hkubectl is a tool which allows you to create most of the operations that you need in order to use hkube

• first lets download hkubectl curl -L https://github.com/kube-HPC/hkubectl/releases/download/v1.1.7/hkubectl.tgz | tar xvz
• add hkubectl to root cp ./hkubectl /usr/local/bin
• run hkubectl -h as you can see there a lot of options available
• configuring hkubectl to the cluster hkubectl config set endpoint http://master/hkube/api-server/
• for example in order to view all available algorithms hkubectl --endpoint=http://master/hkube/api-server/ algorithm get

Create your first algorithm

lets create our first algorithm

Adding algorithm to hkube is a simple process. all you have to do is to implement three apis init,start and stop. for more information please check http://hkube.io/code/api/

Important note: I believe that you wondering how logs can be shown via Hkube. hkube will do it for you, you can see all your logs via kibana including hkube internal data such as pipeline name and algorithm name

hkube already has built in samples in python,javaScript and java. for this tutorial we will use javaScript

• lets clone javascript sample git clone https://github.com/kube-HPC/algorithm-node-example

  • now enter to the folder cd algorithm-node-example and run npm install

  lets open it from the editor viewer. first navigate to folder algorithm-node-example/lib
  as you can see theres couple of files here lets open algorithm-ws.js which implement the api with websocket. as you can see there are couple fo functions there, lets focus on the _handleMessage() function in line 51, the function demonstrate the flow you have to implement in order to use hkube.

  in more descriptive way :

  1. initialize- you do your initialization thing and send initialize message back.
  2. start - after initialize will sent the worker will send start for executing your algorithm and send started back
  3. done - after the algorithm ends done message should be send back to the algorithm with its result

lets focus on _start() function in line 104

as you can see it referencing to myAlgorithm file with inputs. this is actually the algorithm we wish to run. l

• lets move to this file lib/My-algorithm.js
• lets change the return under the process method in line 4 to returnhello ${input[0]}``

deploying algorithm to hkube

lets build our container

• run docker build -t alg-example:1 .
• now hkubectl algorithm add alg-example --image=alg-example:1 --cpu 0.1 --mem 128Mi

let install hkube

since we already started from kubernetes image we need to install hkube as a layer on top of it

hkube has helm chart for installing its third parties and core modules in. but, for this tutorial we will run a slim version of it

• first we will install hkube core modules via helm helm install --name core https://github.com/maty21/katacoda-helm-core/raw/master/core/core-0.1.0.tgz

• now lets install a slim version of hkube third party helm install --name third-party https://github.com/maty21/katacoda-helm-core/raw/master/third-party/third-party-0.1.0.tgz

• run kubectl get po -w and verify that all of the pods in running state

  after hkube is installed on our cluster we are ready to go for the next step

now we are ready to run our first algorithm

• on the editor menu enter to the pipeline folder and on terminal cd pipeline

as you can see there is a JSON file which demonstrates the pipeline flow lets move on the basics

the descriptor has couple of base properties:

• name - describe the pipeline name
• nodes - which describe algorithm we wish to run on this pipeline we will detail about it
• flowInput - an external input you wish to add to the current execution
• webhook - a http server that will be called in two different phases status and result

there are more options . if you wish to learn more visit http://hkube.io/learn/

• lets change flowinput.name to your name
• now in the terminal lets run curl -X POST -H "Content-Type: application/json" -d @"pipeline.json" http://master/hkube/api-server/api/v1/exec/raw/

• now lets get the results for doing it we need the jobid that recived from the previous stage curl http://master/hkube/api-server/api/v1/exec/results/{jobid}