End-to-end tests verify the behavior of the entire system. For FSM, e2e tests will install a control plane, install test workloads and SMI policies, and check that the workload is behaving as expected.
FSM's e2e tests are located in tests/e2e
.
The tests are written using Ginkgo and Gomega so they may also be directly invoked using go test
. Be sure to build the fsm-controller
and init
container images and fsm
CLI before directly invoking the tests (see instructions below).
FSM's framework, helpers and related files are located under tests/framework
.
Once imported, it automatically sets up an init mechanism which will automatically initialize and parse flags and variables from both env
and go test flags
if any are passed to the test. The hooks for initialization and cleanup are set at Ginkgo's BeforeEach
at the top level of test execution (between Ginkgo Describes
); we henceforth recommend keeping every test in its own Describe
section, as well as on a separate file for clarity. You can refer to common.go for more details about the init, setup and cleanup processes.
Tests are organized by top-level Describe
blocks into buckets. Each bucket runs in parallel, and individual tests in the bucket run sequentially.
To help organize the tests, a custom Describe
block named FSMDescribe
is provided which accepts an additional struct parameter which contains fields for test metadata like tier and bucket. FSMDescribe
will construct a well-formatted name including the test metadata which can be used in CI to run tests accordingly. Ginkgo's original Describe
should not be used directly at the top-level and FSMDescribe
should be used instead.
Running the tests will require a running Kubernetes cluster. If you do not have a Kubernetes cluster to run the tests onto, you can choose to run them using Kind
, which will make the test framework initialize a cluster on a local accessible docker client.
Running the tests will also require the Helm CLI to be installed on your machine.
The tests can be run using the test-e2e
Makefile target at repository root level (which defaults to use Kind), or alternatively go test
targetting the test folder, which gives more flexibility but depends on related env
flags given or parsed by the test.
Please refer to the Kind cluster or Other K8s deployment and follow the instructions to setup potential env flags required by either option.
In addition to the flags provided by go test
and Ginkgo, there are several custom command line flags that may be used for e2e tests to configure global parameters like container image locations and cleanup behavior. You can see the list of flags under the flag section below.
The following make
target will create local containers for the FSM components, tagging them with CTR_TAG
, and will launch the tests using Kind cluster. A Kind cluster is created at test start, and requires a docker interface to be available on the host running the test.
When using Kind, we load the images onto the Kind nodes directly (as opposed to providing a registry to pull the images from).
CTR_TAG=not-latest make test-e2e
Note: If you use latest
tag, K8s will try to pull the image by default. If the images are not pushed to a registry accessible by the kind cluster, image pull errors will occur. Or, if an image with the same name is available, like flomesh/fsm-init:latest
, then that publicly available image will be pulled and started instead, which may not be as up-to-date as the local image already loaded onto the cluster.
Have your Kubeconfig file point to your testing cluster of choice.
The following code uses latest
tag by default. Non-Kind deployments do not push the images on the nodes, so make sure to set the registry accordingly.
export CTR_REGISTRY=<myacr>.dockerhub.io # if needed, set CTR_REGISTRY_USER and CTR_REGISTRY_PASSWORD
make build-fsm
make docker-build
go test ./tests/e2e -test.v -ginkgo.v -ginkgo.progress
Currently, test init will load a Kubeconf
based on default kubeconfig loading rules.
If Kind is used, the kubeconf is temporarily replaced and Kind's kubeconf is used instead.
A container registry where to load the images from (FSM, init container, etc.). Credentials are optional if the container registry allows pulling the images publicly:
-ctrRegistry string
Container registry
-ctrRegistrySecret string
Container registry secret
-ctrRegistryUser string
Container registry username
If container registry user and password are provided, the test framework will take care to add those as Docker secret credentials for the given container registry whenever appropriate (tenant namespaces for init
containers, FSM intallation, etc).
Container registry related flags can also be set through env:
export CTR_REGISTRY=<your_cr>.dockerhub.io
export CTR_REGISTRY_USER=<uername> # opt
export CTR_REGISTRY_PASSWORD=<password> # opt
The following flag will refer to the image version of the FSM platform containers (fsm-controller
and init
) and tcp-echo-server
for the tests to use:
-fsmImageTag string
FSM image tag (default "latest")
Make sure you have compiled the images and pushed them on your registry first if you are not using a kind cluster:
export CTR_REGISTRY=myacr.dockerhub.io
export CTR_TAG=mytag # Optional, 'latest' used by default
make docker-build-fsm # Add DOCKER_BUILDX_OUTPUT=type=docker when using kind
Worth mentioning cleanupTest
is especially useful for debugging or leaving the test in a certain state at test-exit.
When using Kind, you need to use cleanupKindCluster
and cleanupKindClusterBetweenTests
in conjunction, or else the cluster
will anyway be destroyed.
-cleanupTest
Cleanup test resources when done (default true)
-meshName string
FSM mesh name (default "fsm-system")
-waitForCleanup
Wait for effective deletion of resources (default true)
Plus, go test
and Ginkgo
specific flags, of course.
The ginkgo.focus
flag can be used to run individual tests. The flag should specify the "Context" of the test they wish to run, which can be found in the .go
file for that test. For instance, if you want to run the e2e_tcp_client_server_test
with SMI policies, you should run:
go test ./tests/e2e -test.v -ginkgo.v -ginkgo.progress -ginkgo.focus="\bSimpleClientServer TCP with SMI policies\b"
The installType
flag can be used to specify whether the tests should install FSM themselves, or if they will be run on a cluster which already has FSM installed.
go test ./tests/e2e -test.v -ginkgo.v -ginkgo.progress -installType=NoInstall
The different values of installType are as follows:
-
installType=NoInstall
The e2es will run on the cluster currently in the user's kubeconfig, which is already expected to have FSM installed prior to the test run. The tests will not install FSM before, or uninstall FSM after any of the tests run. The cluster can be of any CNCF certified distribution (though the k8s version cannot be an outdated version).
-
installType=KindCluster
Each test in the e2e suite will spin up a kind cluster, and run that specific test on that kind cluster. Each test will install FSM via the FSM cli, and then uninstall FSM after the test finishes running. The user is not expected to have created any cluster beforehand.
-
installType=SelfInstall
(default)By default, the e2es will run on the cluster currently in the user's kubeconfig. Each test will install FSM via the FSM cli, and then uninstall FSM after the test finishes running. The cluster can be of any CNCF certified distribution (though the k8s version cannot be an outdated version). The user is should not have FSM installed beforehand.
Testing implements support for Kind. If installType=KindCluster
is enabled, a new Kind cluster will be provisioned and it will be automatically used for the test.
-kindClusterName string
Name of the Kind cluster to be created (default "fsm-e2e")
-cleanupKindCluster
Cleanup kind cluster upon exit (default true)
-cleanupKindClusterBetweenTests
Cleanup kind cluster between tests (default true)
The test.timeout
flag sets a total time limit for all the tests that you are running. If you run the e2es without specifying any timeout limit, the tests will terminate after 10 minutes. To run the tests without any time limit, you should set test.timeout 0
.
To set a specific time limit, a unit must be specified along with a number. For instance, if you want to set the limit to 90 seconds (say for just testing one e2e), you should say test.timeout 90s
. If you want the tests to run for 60 minutes, you should say test.timeout 60m
.
OpenShift compatibility is still a WIP for the e2e tests.
To run these tests on OpenShift
- Install the oc CLI.
- Include
-deployOnOpenShift=true
with yourgo test
command.