Skip to content

Latest commit

 

History

History
87 lines (61 loc) · 3.64 KB

README.md

File metadata and controls

87 lines (61 loc) · 3.64 KB
Raw

Cell Tower Anomaly Detection using Serverless Spark (with Dataproc UI Instructions)

Lab contributed by: TEKsystems

Overview

With the advent of cloud environments, the concept of huge capital investments in infrastructure in terms of capital and maintenance is a thing of the past. Even when it comes to provisioning infrastructure on cloud services, it can get tedious and cumbersome.

In this example, you will look at executing a simple PySpark code which runs on Serverless batch (a fully managed Dataproc cluster). It is similar to executing code on a Dataproc cluster without the need to initialize, deploy or manage the underlying infrastructure.

This usecase is used to check the performance of the celltower, if the performance is good then it means that the maintainence is not required and it functions properly if its performance is bad then it has to be put under maintainence

Services Used

  • Google Cloud Storage
  • Google Cloud Dataproc
  • Google Cloud Bigquery
  • Google Cloud Composer
  • Airflow DAG

3. Permissions / IAM Roles required to run the lab

Following permissions / roles are required to execute the serverless batch

  • Viewer
  • Dataproc Editor
  • BigQuery Data Editor
  • Service Account User
  • Storage Admin

4. Checklist

To perform the lab, below are the list of activities to perform.

1. GCP Prerequisites
2. Spark History Server Setup
3. Uploading scripts and datasets to GCP
4. Creating cloud composer environment
5. Creating a BigQuery Dataset

Note down the values for below variables to get started with the lab:

PROJECT_ID=                                         #Current GCP project where we are building our use case
REGION=                                             #GCP region where all our resources will be created
SUBNET=                                             #subnet which has private google access enabled
BQ_DATASET_NAME=                                    #BigQuery dataset where all the tables will be stored
BUCKET_CODE=                                        #GCP bucket where our code, data and model files will be stored
BUCKET_PHS=                                         #bucket where our application logs created in the history server will be stored
HISTORY_SERVER_NAME=spark-phs                       #name of the history server which will store our application logs
USMA_NAME=serverless-spark                          #user managed service account required for the PySpark job executions
SERVICE_ACCOUNT=$USMA_NAME@$PROJECT_ID.iam.gserviceaccount.com
NAME=<your_name_here>                               #Your Unique Identifier
COMPOSER_SA=composer-sa
COMPOSER_ENV=                                       #Your Cloud Composer Environment Name

5. Lab Modules

The lab consists of the following modules.

  1. Understand the Data
  2. Solution Architecture
  3. Data Preparation
  4. Model Training and Evaluation
  5. Examine the logs
  6. Explore the output

There are 3 ways of performing the lab.

Please chose one of the methods to execute the lab.


6. CleanUp

Delete the resources after finishing the lab.
Refer - Cleanup