Skip to content

Latest commit

 

History

History
317 lines (257 loc) · 11.8 KB

run.md

File metadata and controls

317 lines (257 loc) · 11.8 KB
Raw

Run

Deploying into Knative means you have a Knative cluster ready and your $KUBECONFIG is set or that the ~/.kube/config file is pointing to the cluster's Kubernetes configuration file.

TwitterFn

kn service create twitter-fn \
		   --env TWITTER_API_KEY=$TWITTER_API_KEY \
		   --env TWITTER_API_SECRET_KEY=$TWITTER_API_SECRET_KEY \
		   --env TWITTER_ACCESS_TOKEN=$TWITTER_ACCESS_TOKEN \
		   --env TWITTER_ACCESS_TOKEN_SECRET=$TWITTER_ACCESS_TOKEN_SECRET \
		   --image docker.io/drmax/twitter-fn:latest
Creating service 'twitter-fn' in namespace 'default':

  0.496s Configuration "twitter-fn" is waiting for a Revision to become ready.
  0.521s ...
  6.986s ...
  7.148s ...
  7.268s Ready to serve.

Service 'twitter-fn' created with latest revision 'twitter-fn-njnks-1' and URL:
http://twitter-fn.knative-cluster.us-south.containers.cloud.ibm.com

WatsonFn

kn service create watson-fn \
		   --env WATSON_API_KEY=$WATSON_API_KEY \
		   --env WATSON_API_URL=$WATSON_API_URL \
		   --env WATSON_API_VERSION=$WATSON_API_VERSION \
		   --image docker.io/drmax/watson-fn:latest
Creating service 'watson-fn' in namespace 'default':

  0.197s Configuration "watson-fn" is waiting for a Revision to become ready.
  7.643s ...
  7.752s ...
  7.870s Ready to serve.

Service 'watson-fn' created with latest revision 'watson-fn-dnfxc-1' and URL:
http://watson-fn.default.knative-cluster.us-south.containers.cloud.ibm.com

SummaryFn

You need to then set the environment variables TWITTER_FN_URL and WATSON_FN_URL to the URLs that kn service create ... showed for the respective function creation. For instance:

export TWITTER_FN_URL=twitter-fn.knative-cluster.us-south.containers.cloud.ibm.com
export WATSON_FN_URL=watson-fn.knative-cluster.us-south.containers.cloud.ibm.com

Deploy the summary-fn service with:

kn service create summary-fn \
		   --env TWITTER_FN_URL=$TWITTER_FN_URL \
		   --env WATSON_FN_URL=$WATSON_FN_URL \
		   --image docker.io/drmax/summary-fn:latest
...

You can test the summary-fn function by going to the deployed function URL with your browser or by using curl.

Scaling

While by default, all services deployed to Knative will autoscale on demand, that is, scaling down to 0 if there are no requests to their endpoint, and automatically scaling up when a request arrives again, there are times when you need to control the scaling characteristics for each service.

In the current scenario, the requests to the WatsonFn service will be proportional to the number of images found from the TwitterFn requests. This could be orders of magnitude larger than TwitterFn requests and therefore we may want to parallelize the requests to the WatsonFn function.

We can control the scaling characteristics of any Knative service using the kn client service update command using the following flags:

  • --concurrency-target - recommendation for when to scale up based on the concurrent number of incoming requests. Defaults to --concurrency-limit when given.
  • -concurrency-limit - hard limit of concurrent requests to be processed by a single replica (single service instance).
  • --max-scale - maximum number of replicas.
  • --min-scale - minimum number of replicas.

There are also means to control the amount of memory and cpu to request for and to limit any function. However, for this example, let's use the concurrency flags above to modify the current deployment of the WatsonFn service such that we increase the concurency target and limit as well as the maximum and minimum scale values. The following command will achieve this:

kn service update watson-fn --concurrency-limit 1 \
                            --concurrency-target 1 \
                            --max-scale 10

Updating Service 'watson-fn' in namespace 'default':

  0.124s The Configuration is still working to reflect the latest desired specification.
  6.161s Traffic is not yet migrated to the latest revision.
  6.193s Ingress has not yet been reconciled.
  6.282s Waiting for VirtualService to be ready
  7.597s Ready to serve.

Service 'watson-fn' updated with latest revision 'watson-fn-gmdxr-7' and URL:
http://watson-fn.default.knative-cluster.us-south.containers.cloud.ibm.com

By limiting and targeting the concurrency to 1 and setting the max scale to 10, the WatsonFn will start with 1 replica and autoscale up to 10 replicas as requests comes in while each replica will be allowed to only process one request at a time.

A/B Testing or Blue/Green Deployment

In situation when you want to experiment with different working versions of your functions, Knative supports the ability to setup your service deployment with traffic splitting. This allows some percentage of requests to a service to be routed to a specific version and some other percentage to another. This in effect allows you to deploy your service to implement some
A/B test or do a Blue/Green deploy between revisions.

The Knative client kn has commands to help you configure traffic splitting easily on any of your services and their revisions. Let's explore one example with the summary-fn service. In the default implementation, the requests to twitter-fn and watson-fn are serialized. This results in slow responses in rendering the UI and therefore a suboptimal experience for end users. A better one would be to make the calls to watson-fn services asynchronous and populate the image classifications and the tag cloud dynamically as they become available.

Tagging Stable Revisions

However, let's first tag the current revision as stable. We do this by listing the revisions for summary-fn and tagging the latest one. We first need to find the revision name for the current (latest) and stable one.

kn revision list -s summary-fn
NAME                 SERVICE      GENERATION   AGE   CONDITIONS   READY   REASON
summary-fn-tcthz-1   summary-fn   8            10h   4 OK / 4     True
summary-fn-lljhz-1   summary-fn   7            20h   3 OK / 4     True

We can notice that the revision listed at the top is the most current revision with generation number 8. It is also ready and has name summary-fn-tcthz-1. We can use the name to tag it as latest, sync, and stable.

kn service update summary-fn --tag summary-fn-tcthz-1=latest \
							 --tag summary-fn-tcthz-1=stable \
							 --tag summary-fn-tcthz-1=sync
Updating Service 'summary-fn' in namespace 'default':

  0.310s Waiting for VirtualService to be ready
  1.114s Ready to serve.

Service 'summary-fn' updated with latest revision 'summary-fn-tcthz-1' (unchanged) and URL:
http://summary-fn-default.kndemo-267179.sjc03.containers.appdomain.cloud

Deploy New Async Revision

Next step is to deploy a better, faster, async version of the summary-fn service. The branch summary-fn-async has one such implemenation. The primary change in that branch is in the summary-fn service root / now points to the new async handler so that when users hit the summary-fn endpoint they will be using the async implementation of the UI that make use jQuery to call the watson-fn function versus doing so in a synchronous fashion, as in the stable (or sync) revision. A brief summary of the changes are in the following files.

cat ./funcs/summary/cmds.go
...
func (summaryFn *SummaryFn) summary(cmd *cobra.Command, args []string) error {
	if summaryFn.StartServer {
		http.HandleFunc("/", summaryFn.SummaryAsyncHandler)
...

The other change is in the file ./funcs/summary/async_layout.html which is now used to render the summary-fn UI. In there the primary change is to call the watson-fn function via jQuery.

...
	$(document).ready(function() {
	    $.get( "{{ $WatsonFnURL }}?q={{ $imageURL }}", function( data ) {
	      data.classifiers.forEach(function(c){
	        c.classes.forEach(function(b) {
	            $("#tw{{$i}}_{{$j}}").append("<div>"+b["class"]+" "+b.score+"</div>");
	            words.push({text: b["class"],
	                        weight: b.score*1000,
	                        link: "#tw{{$i}}_{{$j}}"})
	        });
	      });
	      $('#cloud').jQCloud(words);
	    });
	});
...

Next you will need checkout that branch and build the summary-fn function with the code in that branch; and then create an updated Docker image for the SummaryFn service; and push it to docker.io.

git co summary-fn-async
Switched to branch 'summary-fn-async'
./hack/build.sh
🕸️  Update
🧽  Format
⚖️  License
📖 Docs
🚧 Compile
────────────────────────────────────────────
success

./hack/build.sh --docker
🚧 🐳 build images
   🚧 🐳 twitter-fn
Sending build context to Docker daemon  58.05MB
...
🚧 🐳 summary-fn
Sending build context to Docker daemon  58.05MB
...
   🐳 summary-fn
The push refers to repository [docker.io/drmax/summary-fn]
...

Next, let's deploy the new async revision and tag it as such.

kn service update summary-fn \
		   --image docker.io/drmax/summary-fn:latest
...

Service 'watson-fn' updated with latest revision 'summary-fn-jtpyt-1' and URL:
http://watson-fn.default.knative-cluster.us-south.containers.cloud.ibm.com

kn service update summary-fn --tag summary-fn-jtpyt-1=async
Updating Service 'summary-fn' in namespace 'default':

  0.310s Waiting for VirtualService to be ready
  1.114s Ready to serve.

Service 'summary-fn' updated with latest revision 'summary-fn-jtpyt-1' (unchanged) and URL:
http://summary-fn-default.kndemo-267179.sjc03.containers.appdomain.cloud

Splitting Traffic

We can show the details of the summary-fn function so we can see the various revisions and tags in order split traffic 50/50 between the revisions tagged sync and async.

kn service describe summary-fn
Name:       summary-fn
Namespace:  default
Labels:     knfun=demo
Age:        10d
URL:        http://summary-fn-default.kndemo-267179.sjc03.containers.appdomain.cloud
Cluster:    http://summary-fn.default.svc.cluster.local

Revisions:
  100%  @latest (summary-fn-tcthz-1) [8] (24m)
        Image:  docker.io/drmax/summary-fn:latest (pinned to f43b43)
     +  summary-fn-tcthz-1 (current @latest) #latest [8] (24m)
        Image:  docker.io/drmax/summary-fn:latest (pinned to f43b43)
     +  summary-fn-tcthz-1 (current @latest) #stable [8] (24m)
        Image:  docker.io/drmax/summary-fn:latest (pinned to f43b43)
     +  summary-fn-tcthz-1 (current @latest) #sync [8] (24m)
        Image:  docker.io/drmax/summary-fn:latest (pinned to f43b43)
     +  summary-fn-jtpyt-1 (current @latest) #async [9] (28m)
        Image:  docker.io/drmax/summary-fn:latest (pinned to f43b43)

Conditions:
  OK TYPE                   AGE REASON
  ++ Ready                   5s
  ++ ConfigurationsReady    24m
  ++ RoutesReady             5s

Splitting traffic is a service update command that uses the --traffic flag to set the different revisions and the traffic percentage values.

kn service update summary-fn --traffic async=50,sync=50
Updating Service 'summary-fn' in namespace 'default':

  0.310s Waiting for VirtualService to be ready
  1.114s Ready to serve.

Service 'summary-fn' updated with latest revision 'summary-fn-jtpyt-1' (unchanged) and URL:
http://summary-fn-default.kndemo-267179.sjc03.containers.appdomain.cloud

Now when users access the URL for the summary-fn function they will (on average) see 50% the sync page (which takes a few seconds to reder) and 50% the async version which is instantaneous and asynchronously shows image features and the tag cloud of all the images features.