Currently, tests are all CLI driven. That means that the commands executed in test implementations must be made
available in the target container/machine/shell's $PATH
. Future work will address incorporating REST-based tests.
UI-driven tests are considered out of scope for test-network-function
.
In general, tests should adhere to the following principles:
- Tests should be platform independent when possible, and platform-aware when not.
- Tests must be runnable in a variety of contexts (i.e.,
oc
,ssh
, andshell
). Internally, we have developed a variety ofinteractive.Context
implementations for each of these. In general, so long as your command does not depend on specific prompts, the framework handles the context transparently. - Tests must implement the
tnf.Tester
interface. - Tests must implement the
reel.Handler
interface. - Tests must be accompanied by appropriate unit tests.
- Tests adhere to the strict quality and style guidelines set forth in CONTRIBUTING.md.
Each tnf.Tester
implementation must have a unique identifier. In practice, tnf.Tester
implementations are the
building blocks of larger test suites, and each implementation ought to have a means of identification.
An identifier.Identifier
is the mechanism used to hold this meta information.
Please see the implementation for details. Essentially, an Identifier is just a URL and a Semantic Version.
To create an identifier for your test, go to identifiers.go
. Create a constant
for the URL, and add the TestCatalogEntry
to the Catalog
map such as:
listRootDirectoryFilesURL = "http://test-network-function.com/tests/listRootDirectoryFiles"
...
var Catalog = map[string]TestCatalogEntry{
...
listRootDirectoryFilesURL: {
URL: listRootDirectoryFilesURL,
Description: "A test to list the files at the root of the file system.",
Type: Normative,
}
...
}
Reference the exported URL constant in your tnf.Handler
GetIdentifier()
implementation.
Note: JSON tests should also involve creation of an identifier using the same Go-based methodology for 1.0
.
Identifiers can be reused, but they should follow the rules of semantic versioning. Namely, the following versioning should be utilized:
Version Level | Description |
---|---|
Major | API incompatible changes. |
Minor | Add functionality that is backwards compatible. |
Patch | Backwards compatible bug fixes. |
Note: If the premise of the test changes drastically, consider creating a new identifier instead of bumping the major version of an existing one.
There are two options for writing test implementations:
- JSON
- Go
The JSON approach is significantly quicker to implement, and should be preferred when possible.
Most tests just involve sending commands and validating output within a single shell context. For example, open an interactive shell to a container and ping a target host. On the command line, this might be done similar to the following:
oc exec -it <podName> -c <containerName> -- sh
ping -c <count> <destination>
We would expect the ping command to take approximately 5 seconds, as most implementations of ping
default to 1 second
for inter-packet gap. After the command completes, we would expect a summary to be output. Thus, the whole
interaction would be similar to the following:
% oc exec -it test -c test -- sh
sh-4.2# ping -c 5 www.redhat.com
PING e3396.dscx.akamaiedge.net (23.34.95.235) 56(84) bytes of data.
64 bytes from a23-34-95-235.deploy.static.akamaitechnologies.com (23.34.95.235): icmp_seq=1 ttl=61 time=16.1 ms
64 bytes from a23-34-95-235.deploy.static.akamaitechnologies.com (23.34.95.235): icmp_seq=2 ttl=61 time=22.8 ms
64 bytes from a23-34-95-235.deploy.static.akamaitechnologies.com (23.34.95.235): icmp_seq=3 ttl=61 time=24.5 ms
64 bytes from a23-34-95-235.deploy.static.akamaitechnologies.com (23.34.95.235): icmp_seq=4 ttl=61 time=23.6 ms
64 bytes from a23-34-95-235.deploy.static.akamaitechnologies.com (23.34.95.235): icmp_seq=5 ttl=61 time=18.3 ms
--- e3396.dscx.akamaiedge.net ping statistics ---
5 packets transmitted, 5 received, 0% packet loss, time 4007ms
rtt min/avg/max/mdev = 16.163/21.128/24.579/3.276 ms
sh-4.2#
We have now established the needed commands for a basic test. We have established that:
- The test should use an
oc
based context to connect to podtest
containertest
interactively. - The test should issue a
ping
command with a count of5
againstwww.redhat.com
. - After the
ping
command completes, we should inspect the summary to ensure that we received the expected number of packets.
Generic tests must abide by the generic-test.schema.json JSON Schema. Let's
consider the simple ping
example. It describes a test that pings "www.redhat.com" 5 times,
and gives a tnf.SUCCESS
only if all 5 pings receive a response.
Let's walk through ping.json
one key one at a time.
A human-readable description is required for every test. This documentation proves invaluable for later reuse.
testResult
is initialized as 0. As of now, tnf.Tester
mandates Result()
returns an int
value. 0
corresponds
to tnf.ERROR
. In fact, all tests should start report tnf.ERROR
as good practice, and only progress to
tnf.SUCCESS
or tnf.FAILURE
after inspecting a given match.
testTimeout
is the timeout for the test in nanoseconds. For this example, we chose a duration of 10s
to perform
the ping test.
reelFirstStep
is the first reel.Step
in the REEL
finite state machine (FSM). A reel.Step
contains a
mandatory timeout
and optional execute
and expect
. Note: Remember to append \n
to commands specified in
execute
. In this case, execute
is exactly what you might suspect; the ping command to www.redhat.com.
The expect
field deserves further explanation. expect
is an array of regular expressions that might match from
issuing a ping
command. In this case, only one regular expression is expected, the ping summary. However, if more
than one expect
element exists, the matches are determined in-order and the first match found will complete the step.
Development of regular expressions falls outside of this tutorial. regex101.com provides a useful interface to help design regular expressions for Go.
resultContexts
is self describing; if a regular expression in reelFirstStep.expect
is matched, then we will want to
provide further logic to progressively determine whether the test is a tnf.SUCCESS
or tnf.FAILURE
. Importantly,
these concepts are dependent on your own business logic. In this example, we require a response to every ICMP request
packet sent. A different implementation may be more lenient, and require only numRequests - 1
ICMP responses. The
implementation is completely up to the test writer.
Each reelFirst.expect
must have a resultContext
. In this case, we only have one pattern, which is represented by
the ICMP summary regular expression:
{
"pattern": "(?m)(\\d+) packets transmitted, (\\d+)( packets){0,1} received, (?:\\+(\\d+) errors)?.*$",
"defaultResult": 1,
"composedAssertions": [
{
"assertions": [
...
],
"logic": {
"type": "and"
}
}
]
}
defaultResult
is the test result returned if no composedAssertion
s exist. For example, if we omitted
composedAssertions
above, then the mere fact that the ping summary matched would result in tnf.SUCCESS
. However,
since composedAssertions
is provided, we must do further inspection.
In this case, only one composedAssertion
is provided. In case multiple composedAssertion
are provided, each composedAssertion
instances must evaluate
as true
. Here is the single composedAssertion
that we make in the test:
{
"assertions": [
{
"groupIdx": 1,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
},
{
"groupIdx": 2,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
}
],
"logic": {
"type": "and"
}
}
Essentially, a composedAssertion
allows us to make several sub-assertions using logic
. In this case, and
is used,
meaning each assertion
must evaluate as true.
There are two assertions in this example. The first assertion is as follows:
{
"groupIdx": 1,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
}
This means that group 1 of the regular expression match for the ping summary pattern must equal 5
. In this case,
group 1 is the number of ICMP requests transmitted.
The second assertion is as follows:
{
"groupIdx": 2,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
}
This means that group 2 of the regular expression match for the ping summary pattern must equal 5
. IN this case,
group 2 is the number of ICMP requests received.
The whole of the JSON pattern
and composedAssertions
means the test will pass ONLY if:
- exactly 5 pings were sent
- exactly 5 responses were received
Now that you have a sample JSON test defined, you can go ahead and run your JSON test in your development environment. In order to run the test, you must first make the jsontest CLI. Issue the following command:
./tnf jsontest shell examples/ping.json
You will get something similar to the following:
% ./tnf jsontest shell examples/ping.json
INFO[0000] Running examples/ping.json from a local shell context
2020/12/06 13:32:53 Sent: "ping -c 5 www.redhat.com\n"
2020/12/06 13:32:57 Match for RE: "(?m)(\\d+) packets transmitted, (\\d+)( packets){0,1} received, (?:\\+(\\d+) errors)?.*$" found: ["5 packets transmitted, 5 packets received, 0.0% packet loss" "5" "5" " packets" ""] Buffer: "PING e3396.dscx.akamaiedge.net (23.34.95.235): 56 data bytes\n64 bytes from 23.34.95.235: icmp_seq=0 ttl=59 time=17.661 ms\n64 bytes from 23.34.95.235: icmp_seq=1 ttl=59 time=25.993 ms\n64 bytes from 23.34.95.235: icmp_seq=2 ttl=59 time=26.353 ms\n64 bytes from 23.34.95.235: icmp_seq=3 ttl=59 time=25.725 ms\n64 bytes from 23.34.95.235: icmp_seq=4 ttl=59 time=22.403 ms\n\n--- e3396.dscx.akamaiedge.net ping statistics ---\n5 packets transmitted, 5 packets received, 0.0% packet loss\nround-trip min/avg/max/stddev = 17.661/23.627/26.353/3.302 ms\n"
INFO[0004] Test Result: 1
INFO[0004] Test Payload:
INFO[0004] {
"description": "Pings www.redhat.com 5 times using the Unix ping executable.",
"matches": [
{
"pattern": "(?m)(\\d+) packets transmitted, (\\d+)( packets){0,1} received, (?:\\+(\\d+) errors)?.*$",
"before": "PING e3396.dscx.akamaiedge.net (23.34.95.235): 56 data bytes\n64 bytes from 23.34.95.235: icmp_seq=0 ttl=59 time=17.661 ms\n64 bytes from 23.34.95.235: icmp_seq=1 ttl=59 time=25.993 ms\n64 bytes from 23.34.95.235: icmp_seq=2 ttl=59 time=26.353 ms\n64 bytes from 23.34.95.235: icmp_seq=3 ttl=59 time=25.725 ms\n64 bytes from 23.34.95.235: icmp_seq=4 ttl=59 time=22.403 ms\n\n--- e3396.dscx.akamaiedge.net ping statistics ---",
"match": "5 packets transmitted, 5 packets received, 0.0% packet loss"
}
],
"reelFirstStep": {
"execute": "ping -c 5 www.redhat.com\n",
"expect": [
"(?m)(\\d+) packets transmitted, (\\d+)( packets){0,1} received, (?:\\+(\\d+) errors)?.*$"
],
"timeout": 10000000000
},
"resultContexts": [
{
"pattern": "(?m)(\\d+) packets transmitted, (\\d+)( packets){0,1} received, (?:\\+(\\d+) errors)?.*$",
"composedAssertions": [
{
"assertions": [
{
"groupIdx": 1,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
},
{
"groupIdx": 2,
"condition": {
"type": "intComparison",
"input": 5,
"comparison": "=="
}
}
],
"logic": {
"type": "and"
}
}
],
"defaultResult": 1
}
],
"testResult": 1,
"testTimeout": 10000000000
}
Note that testResult
is 1, indicating tnf.SUCCESS
.
If you wish to explore the oc
and ssh
variants of jsontest-cli
, please consult the following:
./jsontest -h
See the diagnostic test suite for an example of this.
Often times, tests require arguments. For example, if you were to write a test which involves testing ping
to a
particular destination, perhaps derived dynamically, the destination would need to be configurable. In this case,
Go templates can be used to render a JSON-based test. ping.json.tpl is an
example of a JSON-test which contains a HOST
argument, and
ping.values.yaml provides the necessary values. Tests can be rendered
using something similar to:
templateFile := path.Join("examples", "generic", "template", "ping.json.tpl")
schemaPath := path.Join("schemas", "generic-test.schema.json")
valuesFile := path.Join("examples", "generic", "template", "ping.values.yaml")
tester, handlers, result, err := generic.NewGenericFromTemplate(templateFile, schemaPath, valuesFile)
A test-network-function
test must implement tnf.Tester
and reel.Handler
Go interface
s. The tnf.Tester
interface defines the contract required for a CLI-based test, and reel.Handler
defines the Finite State Machine (FSM)
contract for executing the test. A basic example is ping.go.
We will go through implementing the required interfaces one at a time below:
For a test to implement the tnf.Tester
interface, it must provide definitions for Args
, Timeout
and Result
.
These are the set of accessor methods used to define characteristics of the test, as well as the actual result of the
test. Note, this does not include any expected results; those need to be defined later.
First create a type called Ping
which is capable of storing result
, timeout
and args
variables. Additionally,
we will restrict our test by mandating that a positive integer count
and destination
string must be provided during
the time of instantiation.
// Ping provides a ping test implemented using command line tool `ping`.
type Ping struct {
result int
timeout time.Duration
args []string
count int
destination string
}
To better enforce data encapsulation, please only export (capitalize) variables that are absolutely needed. For
example, we use count
not Count
above.
If you look at the tnf.Tester
interface definition, you will notice that the data types for
result
, timeout
and args
match the return types for the mandated functions.
type Tester interface {
Args() []string
Timeout() time.Duration
Result() int
}
After creating the struct, define the accessors similar to the following:
// Args returns the command line args for the test.
func (p *Ping) Args() []string {
return []string{"ping", "-c", p.count, p.destination}
}
// Timeout returns the timeout in seconds for the test.
func (p *Ping) Timeout() time.Duration {
return p.timeout
}
// Result returns the test result.
func (p *Ping) Result() int {
return p.result
}
The Args()
implementation deserves some explaining. Args()
is an array of the commands line argument strings. In
this example, the command ping -c 5 www.redhat.com
is represented as string[]{"ping", "-c", "5", "www.redhat.com"}
.
In other words, the elements are all of the white-space separated string components of the command.
That completes our ping.go
tnf.Tester
implementation! Next, implement the logic of the reel.Handler
FSM.
The easy part is out of the way. Implementing reel.Handler
is slightly more involved, but should make sense after
completing this part of the tutorial. reel.go defines the reel.Handler
interface:
// A Handler implements desired programmatic control.
type Handler interface {
// ReelFirst returns the first step to perform.
ReelFirst() *Step
// ReelMatch informs of a match event, returning the next step to perform. ReelMatch takes three arguments:
// `pattern` represents the regular expression pattern which was matched.
// `before` contains all output preceding `match`.
// `match` is the text matched by `pattern`.
ReelMatch(pattern string, before string, match string) *Step
// ReelTimeout informs of a timeout event, returning the next step to perform.
ReelTimeout() *Step
// ReelEOF informs of the eof event.
ReelEOF()
}
We will handle describing implementing each of these methods one by one.
ReelEOF
is used to define the callback executed when EOF is encountered in the context. Unexpected interruptions to
ssh
or oc
session are common reasons for EOF.
For the case of ping we can make this simple. Since we require a count
for ping
, we don't need to do anything
particular for EOF.
// ReelEOF does nothing; ping requires no intervention on EOF.
func (p *Ping) ReelEOF() {
}
ReelTimeout
is used to define the callback executed when a test times out.
When a ping test times out, we probably ought to issue a CTRL+C
in order to exit early and prepare the context for
future commands.
// ReelTimeout returns a step which kills the ping test by sending it ^C.
func (p *Ping) ReelTimeout() *reel.Step {
return &reel.Step{Execute: "\003"}
}
Since we supply tnf.Test
Args()
, we do not need to include anything for Execute
in the returned reel.Step
.
// ReelFirst returns a step which expects the ping statistics within the test timeout.
func (p *Ping) ReelFirst() *reel.Step {
return &reel.Step{
Expect: []string{`(?m)connect: Invalid argument$`, `(?m)(\d+) packets transmitted, (\d+)( packets){0,1} received, (?:\+(\d+) errors)?.*$`},
Timeout: p.timeout,
}
}
Note: The ordering of Expect
matters! The framework matches Expect
elements in index-ascending order.
This is likely the hardest part of any test implementation. ReelMatch
needs to decipher what is matched, and assign
the appropriate result to the tnf.Test
. Let's take a look at the implementation provided for
ping.go:
// ReelMatch parses the ping statistics and set the test result on match.
// The result is success if at least one response was received and the number of
// responses received is at most one less than the number received (the "missing"
// response may be in flight).
// The result is error if ping reported a protocol error (e.g. destination host
// unreachable), no requests were sent or there was some test execution error.
// Otherwise the result is failure.
// Returns no step; the test is complete.
func (p *Ping) ReelMatch(_ string, _ string, match string) *reel.Step {
re := regexp.MustCompile(`(?m)connect: Invalid argument$`)
matched := re.FindStringSubmatch(match)
if matched != nil {
p.result = tnf.ERROR
}
re = regexp.MustCompile(SuccessfulOutputRegex)
matched = re.FindStringSubmatch(match)
if matched != nil {
// Ignore errors in converting matches to decimal integers.
// Regular expression `stat` is required to underwrite this assumption.
p.transmitted, _ = strconv.Atoi(matched[1])
p.received, _ = strconv.Atoi(matched[2])
p.errors, _ = strconv.Atoi(matched[4])
switch {
case p.transmitted == 0 || p.errors > 0:
p.result = tnf.ERROR
case p.received > 0 && (p.transmitted-p.received) <= 1:
p.result = tnf.SUCCESS
default:
p.result = tnf.FAILURE
}
}
return nil
}
Essentially, since ReelMatch()
always returns nil
this function is the final state for the reel.Step
FSM. For
more advanced tests, ReelMatch()
can be called an arbitrary number of times. In this example, ReelMatch()
is only
called once.
The logic for determining the test result is up to the test writer. This particular implementation analyzes the match output to determine the result.
- If the provided
destination
results in anInvalid Argument
, thentnf.ERROR
is returned. - If the ping summary regular expression matched, then:
tnf.ERROR
if there were PING transmit errorstnf.SUCCESS
if a maximum of a single packet was losttnf.FAILURE
for any other case.
An example of using ping.go
from within a Ginkgo test spec is included in
suite.go's testPing
method. Roughly, the code should resemble the
following:
// 1. Create the Test.
pingTester := ping.NewPing(defaultTimeout, targetPodIPAddress, count)
test, err := tnf.NewTest(oc.GetExpecter(), pingTester, []reel.Handler{pingTester}, oc.GetErrorChannel())
gomega.Expect(err).To(gomega.BeNil())
// 2. Run the Test.
testResult, err := test.Run()
gomega.Expect(testResult).To(gomega.Equal(tnf.SUCCESS))
gomega.Expect(err).To(gomega.BeNil())
// 3. Inspect the Results.
transmitted, received, errors := pingTester.GetStats()
gomega.Expect(received).To(gomega.Equal(transmitted))
gomega.Expect(errors).To(gomega.BeZero())
You should now have the appropriate knowledge to write your own test implementation. There are a variety of implementations included out of the box in the handlers directory.
This guide does not cover unit testing the Test, nor does it cover managing test-specific configuration. Please see the examples of existing tests in the codebase for how to do these things.
Although test-network-function
includes built in interactive.Context
implementations for oc
, shell
and ssh
,
there are many cases in which you may need a completely new PTY. For example, networking software (including vpp, Cisco
IOS, etc.) often includes interactive PTY-based menus. In such a case, you will need a new interactive.Context
to
communicate with the underlying Shell.
In such cases, consider using interactive.SpawnGenericPTYFromYAMLFile(...)
or its corollary
interactive.SpawnGenericPTYFromYAMLTemplate(...)
which can be templated using Go text/template
language. Examples
of such PTY implementations can be found in examples/pty.
The current tests frequently use jq
to process structured output from oc -o json
. oc
also allows use of
Go Templates for processing structured output.
This is potentially more powerful than using jq
as it allows building highly customized output of multiple resources
simultaneously without adding dependencies. Conversely jq
is widely available and commonly used, and has been
sufficient for all cases so far. It is up to the author of a contribution to decide which approach is best suited to the
task at hand.
While the likely use of Go Templates is at the complex end of the spectrum, as a simple example the command used in CONTAINER_COUNT
to find the number of containers in a pod is currently using jq
:
oc get pod %s -n %s -o json | jq -r '.spec.containers | length'
The same result could be achieved using a Go Template:
oc get pod %s -n %s -o go-template='{{len .spec.containers}}{{"\n"}}'
To facilitate adding new handlers, the "tnf" utility has been created to help developers to avoid writing repetitive code. The tnf tool source code is here and can be built with the following command:
make build-tnf-tool
To generate a new handler named MyHandler, use the options "generate handler" as in the next example:
./tnf generate handler MyHandler
The generated code has a template and creates the necessary headers. The result is folder "myhandler" located in /pkg/tnf/handlers/myhandler that includes 3 files by handler template. The command relays on golang templates located in pkg/tnf/handlers/handler_template, so in case the "tnf" utility is executed outside the test-network-function root folder, the user can export the environment variable TNF_HANDLERS_SRC pointing to an existing "handlers" relative/absolute folder path.
export TNF_HANDLERS_SRC=other/path/pkg/tnf/handlers
The result of each test execution is included in the claim file.
Sometimes it is convenient to add informational messages regarding the test execution.
In order to add informational messages to your test use the function ginkgo.GinkgoWriter
.
This function adds an additional message that will appear in the CapturedTestOutput
section of the claim file, together with the output of the by directives.
Each added message will be written to claim file even if test failed or error occurred in the middle of the test.
Example usage:
ginkgo.It("Should do what I tell it to do", func(){
// do some more work
// add info
_, err := ginkgo.GinkgoWriter.Write([]byte("important info part 1"))
if err != nil {
log.Errorf("Ginkgo writer could not write because: %s", err)
}
// more work
// more info
_, err := ginkgo.GinkgoWriter.Write([]byte("important info part 2"))
if err != nil {
log.Errorf("Ginkgo writer could not write because: %s", err)
}
// error
if err != nil {
return
}
// last info
_, err := ginkgo.GinkgoWriter.Write([]byte("important info part last"))
if err != nil {
log.Errorf("Ginkgo writer could not write because: %s", err)
}
})