Integrating a "Hello World" Job Type

For definitions of terminology used, please refer to our terminology reference.

When installation (see Installation) and setup of your HySDS cluster (see Cluster Setup) is complete, you can start integrating job types (PGE or non-PGE) into the system. In this tutorial, we integrate our first job type which will simply echo "Hello World".

1. On themozartinstance, create a directory for your new job type

   cd ~/mozart/ops

   
   
   

   
   
   

   
   
   

   mkdir hello_world

   
   
   

   
   
   

   
   
   

   
   
   

   
   
   

   
   
   

   cd hello_world

   
   
   

   
   
   

   
   
   

2. Let's create the script that runs our job type. Open up a new file

   and paste the contents below

   
   
   

   
   
   

   
   
   

   
   
   

3. and save.

4. Set the script to be executable

   chmod 755 run_hello_world.sh

5. Test the script by running it

   ./run_hello_world.sh

   Hello World Hello World to STDERR

6. Now let's create the necessary files for integrating this job into your HySDS cluster. Create the docker directory and cd into it

   mkdir docker

   cd docker

7. Create a Dockerfile for your job type by opening up a new file Dockerfile.<job_type_name>

   vi Dockerfile.hello_world

   paste the following into it

   and save.

8. Create a job-spec file for your job type by opening up a new file job-spec.json.<job_type_name>

   paste the following into it

   and save. For more information on the job-spec format, check out Job and HySDS IO Specifications.

9. Create a hysds-io file for your job type by opening up a new file hysds-io.json.<job_type_name>

   paste the following into it

   and save. For more information on the hysds-io format, check out Job and HySDS IO Specifications.

10. Change directory back up to the root of your job type directory

    and the layout of the contents should be as follows

11. Now we create a git repository from this directory

12. Create a new repository on github, copy the repo URL, and push your existing hello_world repository. In this example, I created a new repo under my github user account, https://github.com/pymonger/hello_world.git

13. Ensure that your HySDS cluster is up and running

    If it is not running, run sds start all.

14. Now we need to configure our new job type for continuous integration and deployment. Since we're still developing this job type, we'll configure it using the master branch instead of release tags

    If jenkins needs to use an OAuth token to access the repo, specify the -k option and ensure that GIT_OAUTH_TOKEN was specified in $HOME/.sds/config:

    You should have output similar to the following

    Follow these instructions to configure the webhook on github so that it can trigger rebuild and deployment.

    Note: If you run into issues like :

    you may have a jenkins-cli.jar version that does not support -http, -auth parameters. You can either get a latest jenkins-cli.jar, or can make the following change line in cluster.py to make it work:

    to:

    If the build fails with following error (aws encryption type) :

    remove the encryption type value as AES256 is default type in cluster.py:

    to:

    if your build fails with import error for python-pyasn1 or rsa:

    you may not have latest version of the packages. use pip to install them (specially in ci and factotum instances):

    yum may not install the latest version of pyasn1_modules.

15. Navigate to your ci instance (e.g. http://<your-ci-instance>:8080) to validate that the jenkins job was configured

    You should see your jenkins job named as "container-builder_<your repo name>_<branch>" (e.g. container-builder_gmanipon_hello_world_master).

16. Verify that the jenkins job runs to completion by manually scheduling a build. The jenkins job will build the docker image, publish it to your S3 code bucket, and register the job types into you HySDS cluster. Click on the green arrow and view the console output to validate.

17. Verify that your new job type was published to your HySDS cluster. Open up a browser and point it to your mozart instance's ElasticSearch head plugin (e.g. http://<your-mozart-instance>:9200/_plugin/head)*

    Alternatively, you can query for the the "Hello World" job-spec using curl

18. Now let's try to run our hello_world job type using the Dataset Faceted Search interface (aka tosca) running on the grq instance. First, let's ingest a dataset so that we have at least 1 dataset in our catalog. On the mozart instance run

19. Navigate to your grq instance (e.g. http://<your-grq-instance>/search)* to validate that the AOI dataset was ingested

20. Next click on the "On-Demand" button. If you don't see the "On-Demand" button, then the AOI dataset wasn't ingested correctly. Otherwise, a modal titled "On-Demand (Process Results)" should pop up as below

    Enter the value "test" into the "Tag" field, select "Hello World [master]" for the "Action" field, and select "job_worker-small" for the "Queue" field (should already be selected because it is the "recommended-queue" based on our job-spec.json.hello_world that we composed earlier). You can leave the "Priority" field at "0" for now

    Click on "Process Now".

21. You will then get another modal providing you with a link to the HySDS Resource Management interface (aka figaro) for the job you just submitted. Click on the link to open up figaro. If the link doesn't open up correctly, there may be a routing issue with the private vs. public IP address in the tosca configuration. Alternately, navigate to your mozartinstance (e.g. http://<your-mozart-instance>/figaro)* to view your job. Click on the "Queued" button to facet down to the jobs that are in a "job-queued" state

    For advanced users, navigate to the RabbitMQ Admin interface (e.g. http://<your-mozart-instance>:15672/#/queues) running on your mozart instance to view the queued jobs and tasks

    It will be stuck in the job-queued because there are no workers configured to process jobs from the "job_worker-small" queue.

22. Let's configure your factotum instance to start up a worker that will pull jobs from the "job_worker-small" queue. What we will do is copy the default supervisord.conf template for factotum to ~/.sds/files, and modify it to include start up of celery workers for the job_worker-small queue:

    Edit the supervisord.conf.factotum file

    append the following lines to the bottom of the file and save

23. Since there are no jobs running in your cluster, we can safely update the factotum instance

    Check the status of supervisord on factotum

    We're ready to update if it shows

    Finally update and start up factotum:

    Rerun sds status factotum to verify that your new worker for job_worker-small is RUNNING

24. Navigate to your mozart instance (e.g. http://<your-mozart-instance>/figaro)* to view your job again. Click on the "Started" button to facet down to the jobs that are in a "job-started" state

    When the job is finished executing, you should have a job that is in a "job-completed" state

25. Click on the "work directory" link to view the live work directory on factotum. All STDOUT output will go into the _stdout.txt file and all STDERR output will go into the _stderr.txt. Here's what _stdout.txt should show

    and what the _stderr.txt should show

    Note:

    If you run into issues like:

    it could be due to permission issue in aws. Check the permission tab in 'grfn-v2-ops-product-bucket' to ensure you have the same configuration, specially, the policy file (dont forget to edit the bucket name in your policy file). Also download the index.html file from there and add to your bucket (do necessary edits to reflect your bucket)

That's it! Congratulations, you've integrated a new job type into your HySDS cluster and successfully executed it.

Next Steps

Now that you've integrated your first job type, continue on to Step 5: Integrating a "Hello World" Dataset Type to integrate a dataset produced by this job type.