This example builds on top of solution 5.
In this example we substitute the basic console client with a Web Api - the cluster is called from a very simple Web Api containing only one GET action. The example also demonstrates the usage of CancellationTokens and the Orleans wrapper GrainCancellationTokens.
The demo illustrates the usage of a cluster of silos, one client and a load test running against that client (for the sake of the illustration, we activate a grain for each call the api makes to the silo cluster).
The Web Api's GET action can be accessed by default using the following URL: http://localhost:5432/helloworld?name=Piotr.
As in example 5, following scripts can be used to run silos:
./run-docker-silo.sh./run-local-silo.sh
Important thing to note is that if one wants to run multiple silos which form a cluster, following variables need to be made unique for every silo:
GATEWAYPORT(e.g.3001,3002, etc.)SILOPORT(e.g.2001,2002, etc.)DASHBOARDPORT(e.g.8081,8082, etc.)
As in example 5, following scripts can be used to run clients:
./run-client-docker.sh./run-client-local.sh
Now we get to a bit more serious testing as it's easier to laod test the setup. Here is what's happening:
- left: cluster of silos running.
- right: Web Api running.
Load test results - vegeta:
Load test results - orleans dashboard:
Load test results - orleans dashboard:
Grain distribution across the cluster of silos:
Ideally vegeta, but in case it is not installed, ./run-demo-load-test.sh should be modified to invoke the preferred load tester.
- run the cluster:
./run-demo-cluster.sh(3 silos: 2 hosted in docker, 1 hosted locally) - run the client:
./run-demo-client.sh(only one client needed in this case, vegeta is going to do the heavy lifting) - run the load test:
./run-demo-load-test.sh(vegeta script generating 100 requests per second of constant load)