Developing Plugins
It's easy to create a plugin and use it immediately in your project. First choose a directory outside your project and run :
$ ignite plugin scaffold my-plugin
This will create a new directory my-plugin
that contains the plugin's code,
and will output some instructions about how to use your plugin with the
ignite
command. Indeed, a plugin path can be a local directory, which has
several benefits:
- you don't need to use a git repository during the development of your plugin.
- the plugin is recompiled each time you run the
ignite
binary in your project, if the source files are older than the plugin binary.
Thus, the plugin development workflow is as simple as :
- scaffold a plugin with
ignite plugin scaffold my-plugin
- add it to your config via
ignite plugin add -g /path/to/my-plugin
- update plugin code
- run
ignite my-plugin
binary to compile and run the plugin. - go back to 3.
Once your plugin is ready, you can publish it to a git repository, and the
community can use it by calling ignite plugin add github.com/foo/my-plugin
.
Now let's detail how to update your plugin's code.
The plugin interface
The ignite
plugin system uses github.com/hashicorp/go-plugin
under the hood,
which implies to implement a predefined interface:
// An ignite plugin must implements the Plugin interface.
type Interface interface {
// Manifest declares the plugin's Command(s) and Hook(s).
Manifest() (Manifest, error)
// Execute will be invoked by ignite when a plugin Command is executed.
// It is global for all commands declared in Manifest, if you have declared
// multiple commands, use cmd.Path to distinguish them.
Execute(cmd ExecutedCommand) error
// ExecuteHookPre is invoked by ignite when a command specified by the Hook
// path is invoked.
// It is global for all hooks declared in Manifest, if you have declared
// multiple hooks, use hook.Name to distinguish them.
ExecuteHookPre(hook ExecutedHook) error
// ExecuteHookPost is invoked by ignite when a command specified by the hook
// path is invoked.
// It is global for all hooks declared in Manifest, if you have declared
// multiple hooks, use hook.Name to distinguish them.
ExecuteHookPost(hook ExecutedHook) error
// ExecuteHookCleanUp is invoked by ignite when a command specified by the
// hook path is invoked. Unlike ExecuteHookPost, it is invoked regardless of
// execution status of the command and hooks.
// It is global for all hooks declared in Manifest, if you have declared
// multiple hooks, use hook.Name to distinguish them.
ExecuteHookCleanUp(hook ExecutedHook) error
}
The code scaffolded already implements this interface, you just need to update the methods' body.
Defining plugin's manifest
Here is the Manifest
struct :
type Manifest struct {
Name string
// Commands contains the commands that will be added to the list of ignite
// commands. Each commands are independent, for nested commands use the
// inner Commands field.
Commands []Command
// Hooks contains the hooks that will be attached to the existing ignite
// commands.
Hooks []Hook
// SharedHost enables sharing a single plugin server across all running instances
// of a plugin. Useful if a plugin adds or extends long running commands
//
// Example: if a plugin defines a hook on `ignite chain serve`, a plugin server is instanciated
// when the command is run. Now if you want to interact with that instance from commands
// defined in that plugin, you need to enable `SharedHost`, or else the commands will just
// instantiate separate plugin servers.
//
// When enabled, all plugins of the same `Path` loaded from the same configuration will
// attach it's rpc client to a an existing rpc server.
//
// If a plugin instance has no other running plugin servers, it will create one and it will be the host.
SharedHost bool `yaml:"shared_host"`
}
In your plugin's code, the Manifest
method already returns a predefined
Manifest
struct as an example. Adapt it according to your need.
If your plugin adds one or more new commands to ignite
, feeds the Commands
field.
If your plugin adds features to existing commands, feeds the Hooks
field.
Of course a plugin can declare Commands
and Hooks
.
A plugin may also share a host process by setting SharedHost
to true
.
SharedHost
is desirable if a plugin hooks into, or declares long running commands.
Commands executed from the same plugin context interact with the same plugin server.
Allowing all executing commands to share the same server instance, giving shared execution context.
Adding new command
Plugin commands are custom commands added to the ignite cli by a registered
plugin. Commands can be of any path not defined already by ignite. All plugin
commands will extend of the command root ignite
.
For instance, let's say your plugin adds a new oracle
command to ignite
scaffold
, the Manifest()
method will look like :
func (p) Manifest() (plugin.Manifest, error) {
return plugin.Manifest{
Name: "oracle",
Commands: []plugin.Command{
{
Use: "oracle [name]",
Short: "Scaffold an oracle module",
Long: "Long description goes here...",
// Optionnal flags is required
Flags: []plugin.Flag{
{Name: "source", Type: plugin.FlagTypeString, Usage: "the oracle source"},
},
// Attach the command to `scaffold`
PlaceCommandUnder: "ignite scaffold",
},
},
}, nil
}
To update the plugin execution, you have to change the plugin Execute
command,
for instance :
func (p) Execute(cmd plugin.ExecutedCommand) error {
if len(cmd.Args) == 0 {
return fmt.Errorf("oracle name missing")
}
var (
name = cmd.Args[0]
source, _ = cmd.Flags().GetString("source")
)
// Read chain information
c, err := getChain(cmd)
if err != nil {
return err
}
//...
}
Then, run ignite scaffold oracle
to execute the plugin.
Adding hooks
Plugin Hooks
allow existing ignite commands to be extended with new
functionality. Hooks are useful when you want to streamline functionality
without needing to run custom scripts after or before a command has been run.
this can streamline processes that where once error prone or forgotten all
together.
The following are hooks defined which will run on a registered ignite
commands
Name | Description |
---|---|
Pre | Runs before a commands main functionality is invoked in the PreRun scope |
Post | Runs after a commands main functionality is invoked in the PostRun scope |
Clean Up | Runs after a commands main functionality is invoked. if the command returns an error it will run before the error is returned to guarantee execution. |
Note: If a hook causes an error in the pre step the command will not run
resulting in post
and clean up
not executing.
The following is an example of a hook
definition.
func (p) Manifest() (plugin.Manifest, error) {
return plugin.Manifest{
Name: "oracle",
Hooks: []plugin.Hook{
{
Name: "my-hook",
PlaceHookOn: "ignite chain build",
},
},
}, nil
}
func (p) ExecuteHookPre(hook plugin.ExecutedHook) error {
switch hook.Name {
case "my-hook":
fmt.Println("I'm executed before ignite chain build")
default:
return fmt.Errorf("hook not defined")
}
return nil
}
func (p) ExecuteHookPost(hook plugin.ExecutedHook) error {
switch hook.Name {
case "my-hook":
fmt.Println("I'm executed after ignite chain build (if no error)")
default:
return fmt.Errorf("hook not defined")
}
return nil
}
func (p) ExecuteHookCleanUp(hook plugin.ExecutedHook) error {
switch hook.Name {
case "my-hook":
fmt.Println("I'm executed after ignite chain build (regardless errors)")
default:
return fmt.Errorf("hook not defined")
}
return nil
}
Above we can see a similar definition to Command
where a hook has a Name
and
a PlaceHookOn
. You'll notice that the Execute*
methods map directly to each
life cycle of the hook. All hooks defined within the plugin will invoke these
methods.