README ¶
Integration Tests
This is the home of the system integrations tests.
Running the tests
They can be run from the root of vega with:
make integrationtest
or
go test ./...
Running just the integration tests
The integration tests have been hooked up to run as regular unit tests, so you can run just the integration tests with a simple command:
go test ./integration/...
When running these tests, you'll probably want to get a more verbose output (showing which steps of the tests passed and failed), which can be done by adding 2 flags:
go test -v ./integration/... --godog.format=pretty
The -v
flag tells go test
to run with verbose output (sending logging to stdout). The -godog.format=pretty
flag (
which must be put at the end) instructs godog to print out the scenario's and, in case an assertion fails, show which
particular step of a given scenario didn't work.
Running specific scenario's
To run only certain tests (feature files), you can simply add the paths to a given feature file to the command:
go test -v ./integration/... --godog.format=pretty $(pwd)/integration/features/my-feature.feature
Race detection and cache
For performance reasons, go test
will check whether the source of a package has changed, and reuse compiled objects or
even test results in case it determines nothing has changed. Because the integration tests are tucked away in their own
package, changes to other packages might not be compiled, and tests could possibly pass without changes being applied.
To ensure no cached results are used, the -count
flag can be used:
go test -v -count=1 ./integration/... --godog.format=pretty
Should there be tests that are intermittently failing, this could indicate a data race somewhere in the code. To use the
race detector to check for this, you can add the -race
flag to the command. The full commands then would be:
# Run all integration tests, verbose mode, ensure recompiled binaries, enable race detection, and use godog pretty formatting
go test -v -count=1 -race ./integration/... --godog.format=pretty
# Same as above, but only run a specific feature file:
go test -v -count=1 -race ./integration/... --godog.format=pretty $(pwd)/integration/feature/my-feature.feature
Race detection is a complex thing to do, so it will make running tests significantly slower. The pipeline runs the tests with race detection, so this shouldn't be required to do locally.
Reproducing/replicating system tests
The system tests run on a higher level. They submit a new market proposal, get said market accepted through governance,
and then start trading. They use a LogNormal
risk model, and specific fee parameters. David kindly provided the
long/short risk factors for a simple risk model that result in the same margin requirements and same fees being applied
to the trades. To create an integration test that replicates the system test results (transfers, balances, fees, etc...)
, simply start your feature file with the following:
Feature: A feature that reproduces some system test
Background:
Given the markets:
| id | quote name | asset | risk model | margin calculator | auction duration | price monitoring | data source config |
| ETH/DEC20 | ETH | ETH | default-simple-risk-model | default-margin-calculator | 1 | default-none | default-for-future |
And the initial insurance pool balance is "0" for all the markets
Debug problems with VSCode
You might have a situation where you need to closely investigate the state of a tested application or the tests script itself. You might want to set some breakpoints, at which test execution stops, and you can view that state.
This process is called debugging, and we have an initial/template configuration for you. Please follow these steps:
- Open and edit .vscode/launch.json config file (do not commit changes to this file):
- to run one
.feature
file, then editDebug .feature test
section and point to your.feature
file, - to run single Scenario in a
.feature
file, then change.feature
file path and specify the line of a Scenario in that file. All in sectionDebug single Scenario in .feature test
,
- Set breakpoints in tests code or application code (note: it has to be in
.go
files, no.feature
), - In VSCode switch to
Run and Debug
view (an icon with a bug from left side-bar), - From top drop-down select which option you want to run, e.g.
Debug .feature test
, - Click green Play button,
- Observe results in
Debug Console
tab at bottom.
Life cycle
To get a market up and running, here is the process:
- Configuration of network parameters. They have default values so it's not required, but if we want to override them, it should be done in the first step.
- Configuration of market.
- Declaration of the traders and their general account balance.
- Placement of orders by the traders, so the market can have a mark price.
Once these steps are done, the market should be in a proper state.
Steps
The list of steps is located in ./main_test.go
.
Market instantiation
Setting up a market is complex and the base for everything. As a result, we created a "lego-like" system to help us strike the balance between flexibility and re-usability.
Flexibility with steps
A market is composed of several sets of parameters grouped by domain, such as margin, risk model, fees, and so on.
Each set of parameters is declared in its own step into which a custom name is given. In our "lego" analogy, these named sets would be the "blocks".
To declare a market, we tell to our market which "blocks" to use.
Here is an example where we declare a risk model named "simple-risk-model-1". Then, we declare a "BTC" market, to which we associate the risk model "simple-risk-model-1".
Given the simple risk model named "simple-risk-model-1":
| long | short | max move up | min move down | probability of trading |
| 0.1 | 0.1 | 10 | -10 | 0.1 |
And the markets:
| id | quote name | asset | risk model |
| ETH/DEC21 | BTC | BTC | simple-risk-model-1 |
Re-usability with defaults
Because markets are tedious to instantiate, most of the time, we instantiate them using defaults stored in JSON files
inside the folder steps/market/defaults
.
Each sub-folders contain the defaults for their domain. Referencing a default for the price monitoring that is not in
the price-monitoring
folder will result in failure.
Using defaults works just like the named set, except that the file name will be used as the name. As a result, if the
file containing the defaults is named default-basic.json
, then the name to fill in will be default-basic
.
This is the recommended way. It's also fine to introduce a new defaults as long as it's used more than a couple of times.
You can mix the use of steps and defaults in market declaration.
Debug
Sometimes, you need to log some state. For this, you can use the debug ...
steps.
Convention
Glossary
We should move toward building our ubiquitous language and use domain language and avoid the use of synonyms.
If we talk about submitting an order
, we avoid using placing an order.
Structuring a feature test
File
A feature test's file should be named by the feature / command it's testing, such
has: maintenance_call_for_margin_account.go
.
The file name should match, or at least be close to, the description of the Feature
keyword.
To be avoided:
- A prefix with a pull-request or an issue number, such as
4284-cancel-order.go
. - A vague name, or context, such as
orders.go
orcancellation.go
Feature
The Feature
keyword should describe the feature to be tested in proper sentences, with context.
It should match, or at least be close to, the name of the file it lives in.
Let's assume, we have a file called trader_cancels_orders.go
Feature: Traders can cancel his orders under certain conditions
By reading this, we get to know who's the main actor, the action and the target. Saying "Under certain conditions" is
vague, but it's enough as that's the purpose of the Scenario
to be more specific. At least, we know there are
conditions.
Feature: cancel orders
This is too vague.
Feature: Should monitor prices
This seems completely unrelated to what the file name mentions.
Scenario
The Scenario
keyword should describe the tested case of the command, and, thus, should never be empty.
A file can contain multiple scenarios if they test the same feature. Unrelated tests should live in a dedicated file.
If the feature to be tested is order cancellation, we could have:
Feature: Trader can cancel orders under certain condition
Scenario: Trader can cancel his order if not matched
...
Scenario: Trader cannot cancel another trader's order
...
We know who is doing what on what.
Scenario: Works
...
Scenario: fail !
...
Oh yeah ?
Scenario:
...
Okay...
Given
Given
should only be used for prerequisite declaration. Arguably, it's a bit tricky to distinguish a prerequisite from
what's not. For now, as a rule of thumb, we consider market declaration and traders initial deposit to the general
account to be the pre-requisites. Other steps should use the keywords below.
Given the market ...
And the traders general account balance ...
When
When
should only be used when issuing a command. It shouldn't be used for assertions. The preferred construct of
command steps is:
<actor> <action verb> <target>
Construction with passive voice is accepted, if it makes more sense than the active voice.
When traders submit the following orders
We know who does what.
When an oracle data is submitted
The passive voice sounds better The system receives the following oracle data
.
Then
Then
should only be used when asserting a state. It shouldn't be use for commands. The preferred construct of
assertion steps is:
<actor> should <state verb> <target>
Then trader trader-1 should have a balance of 100 ETH
We know what we expect from whom.
Then trader trader-1 have a balance of 100 ETH
We miss the should
that emphasize the expectation.
Then the orders should fails
This is too vague.
Then the trader places an order
It's a command. The keywords should be used to help the reader to distinguish a command from an assertion. Even if the
above sentence makes sense, it breaks the structure When command > Then assertion
, and we might end up with a list
of Then
:
Then the trader places an order
Then the trader should have balance ...
Then an oracle data is sent
Then the settlement data should be updated
We are no longer be able to sort out the commands from the assertions at first glance.
And / But
And
can be used by any of the previous keywords and should follow the sentence construction of the keyword it is
backing. Use But
for negative outcomes.
Step
Text
- The first word should start we a lower-case letter.
- Words (and table columns) should be lower-case with space separation, like plain human style. No upper-case location to be remembered.
- Acronyms should be lower-case, like the rest, without trailing dot. We want to avoid interrogation such as :
ID
orId
orId.
?
When the market id should contain the asset "..."
All lower-case.
Then The Market Id should appears in U.R.L with QuoteName
Single-line step
Error
We should verify the error message on every expected failure using because
followed by the error message.
Then the order "1234" should be rejected because "....."
We ensure the error is the expected one, and the context is clear, no need for additional comments.
Then the order "1234" should be rejected
It may have not failed for the reason we expected. And, we may be tempted to add a comment to explain the reason of the failure.
Table step
Error
The column to verify the error should always be named error
.
Date
Prefer expiration date
over expires at
or started at
.