This repository contains a development branch of the second major version of Variant.
See https://github.com/mumoshu/variant for more information on the first version.
Once finished, this repository will eventually take over the
master
branch of the original variant repository.
- HCL-based DSL: Terraform-like strongly-typed DSL on top of HCL to define your command. See
Configuration Language
below. - Concurrency and Workflow: Embedded workflow engine with concurrency. See
Concurrency
below. Example: concurrency - Auto-prompt: Variant prompts for missing arguments to your command so that your user is only needed to provide missing arguments on the fly, instead of rerunning the whole command by repeating every argument.
- Configs: Deep-merging YAML configuration files. Example: config
- Secrets: Deep-merging secret values from Vault, AWS SecretsManager, SOPS, etc. powered by vals. Example: secret
- Testing: Test framework with
go test
-compatible test runner. Example: simple - Embeddable: Easy embedding in any Golang application
- Easy distribution: Build a single-executable of your command with Golang
- Dependency Management: Dependent files, executable binaries and docker-run shims can be automatically installed and updated with the variantdev/mod integration. Example: module
- Run as a Kubernetes controller: You can easily turn your Variant command into a Kubernetes controller. See examples/controller
- Integrations: Integrates nicely with Slack, GitHub. You can run Variant command in response to Slack message, GitHub issue comment, commit push, etc.
Create an variant
file that contains the following:
examples/getting-started/getting-started.variant
:
option "namespace" {
description = "Namespace to interact with"
type = string
default = "default"
short = "n"
}
job "kubectl" {
parameter "dir" {
type = string
}
exec {
command = "kubectl"
args = ["-n", opt.namespace, "-f", param.dir]
}
}
job "helm" {
parameter "release" {
type = string
}
parameter "chart" {
type = string
}
option "values" {
type = list(string)
}
exec {
command = "helm"
args = ["upgrade", "--install", "-n", opt.namespace, param.release, param.chart]
}
}
job "deploy" {
description = "Deploys our application and the infrastructure onto the K8s cluster"
step "deploy infra" {
run "helm" {
release = "app1"
chart = "app1"
}
}
step "deploy apps" {
run "kubectl" {
dir = "deploy/environments/${opt.env}/manifests"
}
}
}
Now you can run it with variant
:
variant run -h
will show you that all the jobs are available via sub-commands:
$ variant run -h
Usage:
variant run [flags]
variant run [command]
Available Commands:
deploy
helm
kubectl
Flags:
-h, --help help for run
-n, --namespace string Namespace to interact with
Use "variant run [command] --help" for more information about a command.
And variant run deploy -h
for the usage for the specific job = sub-command named deploy
:
Deploys our application and the infrastructure onto the K8s cluster
Usage:
variant run deploy [flags]
Flags:
-h, --help help for deploy
Global Flags:
-n, --namespace string Namespace to interact with
As you've seen in the help output, variant run deploy
runs the deploy
job, which in turn runs kubectl
and helm
to install your apps onto the K8s cluster:
$ variant run deploy
Once you're finished developing the command, let's build a single executable binary of the command for easy distribution:
$ variant export binary ./ build/myapp
The exported executable binary accepts the same arguments as variant run
:
$ ./build/myapp -h
$ ./build/myapp run deploy
Congratulations! You're now ready to dive deep and solve your own problems with Variant.
Still curious how Variant helps developing your own command as it grows?
Head over to the following per-topic sections for more features:
- Writing Commands to learn the Variant DSL for writing commands
- Debugging Commands to learn how to debug your Variant command
- Writing Tests to learn the Variant DSL for writing tests
- Generating Shims to make your Variant command look native
- Compiling Command to export Go source or an executable binary
- Running Command From Other Directory
- Concurrency section to make
kubectl
andhelm
concurrent so that the installation time becomes minimal - Log Collection to filter and forward log of commands and the arguments passed to them along with their outputs
- Use Split, Merge and Import to split, compose and tidy Variant commands
- JSON Configuration Syntax can be used as an alternative to HCL2-based one
- Slack integration to turn your command into a Slack bot
If you're distributing this command with your teammates, do use variant generate shim
to create a shim to make it look like a native command:
$ variant generate shim examples/getting-started/
$ cat ./examples/getting-started/getting-started
#!/usr/bin/env variant
import = "."
$ ./examples/getting-started/getting-started
Usage:
getting-started [flags]
getting-started [command]
Available Commands:
deploy Deploys our application and the infrastructure onto the K8s cluster
helm
help Help about any command
kubectl
Flags:
-h, --help help for getting-started
-n, --namespace string Namespace to interact with
Use "getting-started [command] --help" for more information about a command.
As we've covered in the Getting Started guide, variant export
sub-commands can be used to export
your command in various formats.
variant export binary SRC CMD
generates an executable binary at the path CMD
from your command defined under the
directory SRC
:
$ variant export binary ./ build/myapp
The exported executable binary accepts the same arguments as variant run
.
So myapp -h
corresponds to cd $SRC; variant run -h
, while
myapp run deploy
corresponds to cd $SRC; variant run deploy
.
variant export go SRC/CMD PKG
generates a directory at PKG/CMD
that contains Go source files that can be built by
running go build PKG/CMD
.
Assuming you already have go
installed, you can run variint export go src/myapp build
, then edit code under
build/myapp
to make any customization that can't be done with shims, and finally build
an executable with go build -o myapp ./build/myapp
.
Usually, when your command has been defined under the directory path/to/your/command
, variant run
requires you to chdir
to
it before running.
To be clear, cd $SRC; variant run
can be used to run it from another directory. However with that you command cannot
access the actual current directory, as you've already cd
ed.
You can also use a shim or an exported binary to make it runnable from any
directory. But it takes some time so probably you'd like to run it from any directory without a variant export
step
while developing?
The VARIANT_DIR
environment variable might be the solution. When variant
recognizes it, it reads the command from the directory
specified by it.
Just run:
VARIANT_DIR=$SRC variant run
Your command can now be run without cd
and still has access to the current directory.
Do you have a huge yourcmd.variant
that needs to be split for readability?
path/to/yourcmd.variant
:
job "foo" {
# snip
}
job "bar" {
# snip
}
job "baz" {
# snip
}
Variant usually works per-directory basis. That is, it loads and merges all the .variant
files in a directory to form a single command.
That is, you can just split the file into three .variant
files in the same directory to split the huge file:
path/to/yourcmd_foo.variant
:
job "foo" {
parameter "param1" {
# snip
}
# snip
}
path/to/yourcmd_bar.variant
:
job "bar" {
# snip
}
path/to/yourcmd_baz.variant
:
job "baz" {
# snip
}
Okay that works. But you ended up too many files in a single directory?
A "parent" variant file containing import
or imports
can be used to load all the *.variant
files in the directory into the current job
.
path/to/yourcmd.variant
:
job "foo" {
import = "./foo"
}
job "bar" {
import = "./bar"
}
path/to/foo/foo.variant
:
parameter "param1" {
# snip
}
# snip
Note that imports
is the newer variant of import
that supports multiple sources to be imported.
Also, you can import following sources:
- Relative path to local directory (A local path that doesn't start with
/
, likefoo/bar
) - Absolute path to local directory (An absolute path that starts with
/
, like/variant/modules/example.com/foo/bar
) - URLs to Git repository (
REPO_URL@PATH/TO/DIR?ref=BRANCH
, e.g.git::ssh://[email protected]/mumoshu/variant2@examples/advanced/import/foo?ref=master
)
See the import example for the full declaration of this command for reference.
The example in the Getting Started
guide can be modified by adding needs
to build a DAG of steps and concurrency
for setting the desired number of concurrency:
BEFORE:
job "deploy" {
step "deploy infra" {
run "helm" {
release = "app1"
chart = "app1"
}
}
step "deploy apps" {
run "kubectl" {
dir = "deploy/environments/${opt.env}/manifests"
}
}
}
AFTER:
job "deploy" {
concurrency = 2
step "deploy fluentd" {
run "helm" {
release = "fluentd"
chart = "fluentd"
}
}
step "deploy prometheus" {
run "helm" {
release = "prometheus"
chart = "prometheus"
}
}
step "deploy apps" {
run "kubectl" {
dir = "deploy/environments/${opt.env}/manifests"
}
needs = ["deploy fluentd", "deploy prometheus"]
}
}
Now, running variant run deploy
deploys fluend and prometheus concurrently.
Once finished, it deploys your app, as you've declared so in the needs
attribute of the run "deploy apps" {}
block.
log
block(s) placed under a job
can be used to forward log of commands and the arguments passed to them along with their outputs.
Let's assume you'd like to collect log for the test
job:
job "test" {
run "echo" {
message = "foo"
}
}
Collecting all the run
and exec
that are executed as part of this job
along with all the metadata attached to respective event can be achieved by filtering events by condition
and formatting log messages with format
like:
log {
collect {
condition = event.type == "exec"
format = "exec=${jsonencode(event.exec)}"
}
collect {
condition = event.type == "run"
format = "run=${jsonencode(event.run)}"
}
# The log file to be created
file = "${context.sourcedir}/log.txt"
}
In case you need to post-process or upload the log file, use forward
to pass the log file to any Variant job
:
file = "${context.sourcedir}/log.txt"
forward {
run "save-logs" {
logfile = log.file
}
}
# ...
job "save-logs" {
option "logfile" {
type = string
}
exec {
command = "upload-to-jira"
args = ["-f", opt.logfile]
}
}
See the logcollection example for the full declaration of this command for reference.
Slack integration turns your Variant command into a Slack bot.
The bot is installed onto a specific Slack channel as a Slack app to listen any message to trigger a command.
Once you run:
SLACK_BOT_TOKEN=... SLACK_VERIFICATION_TOKEN=... variant start slackbot -n mycmd
in a directory that has a valid Variant command, it starts a RTM session to listen messages posted to the Slack channel.
When a message starting with:
/mycmd
is posted to the channel, the bot inteprets
/mycmd <CMD> <FLAGS>
as:
variant run <CMD> <FLAGS>
and runs it.
Auto-prompting via interactive messages
One of cool features of the bot is that when you missed to specify values for certain options, it will automatically start a interactive session to let you select and input missing values within Slack. You don't need to remember all the flags nor repeat lengthy commands anymore.
Variant uses its own configuration language based on the HashiCorp configuration language 2.
It is designed to allow concise descriptions of your command. The Variant language is declarative, describing an intended goal of the command rather than the steps to compose your command.
In addition to everything available via the native HCL syntax, Variant language provides the following HCL blocks
and functions
to declare your command.
job "CMD SUBCMD {}
is a job that can be run via run "CMD SUBCMD" {}
or variant run CMD SUBCMD
Do only one thing in each "job"
Each job can contain any of the followings, but not two or more of them:
- assert
- exec
- run
- step
This restriction ensures that you can do only one thing in each job, which makes testing and reading the code easier.
That is, a job containing assert
can be used as a custom assertion "function" and nothing else.
A job containing one or more step
s can be used as a workflow composed of multiple jobs. Each step
is restricted to call a single job
. As each job
is easily unit testable, this ensures that you can test the workflow without dealing with each job's implementation.
Options:
job
has the following attributes:
private
: when set totrue
by writingprivate = true
, the job is hidden from the command-line help.
parameter "NAME" {}
is Nth positional argument to job
that can be pased via run "the job" { NAME = "val1" }
or variant run the job val1
option "NAME" {}
is a named argument to job
that can be passed via run "the job" { NAME = "val1" }
or varuant run the job --NAME val1
config "NAME" {}
is a layered configuration named NAME
Let's assume defaults.yaml
has the following content:
api:
endpoint: api.example.com
replicas: 2
And a Variant command that contains a job named thing
that has a config
block named conf
:
job "thing" {
option "env" {
type = string
}
config "myconf" {
source file {
path = "defaults.yaml"
}
source job {
name = "values"
args = {
env = opt.env
}
}
source job {
name = "value"
args = {
}
key = "key1"
format = "text"
}
}
exec {
command = "echo"
args = list("apiendpoint=${conf.myconf.api.endpoint}, replicas=${conf.myconf.replicas}, env=${conf.myconf.env}, key1=${conf.myconf.key1}")
}
}
job "values" {
option "env" {
type = string
}
exec {
command = "echo"
args = list("env: ${opt.env}")
}
}
job "value" {
exec {
command = "echo"
args = list("val1")
}
}
The first source loads a file at defaults.yaml
and merges the keys and values in it into conf
.
Similarly, the second source loads keys and values from a job named values
and merges them into conf
.
The third sets the value for the key key1
where the value is the output of a job named value
.
The configuration myconf
can be referenced by ${conf.myconf.KEY}
as shown in the exec
block within the thing
job above.
Let's run variant run thing --env prod
and the output should be:
apiendpoint=api.example.com, replicas=2, env=prod, key1=val1
run
runs a job with args. run
is available within job
and test
.
job "a job" {
run "another job" {
param1 = "val1"
opt1 = "val2"
}
}
job "another job" {
parameter "param1" {
type = string
}
option "opt1" {
type = string
}
exec {
// ...
}
}
Sometimes you'd encounter a situation that you need indirection.
That is, the job name must be dynamically determined depending on a variable, parameter, or an option.
Variant2 provides an alternative run
block syntax for that, which basically omits the NAME label in the run "NAME" { }
you've previously seen:
To call the example
job with the parameter foo
set to "FOO"
and the option bar
set to "BAR"
, you write the run
block like the below:
run {
job = "example"
with = {
foo = "FOO"
bar = "BAR"
}
}
TODO
An exec
block executes the OS command.
Available attributes:
cmd
: The path to the executable binary/scriptargs
: The arguments to be passed to the commandenv
: The environment variables given to the commanddir
: The working directory
- All the Terraform built-in functions
- Plus a few Variant-specific functions
jsonpath
(definition, example)
To learn more, see examples for working examples covering all these blocks
and functions
.
Optionally read the following for overviews on each type of block and functions.
Variant
has a JSON-based dialect of its DSL. It's based on the HCL's build-in feature to natively support JSON, so the conversion rules between HCL and JSON is very similar to famous HCL-based language like Terraform.
Generally speaking, you can use blocks, attributes and expressions with a little overhead. See the options-json example for more details.
Setting the environment variable VARIANT_TRACE
to a non-empty value enables the trace logging.
It might be handy when you are curious how a third-party Variant command works, or when it seems like your command has an unknown bug that must be fixed.
Example:
$ VARIANT_TRACE=1 VARIANT_DIR=examples/issues/8-logging ./variant run example
go build -o variant ./pkg/cmd
foobar
TRACE {"Type":"exec","Time":"2020-04-09T16:01:37.436145+09:00","Run":null,"Exec":{"Command":"echo","Args":["foobar"]}}exec={"args":["foobar"],"command":"echo"}
Variant
has its own testing framework composed of the test runner and the config syntax.
The test runner can be triggered by executing variant test
. Upon run, the runner searches for test files whose names suffixed with _test.variant
, parses it and run all the tests defined within it.
The test configuration syntax is as simple as declaring test "CMD1 SUBCMD1 {}
, which is a set of unit tests for job "CMD1 SUBCMD1"
.
In each test
, you can use case
, run
and assert
blocks to define tests.
case "case1" { varname = value }
defines a test case named case1
with the test data consists of a single variable called varname
whose value is value
. You can include one or more variables in a case
.
run "jobname" { arg1 = value }
specifies which job
is run and which args is passed to the job run. This syntax is equivalent to run
under job
s and step
s.
Each assert { condition = expression }
block adds a assertion that is run by the test framework. When the expression contained in the condition
evaluated to false
, the test fails.
Example:
test "options" {
case "ok1" {
exitstatus = 0
err = ""
out = trimspace(<<EOS
1 2 3 a b|c
EOS
)
}
run "test" {
int1 = 1
ints1 = list(2,3)
str1 = "a"
strs1 = list("b","c")
}
assert "error" {
condition = run.err == case.err
}
assert "out" {
condition = (run.res.set && run.res.stdout == case.out) || !run.res.set
}
assert "exitstatus" {
condition = run.res.exitstatus == case.exitstatus
}
}
For more examples, please browse examples
directories and look for any files ending with _test.variant
.
The Variant
DSL is very much based on HCL and CTY.
The following is a set of recommended reading for learning HCL and CTY.
hcl
cty