Techdemo for Microprofile stuff

What is shown:

OpenAPI

Both, frontend and backend services, have the OpenAPI available under /openapi. There is also a UI under /openapi/ui.

Frontend

Uses a contract first approach and generates the REST interfaces from a given openapi.yaml file. Then it uses the interfaces to implement the REST service.

Backend

Defines JAX-RS REST services and enriches them with OpenAPI annotations.

Config

For example, the backend URL is configured in the frontend service in the class TolerantResource.

Health

The backend service has a HealthProbe that can be configured via REST call to fail a couple of times. Can be queried under /health.

Metrics

The REST services are @Timed and there is a @Counted on the health probe. Can be queried under /metrics.

REST client

The frontend uses the Microprofile REST-Client API to call the backend. This also has a "special" config property for the backend URL. (Currently it is commented out and replace with an implementation that allows to send the tracing context to the next server).

Fault Tolerance

The fault tolerant stuff is also visible in the metrics.

Demonstration 1: Retry

The tolerantRetry operation fails 80% of the time. With a @Retry you can show that it gets called multiple times (yes, there is an ugly hack to count the calls). To configure the retry behaviour you can change the value .....tolerantRetry/Retry/maxRetries=100 in the microprofile-config.properties file.

Demonstration 2: Timeout / Fallback

The tolerantTimeout operation takes a parameter that tells the backend how long to stall. If you make that larger than 1000ms then the frontent will timeout and use the fallback method.

Demonstration 3: CircuitBreaker / Fallback

This one is a bit more complicated. The tolerantCircuitBreaker method takes a parameter that tells it to fail or not. You should be able to observe the following behaviours:

Calling it with false will return status code 200 and an increasing counter
Hitting it with true will only count up three more times. The status code is 503 (service unavailable).
Waiting for 10 seconds will bring the circuit into its "half closed" state. Hitting it with true will count up one time (because it fails and will open the circuit again).
Switching back to false should close the circuit after two callls and after 10 second. (it happens earlier, but I don't know why yet)

Demonstration 4: Bulkhead

There is a class BadBulkheadClient that tries to bombard the frontend with calls. Depending on a flag you can set in the source, it will either use the method with or without a bulkhead. It will also try to get the system properties once every second. The program ends after 10 seconds.

This is what you should be able to observe.

With bulkhead

Many of the calls are not successful because the bulkhead "is full"
The system properties get queried 10 times

Without bulkhead

The calls take a long time
There are less than 10 calls to the system properties

Demonstration 5: Tracing

Both services support tracing. You need a running Zipkin server to record the calls.

There was a slight problem: currently only the ClientBuilder can be configured to transfer the tracing context to the next server. Have a look at BackendClientWithTracing class.

Future stuff

JWT
gRPC / RSocket?
Comparison to https://12factor.net/ and "Beyond The 12 Factor App"

OpenLiberty Bugs?

JSON Metrics ignore TimeUnit
Fallback does not work with private fallback method even though the example in the specs is made that way
Does the circuit breaker close too early?

Development

Project creation

mvn archetype:generate \
    -DarchetypeGroupId=net.wasdev.wlp.maven \
    -DarchetypeArtifactId=liberty-archetype-webapp \
    -DarchetypeVersion=2.6.1 \
    -DlibertyPluginVersion=2.6.1 \
    -DruntimeVersion=18.0.0.3 \
    -DgroupId=ch.ocram.microprofile.techdemo \
    -DartifactId=frontend \
    -Dversion=1.0.0-SNAPSHOT

Start a Swagger editor

docker run -d -p 7510:8080 swaggerapi/swagger-editor

PoliM/microprofile-techdemo