark/vast/*.jar | tr ' ' ':') \
      --jars $(echo /opt/spark/vast/*.jar | tr ' ' ',') \
      --conf spark.executor.userClassPathFirst=true \
      --conf spark.driver.userClassPathFirst=true \
      --conf spark.driver.maxResultSize=4g \
      --conf spark.driver.memory=16g \
      --conf spark.executor.cores=2 \
      --conf spark.executor.memory=4g \
      --conf spark.ndb.endpoint=http://172.19.197.1 \
      --conf spark.ndb.data_endpoints=http://127.0.0.1 \
      --conf spark.ndb.access_key_id=Q00Q0.. \
      --conf spark.ndb.secret_access_key=IpKLQnvx.. \
      --conf spark.ndb.num_of_splits=64 \
      --conf spark.ndb.num_of_sub_splits=8 \
      --conf spark.ndb.rowgroups_per_subsplit=1 \
      --conf spark.ndb.query_data_rows_per_split=4000000 \
      --conf spark.ndb.retry_max_count=3 \
      --conf spark.ndb.retry_sleep_duration=1 \
      --conf spark.ndb.parallel_import=true \
      --conf spark.ndb.dynamic_filter_compaction_threshold=100 \
      --conf spark.ndb.dynamic_filtering_wait_timeout=2 \
      --conf spark.sql.catalog.ndb=spark.sql.catalog.ndb.VastCatalog \
      --conf spark.sql.extensions=ndb.NDBSparkSessionExtension \
      --conf spark.python.authenticate.socketTimeout=1m \
      --conf spark.driver.host=10.71.16.91 \
      --conf spark.hadoop.fs.s3a.access.key=Q00Q0.. \
      --conf spark.hadoop.fs.s3a.secret.key=IpKLQnvx.. \
      --conf spark.hadoop.fs.s3a.impl=org.apache.hadoop.fs.s3a.S3AFileSystem \
      --conf spark.hadoop.fs.s3a.endpoint=http://172.19.197.1 \
      --conf spark.hadoop.fs.s3a.path.style.access=true \
      --conf spark.hadoop.mapreduce.outputcommitter.factory.scheme.s3a=org.apache.hadoop.fs.s3a.commit.S3ACommitterFactory \
      --conf spark.hadoop.fs.s3a.ssl.enabled=false \
      --conf spark.hadoop.hive.metastore.uris=thrift://v197lg6:9083 \
      --conf spark.sql.catalogImplementation=hive

Accessing the VAST Database using the Spark Driver Web UI

You can monitor progress of Spark sessions submitted using pyspark or spark-submit using the Spark Driver Web UI.

Browse to the client IP on port 4040 (the default port for the Spark Web UI).

Removing the Spark Application

  1. On the Data Engine page, select the Applications tab.

  2. Right-click on the application, and click Remove.

    Monitor progress on the Activities page of the VAST Web UI: events appear indicating that the application removal has completed (event name: delete_managed_application). When this process is complete, the application is removed from the list of applications in the Applications page.

HA Operability Issues

At least two CNodes must be selected to host applications, to allow for continued operation in the event one CNode fails.

If the CNode running the app fails, the app will be started on another CNode designated for apps.