Skip to content

README.md

Latest commit

 

History

History
78 lines (64 loc) · 6.89 KB

README.md

File metadata and controls

78 lines (64 loc) · 6.89 KB

Adding Smoke E2E for a Language SDK

For the steps below, take statefun-smoke-e2e-golang module as a reference implementation. You can copy the static config/protobuf files from there.

Implementation

Step 1: Create a new module and generate the protobuf bindings

  • Create a new module named statefun-smoke-e2e-LANGUAGE under statefun-e2e-tests.
  • Make statefun-smoke-e2e-multilang-base as parent of the created module in pom.xml.
  • Copy commands.proto from the reference implementation into the created module under src/main/protobuf.
  • Make language specific changes to the commands.proto. For example, the following line is added into commands.proto to generate GoLang bindings:
option go_package = ".;app";
  • Generate language specific protobuf bindings under src/main/LANGUAGE. This enables the function to serialize/deserialize the messages sent/received.
  • Copy remote-module/module.yaml, Dockerfile, log4j.properties into src/test/resources. These files are for launching the Smoke E2E driver, hence no code changes needed.

Step 2: Implement the CommandInterpreterFn as an HTTP endpoint

  • CommandInterpreterFn is a remote function that performs the operation based on the command received. Following is the high level introduction to the logic to be implemented in the function. One can also look into other Smoke E2E's implementation for reference.
    • The function receives SourceCommands generated by the CommandGenerator(see details in statefun-smoke-e2e-driver).
    • The function extracts Commands from the SourceCommand, and performs corresponding operations based on the command types:
      • Command.Send: send the enclosing Commands to the target address.
      • Command.SendAfter: send the enclosing Commands to the target address after a certain delay of time(defined as a constant, say, 1 millisecond).
      • Command.IncrementState: increment the counter(state) by 1.
      • Command.SendEgress: send a message to the DISCARD_EGRESS. The message can be just a simple string with whatever content.
      • Command.Verify: build a VerificationResult by filling in the expected count from the Command.Verify, and the actual count from function's counter(state). Then, send it over to the VERIFICATION_EGRESS.
    • Make sure the typename of the function, message, and ingress/egress are all aligned with the driver's definition. For example, the typename of VerificationResult should be statefun.smoke.e2e/verification-result. Think of the typename is the unique identifier of messages across different language bindings. You can find all the naming in Constants under statefun-smoke-e2e-driver.
  • Finally, wrap the CommandInterpreterFn as an HTTP endpoint using a simple web container. The endpoint should listen on 8000 port, aligning to the module.yaml setting.

Step 3: Orchestrate pieces into a SmokeVerificationE2E

  • Create a language specific SmokeVerificationE2E under src/test/java. Most of the code can just be copied from other Smoke E2E implementations. Particularly, you should focus on preparing resources for launching the language specific HTTP endpoint in the configureRemoteFunction method.
  • Next, create a Dockerfile.remote-function under src/test/resources, which builds the resources prepared by SmokeVerificationE2E into a Docker image. The image is then launched as an HTTP endpoint inside the container serving the CommandInterpreterFn.

Execution

Running the Harness test

  • The Harness test is to run the Smoke E2E test as a JUnit process against the remote function Java process. A nice feature it provides is you can run both the Harness and the remote function on an IDE, therefore allows you to set breakpoints on either one and debug the code efficiently.
  • How To:
    • Startup the HTTP endpoint that serve the remote function(implemented in Step 2) at port 8000, which aligned with the module.yaml defined in statefun-smoke-e2e-multilang-harness.
    • Comment out @Ignore in MultiLangSmokeHarnessTest.
    • Run MultiLangSmokeHarnessTest.
  • Noted that the Harness requires only Step 1 and Step 2 to be implemented, hence you can use the Harness to test out your implementation.

Running the SmokeVerificationE2E

  • The SmokeVerificationE2E, unlike Harness, spawn up the Flink cluster(master, workers) and the remote function as dedicated containers for testing. Such a setup is to simulate the production environment.
  • Once the Step 1,2,3 are implemented, you should be able to run the language specific SmokeVerificationE2E directly.

Runtime Architecture

Flink StateFun cluster -> SimpleVerificationServer
    (Containers)              (JUnit Process)
          ^
          |
          ˇ
   Remote Function
    (Container)

The SmokeRunner orchestrates the entire Smoke E2E runtime and does the following:

  • Launch the Flink StateFun cluster and the remote function, those are defined by StatefulFunctionsAppContainers.Builder and passed to the SmokeRunner as an input parameter. The detailed configuration can be found in concrete implementation such as SmokeVerificationGolangE2E
  • The CommandGenerator wrapped by CommandFlinkSource starts to generate commands and sent them over to the remote function for state manipulations(counters). At the same time it also applies the commands internally to its internal states.
  • After commands are generated, The CommandFlinkSource enters the verification stage and starts to send out Command.Verify messages along with the counts stored in its internal states as expected counts.
  • The remote function encloses the expected counts(from Command.Verify messages), and the actual counts(from remote function states) into VerificationResult messages.
  • The VerificationResult messages are then sent to SocketClientSink, and further sent to SimpleVerificationServer that collects the verification results.
  • The awaitVerificationSuccess method in SmokeRunner keeps polling the result arrived at SimpleVerificationServer and verifies it by simply doing actual == expected evaluation. When all the messages are successfully verified, the program exits.

Framework Modules

statefun-smoke-e2e-common

Testing utilities such asSmokeRunner, which is the facet class that organizes the Smoke E2E runtime architecture.

statefun-smoke-e2e-driver

The core logic of the Smoke E2E driver. The driver code here is built into a self-contained jar that can be loaded and ran by the Flink StateFun cluster directly.

statefun-smoke-e2e-multilang-base

This module contains a generic pom.xml that have the dependencies and the driver jar downloaded to run the Smoke E2E.

statefun-smoke-e2e-multilang-harness

The Harness test that run the Smoke E2E as a JUnit process. Noted that one should have a remote function running at localhost 8000 port for Harness test to interact with.