This document provides descriptions of Gerrit end-to-end (e2e
) test scenarios implemented using
the Gatling framework.
Similar scenarios have been successfully used to compare performance of different Gerrit versions or study the Gerrit response under different load profiles. Although mostly for load, scenarios can either be for load or functional (e2e) testing purposes. Functional scenarios may then reuse this framework and Gatling’s usability features such as its protocols (more below) and DSL.
That cross test-scope reusability applies to both Gerrit core scenarios and non-core ones, such as
for Gerrit plugins or other potential extensions. End-to-end testing may then include scopes like
feature integration, deployment, smoke (and load) testing. These load and functional test scopes
should remain orthogonal to the unit and component (aka Gerrit IT
-suffixed or acceptance
) ones.
The term acceptance
though may still be coined by organizations to target e2e functional testing.
What is Gatling?
Gatling is mostly a load testing tool which provides out of the box support for the HTTP protocol. Documentation on how to write an HTTP load test can be found here. However, in the scenarios that were initially proposed, the Gatling Git extension was leveraged to run tests at the Git protocol level.
Gatling is written in Scala, but the abstraction provided by the Gatling DSL makes the scenarios
implementation easy even without any Scala knowledge. The online End-to-end tests
presentation links
posted on the homepage have more introductory information.
IDE: IntelliJ
Examples of scenarios can be found in the e2e-tests
directory. The files in that directory should
be formatted using the mainstream
Scala plugin for IntelliJ. The latter is not
mandatory but preferred for sbt
and Scala IDE purposes in this project. So, Eclipse can also be
used alongside as a development IDE; this is described below.
Eclipse
-
Install the Scala plugin for Eclipse.
-
Run
sbt eclipse
from thee2e-tests
root directory. -
Import the resulting
e2e-tests
eclipse file inside the Gerrit project, in Eclipse. -
You should see errors in Eclipse telling you there are missing packages.
-
This is due to the sbt-eclipse plugin not properly linking the Gerrit Gatling e2e tests with Gatling Git plugin.
-
You then have to right-click on the root directory and choose the build path→link source option.
-
Then you have to browse to
.sbt/1.0/staging
, find the folder where gatling-git is contained, and choose that. -
That last step should link the gatling-git plugin to the project; e2e tests should not show errors anymore.
-
You may get errors in the gatling-git directory; these should not affect Gerrit Gatling development and can be ignored.
How to build the tests
An sbt-based installation of Scala is required.
The scalaVersion
used by sbt
once installed is defined in the build.sbt
file. That specific
version of Scala is automatically used by sbt
while building:
sbt compile
The following warning, if present when executing sbt
commands, can be removed by creating the
related credentials file
locally. Dummy values for user
and password
in that file can be used initially.
[warn] Credentials file ~/.sbt/sonatype_credentials does not exist
The other warning below can be safely ignored, so far. Running the proposed sbt evicted
command
should only list scala-java8-compat_2.12
as [warn]
. The other dependency conflicts should show
as [info]
. All of the listed conflicts get usually resolved seamlessly or so.
[warn] There may be incompatibilities among your library dependencies; run 'evicted' to see detailed eviction warnings.
Every sbt
command can include an optional log level
argument.
Below, [info]
logs are no longer shown:
sbt --warn compile
How to build using Docker
docker build . -t e2e-tests
How to set-up
SSH keys
If you are running SSH commands, the private keys of the users used for testing need to go in
/tmp/ssh-keys
. The keys need to be generated this way (JSch won’t validate them
otherwise):
mkdir /tmp/ssh-keys ssh-keygen -m PEM -t rsa -C "test@mail.com" -f /tmp/ssh-keys/id_rsa
The public key in /tmp/ssh-keys/id_rsa.pub
has to be added to the test user(s) SSH Keys
in
Gerrit. Now, the host from which the latter runs may need public key scanning to become known.
This applies to the local user that runs the forthcoming sbt
testing commands. An example
assuming localhost
follows:
ssh-keyscan -t rsa -p 29418 localhost > ~/.ssh/known_hosts
Input file
The CloneUsingBothProtocols
scenario is fed with the data coming from the
src/test/resources/data/com/google/gerrit/scenarios/CloneUsingBothProtocols.json
file. Such a
file contains the commands and repository used during the e2e test. That file currently looks like
below. This scenario serves as a simple example with no actual load in it. It can be used to test
or validate the local setup. More complex scenarios can be further developed, under the
com.google.gerrit.scenarios
package. The uppercase keywords are set through
environment properties.
[ { "url": "ssh://admin@HOSTNAME:SSH_PORT/_PROJECT", "cmd": "clone" }, { "url": "HTTP_SCHEME://HOSTNAME:HTTP_PORT/_PROJECT", "cmd": "clone" } ]
Valid commands are:
-
clone
-
fetch
-
pull
-
push
Project and HTTP credentials
The example above assumes that the loadtest-repo
project exists in the Gerrit under test. The
CloneUsingBothProtocols
scenario already includes creating that project and deleting it once done
with it. That scenario class can be used as an example of how a scenario can compose itself
alongside other scenarios (here, CreateProject
and DeleteProject
).
The HTTP Credentials
or password obtained from test user’s Settings
(in Gerrit) may be
required, in src/test/resources/application.conf
, depending on the above commands used. That
file’s http
section shows which shell environment variables can be used to set those credentials.
Executing the CloneUsingBothProtocols
scenario, as is, does require setting the http credentials.
That is because of the aforementioned create/delete project (http) scenarios composed within it.
Environment properties
The JAVA_OPTS
environment variable
can optionally be used to define
non-default values for keys found in scenario json
data files. That variable can currently be set
with either one or many of these supported properties, from the core framework:
-
-Dcom.google.gerrit.scenarios.hostname=localhost
-
-Dcom.google.gerrit.scenarios.ssh_port=29418
-
-Dcom.google.gerrit.scenarios.http_port=8080
-
-Dcom.google.gerrit.scenarios.http_scheme=http
Above, the properties can be set with values matching specific deployment topologies under test.
The name of the property corresponds to the uppercase keyword found in the json file. For example,
hostname
above will set the value of HOSTNAME
in the aforementioned example.
The example values shown above are the currently coded default ones. For example, the http
scheme
above could be replaced with https
. The framework may support differing or more properties over time.
Replication delay
The replication_delay
property allows test scenario steps to wait for that many seconds, prior to
expecting a done replication. Its default is 15
seconds and can be set using another value:
-
-Dcom.google.gerrit.scenarios.replication_delay=15
There is a short time buffer added to this property. Now, the replication starts after replication
plugin’s own replicationDelay
, in seconds, and typically takes some more seconds to complete.
That whole replication time depends on the system under test. Therefore, this property here should
be set to a value high enough, so that the test checks for a done replication at the right time.
Automatic properties
The example keywords also include _PROJECT
, prefixed with an underscore, which
means that its value gets automatically generated by the scenario. Any property setting for it is
therefore not applicable. Its usage differs from the non-prefixed PROJECT
keyword, in that sense.
Using the latter instead in json files requires setting this JAVA_OPTS
property:
-
-Dcom.google.gerrit.scenarios.project=myOwnTestRepoProjectName
Other automatic keys may be used and implemented, always prefixed with an underscore that tells so.
Plugin scenarios
Plugin or otherwise non-core scenarios can also use such properties. The core java package
com.google.gerrit.scenarios
from the example above has to be replaced with the one under which
those scenario classes are. Such extending scenarios can also add extension-specific properties.
Examples of this can be found in these Gerrit plugins test code:
Power factor
The following core property can be optionally set depending on the runtime environment. The test
environments used as reference for scenarios development assume its default value, 1.0
. For
slower or more complex execution environments, the value can be increased this way for example:
-
-Dcom.google.gerrit.scenarios.power_factor=1.5
This will make the scenario steps take half more time to expect proper completion. A value smaller
than the default, say 0.8
, will make scenarios wait somewhat less than how they were developed.
Scenario development is often done using locally running Gerrit systems under test, which are
sometimes dockerized.
Number of users
The number_of_users
property can be used to scale scenario steps to run with the specified number
of concurrent users. The value of this property remains 1
by default. For example, this sets the
number of concurrent users to 10:
-
-Dcom.google.gerrit.scenarios.number_of_users=10
This will make scenarios that support the number_of_users
property to inject that many users
concurrently for load testing.
How to run tests
Run all tests:
sbt "gatling:test"
Run a single test:
sbt "gatling:testOnly com.google.gerrit.scenarios.CloneUsingBothProtocols"
Generate the last report:
sbt "gatling:lastReport"
The src/test/resources/logback.xml
file
configures Gatling’s logging level. To quickly
enable detailed logging of http
requests and responses, the root level
can be set to trace
in that file.
How to run using Docker
docker run -it e2e-tests -s com.google.gerrit.scenarios.CloneUsingBothProtocols
How to run non-core scenarios
Locally adding non-core scenarios, for example from Gerrit plugins, is as simple as copying such files in. Copying is necessary over linking, unless running using Docker (above) is not required. Docker does not support links for files it has to copy over through the Dockerfile (here, the scenario files). Here is how to proceed for adding such external (e.g., plugin) scenario files in:
pushd e2e-tests/src/test/scala cp -r (or, ln -s) scalaPackageStructure . popd pushd e2e-tests/src/test/resources/data cp -r (or, ln -s) jsonFilesPackageStructure . popd
The destination folders above readily git-ignore every non-core scenario file added under them. If
running using Docker, e2e-tests/Dockerfile
may require another COPY
line for the hereby added
scenarios. Aforementioned sbt
or docker
commands can then be used to run the added tests.
Part of Gerrit Code Review
[scala]: